A method for predicting motor imagery EEG signals

By using the support vector neural network solver in the prediction of electroencephalogram signal, the interval maximization criterion and Lagrangian multiplier method are used to transform it into linear variational inequality, and the segmented linear projection equation is designed, which solves the solution efficiency of traditional methods. The problem of low efficiency and real-time prediction of EEG signal is achieved.

CN114565028BActive Publication Date: 2024-11-29SOUTH CHINA UNIV OF TECH
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202210143506.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-02-16
Publication Date
2024-11-29
Estimated Expiration
2042-02-16

AI Technical Summary

Technical Problem

In the existing methods of predicting electroencephalogram signals in motion imagination, traditional support vector machines use serial solvers to make the solution efficiency low, making it difficult to meet the needs of real-time and high efficiency.

Method used

Using a support vector neural network solver based on neural mechanics, a quadratic planning problem is established through the interval maximization criterion, a Lagrangian multiplication method is used to transform it into a linear variational inequality, and a segmented linear projection equation is designed to construct the original dual. The neural network solves it to obtain the classification function of the electroencephalogram signal of motor imagination.

Benefits of technology

It realizes a method of predicting electroencephalogram signal with faster convergence speed, higher classification accuracy and better robustness, and has the characteristics of strong real-time, small calculation amount and high prediction accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114565028B_ABST
    Figure CN114565028B_ABST
Patent Text Reader

Abstract

The present invention provides a method for predicting motor imagery EEG signals, comprising the following steps: preprocessing the collected motor imagery EEG signal data of the user; obtaining the features and labels of the motor imagery EEG signal data, and establishing a quadratic programming problem by using the interval maximization criterion of two types of samples; converting the quadratic programming problem into a piecewise linear projection equation by Lagrange multiplier method and equivalent transformation; designing the original dual neural network based on the piecewise linear projection equation and solving the problem; and passing the solution result of the dual neural network to the motor imagery EEG signal classification function, thereby obtaining the prediction result of the EEG signal. The present invention is used for the prediction problem of motor imagery EEG signal category, and the support vector neural network method adopted is a neural network solver developed based on the support vector machine model, which has high prediction accuracy, does not involve complex matrix operations, has fast convergence speed, and also has the advantages of strong real-time performance and easy implementation.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the field of motor imagery EEG signal prediction, and in particular is a motor imagery EEG signal prediction method. Background Art

[0002] Motor imagery EEG signal prediction is to predict the category of the user's imagined action based on the collected user EEG signals. Motor imagery refers to the activation of specific brain areas when the user imagines his or her limbs (or muscles) moving but there is no actual movement output. By analyzing EEG signals, detecting and identifying the activation effects of different brain areas to judge the user's intention, and then realizing direct communication and control between the human brain and external devices. Compared with P300 EEG signal acquisition, which requires users to concentrate on the screen prompt information, the acquisition of motor imagery EEG signals only requires users to imagine the action after the prompt, which has a lower burden on users. Therefore, the classification of motor imagery EEG signals has always been a hot research field at home and abroad (P.Gaur, RBPachori, H.Wang, G.Prasad.An Automatic Subject Specific Intrinsic Mode Function Selection for Enhancing Two-Class EEG-Based Motor Imagery-BrainComputer Interface[J].IEEE Sensors Journal, 2019, 19(16): 6938-6947.). Motor imagery EEG signal predicts human imagery category. Most of the existing prediction and classification methods are support vector machines (A. Jiang, J. Shang, X. Liu, Y. Tang, HK Kwan, Y. Zhu. Efficient CSP Algorithm With Spatio-Temporal Filtering for Motor Imagery Classification [J]. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2020, 28 (4): 1006-1016.), but the traditional support vector machine uses a serial solver for solving, and the solving efficiency is low. Compared with the traditional method, the method of the present invention is a support vector neural network solver based on neuromechanics developed from the support vector machine model, which has the characteristics of faster convergence speed, higher classification accuracy, better robustness and real-time performance. Summary of the invention

[0003] The purpose of the present invention is to propose a method for predicting motor imagery EEG signals based on a support vector neural network with high classification accuracy, global convergence and fast convergence speed. The technical solution of the present invention has small calculation amount, strong real-time performance and high prediction accuracy.

[0004] The present invention is achieved by at least one of the following technical solutions.

[0005] A method for predicting motor imagery EEG signals, comprising the following steps:

[0006] 1) Preprocessing the collected user's motor imagery EEG signal data;

[0007] 2) Obtain the features and labels of the preprocessed data, and establish a quadratic programming problem using the interval maximization criterion of the two types of samples;

[0008] 3) Convert the quadratic programming problem into a linear variational inequality using the Lagrange multiplier method;

[0009] 4) Convert the linear variational inequality into a piecewise linear projection equation;

[0010] 5) Design the primal-dual neural network based on the piecewise linear projection equation and solve the problem;

[0011] 6) The solution results of the original dual neural network are used to obtain the weight vector and offset of the motor imagery EEG signal classification function;

[0012] 7) passing the weight factor and the offset to the motor imagery EEG signal classification function to obtain the prediction result of the EEG signal;

[0013] 8) The prediction results are output and displayed on the computer screen.

[0014] Preferably, the preprocessing is to perform z-score standardization on the collected user motor imagery EEG signal data in sequence.

[0015] Preferably, the quadratic programming problem is a minimization problem:

[0016] α T Wα / 2+q T α

[0017] The quadratic programming problem is subject to the constraints: T α=0,α≥0、0≤α≤C,where α∈R m represents the Lagrange multiplier, m is the number of data samples, R m represents the m-dimensional real number field, y∈R m is the motor imagery EEG signal label vector, W = A·B∈R m×m And A=(K(x i , xj )) m×m The Gaussian kernel function matrix representing the motor imagery EEG signal feature vector, B = yy T ∈R m×m , q = (-1, -1, ..., -1) T ∈R m , where x i 、x j Represents the feature vector of the i-th and j-th data, K(x i , x j ) is the Gaussian kernel function, B is the label matrix composed of the label vector y, A is the Gaussian kernel function matrix of the motor imagery EEG signal feature vector, q is the m-dimensional column vector, where the value of q is -1, and the parameter C is the penalty factor.

[0018] Preferably, the quadratic programming problem in step 3) is transformed into a linear variational inequality:

[0019] (dd * ) T (Hd * +p)≥0

[0020]

[0021]

[0022] where d is the primal-dual decision vector, d * represents the optimal solution of the original dual decision vector; the matrix H = (W, y; -y T , 0)∈R (m+1)×(m+1) , W is the Gaussian kernel function matrix of the motor imagery EEG signal feature vector, y∈R m is the motor imagery EEG signal label vector, vector p=(q;0)∈R m+1 , q = (-1, -1, ..., -1) T ∈R m , R m represents the m-dimensional real number field, α∈R m represents the Lagrange multiplier, u represents the Lagrange multiplier for the constraint condition, α * and u * represents the optimal solution of α and u.

[0023] Preferably, the quadratic programming problem in step 3) is transformed into a piecewise linear projection equation:

[0024] P Ω (d-(Hd+p))-d=0

[0025] Where P Ω (·)∈R m+1represents the piecewise linear projection function, d represents the primal-dual decision vector matrix, and the matrix H = (W, y; -y T , 0)∈R (m+1)×(m+1) , W represents the Gaussian kernel function matrix of the motor imagery EEG signal feature vector, y∈R m represents the label vector of the motor imagery EEG signal; R m represents the m-dimensional real number field; vector p=(q;0)∈R m+1 , q=(-1,-1,…,-1) T ∈R m .

[0026] Preferably, the piecewise projection function is:

[0027]

[0028] Among them, d i represents the i-th element in d, Indicates d i The maximum and minimum boundary values.

[0029] Preferably, in step 4), a support vector neural network solver is introduced to solve the piecewise linear projection equation using the support vector neural network solver, and the dynamic equation of the solver is:

[0030]

[0031] Where θ>0 is a parameter used to measure the convergence rate of the original dual neural network, I is the identity matrix; the matrix H = (W, y; -y T , 0)∈R (m+1)×(m+1) , W is the Gaussian kernel function matrix of the motor imagery EEG signal feature vector, y∈R m is the motor imagery EEG signal label vector, d is the decision variable of the Lagrange multiplier α containing the quadratic programming problem, and the vector p=(q;0)∈R m+1 , represents the derivative of vector d, P Ω (·)∈R m+1 Represents a piecewise projection function.

[0032] Preferably, the support vector neural network solver uses the solved result d * Get the first m Lagrange multipliers α * , and further obtain the normal vector ω in the prediction function * and the displacement term b * :

[0033]

[0034]

[0035] Where m is the number of data samples, and represents the i-th and j-th optimal solutions of the Lagrange multiplier α of the quadratic programming problem, x i 、x j represents the feature vector of the i-th and j-th data, the parameter C is the penalty factor, and SVS represents the set of support vectors in the data sample.

[0036] Preferably, the motor imagery EEG signal prediction result is:

[0037] f(x)=sgn(ω T ·x)+b * )

[0038] Where b is the displacement term, x represents the characteristic data of the motor imagery EEG signal, ω∈R n is the normal vector that determines the direction of the hyperplane.

[0039] Preferably, the classification result of the input sample is obtained according to the prediction function f(x), and the result is transmitted and displayed on a computer screen.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] The present invention can effectively overcome the shortcomings of existing motor imagery EEG signal prediction technologies and provide a motor imagery EEG signal prediction method with good real-time performance, small amount of calculation, fast speed, higher efficiency and easy solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A flowchart of a method for predicting motor imagery EEG signals according to the present invention;

[0043] Figure 2 This is the electrode diagram for collecting EEG signals of the present invention;

[0044] Figure 3 This is the electrode diagram for collecting 64 channels of motor imagery EEG signals in the example. DETAILED DESCRIPTION

[0045] In order to enable those skilled in the art to better understand the scheme of the present invention, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation methods. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present invention. The present invention will be further described below in conjunction with the accompanying drawings.

[0046] Example 1

[0047] like Figure 1 The present invention shown in the figure comprises five parts: data preprocessing, quadratic programming problem, piecewise linear projection equation, primal-dual neural network solver, and motor imagery EEG signal prediction. The prediction method comprises the following steps:

[0048] 1) Preprocessing the collected user's motor imagery EEG signal data;

[0049] 2) Obtaining the features and labels of the motor imagery EEG signal data in step 1), and establishing a quadratic programming problem using the interval maximization criterion of two types of samples (i.e., left-hand imagery EEG data and right-hand imagery EEG data);

[0050] 3) converting the quadratic programming problem in step 2) into a piecewise linear projection equation by Lagrange multiplier method and equivalent transformation;

[0051] 4) Design the primal-dual neural network based on the piecewise linear projection equation in step 3) and solve the problem;

[0052] 5) The solution of the original dual neural network in step 4) is passed to the motor imagery EEG signal classification function to obtain the prediction result of the EEG signal.

[0053] In the above technical solution, the step 1) pre-processes the user's motor imagery EEG signal data, and the specific operation is to perform z-score standardization in sequence, which is specifically expressed as:

[0054]

[0055] Where x represents the user's motor imagery EEG signal data, μ represents the data mean, and σ represents the data standard deviation. In this embodiment, the data is divided into a training set and a test set in a ratio of 8:7.

[0056] The step 2) extracts the features of the motor imagery EEG signals and the corresponding labels to establish a quadratic programming problem, specifically, the features of each motor imagery EEG signal are used as the classification feature vector, and the imagined action category (i.e., the two categories of imagined left hand and imagined right hand) are used as the classification result label, and the idea established by the support vector machine is used to transform it into a quadratic programming problem that only needs to solve the Lagrangian parameter α.

[0057] The step 3) converts the quadratic programming problem into a piecewise linear projection equation, specifically, first converting the quadratic programming problem into a linear variational inequality using Lagrange number multiplication and equivalent transformation, and then converting it into a piecewise linear projection equation.

[0058] The step 4) designs the primal-dual neural network based on the piecewise linear projection equation and solves the problem, and introduces a support vector neural network solver to solve the piecewise linear projection equation.

[0059] The step 5) passes the solution result of the original dual neural network to the motor imagery EEG signal decision function, thereby obtaining the prediction result of the EEG signal.

[0060] Example 2

[0061] like Figure 2 As shown, the 60-channel acquisition electrode diagram of the motor imagery EEG signal in this embodiment. First, the user's motor imagery EEG signal data is preprocessed, and then a quadratic programming problem is established according to the characteristics of the data and the corresponding labels. Then, the quadratic programming problem is converted into a piecewise linear projection equation through Lagrange multipliers and equivalent transformations, and then the original dual neural network solver is used to calculate the optimal solution, and finally the new motor imagery EEG signal prediction result is obtained. The details are as follows:

[0062] First, given a motor imagery EEG signal dataset containing m samples, collected from Figure 2 The 60-channel acquisition electrode diagram shown in the figure defines the data set as S:

[0063] S={(x1,y1),(x2,y2),…,(x m ,y m )} (1)

[0064] Among them, x i ∈R n , i = 1, 2, ..., m is the feature vector of the i-th data sample, y i ∈{-1, 1}, i=1, 2, ..., m is the category label of the i-th data sample, m represents the number of data samples, R n represents the n-dimensional real number domain. The operations for preprocessing the data set include standardization and data partitioning. The standardization method used in the present invention is the most commonly used z-score standardization, which is specifically expressed as Where x represents the characteristic data of the motor imagery EEG signal, μ represents the data mean, and σ represents the data standard deviation, so that the processed data has a mean of 0 and a standard deviation of 1.

[0065] Secondly, a quadratic programming problem is established for the motor imagery EEG signal dataset. The motor imagery EEG signal prediction decision function is:

[0066] f(x)=sgn(ω T ·x)+b)

[0067] where ω∈R n is the normal vector that determines the direction of the hyperplane, and b is the displacement term. Let y = (y1, y2, ..., y m ) T ∈R m, q = (-1, -1, ..., -1) T ∈R m , R m represents the m-dimensional real number field;

[0068]

[0069]

[0070] Where K(x i , x j ) is the Gaussian kernel function of the two samples, B is the label matrix composed of the label vector y, A is the Gaussian kernel function matrix of the motor imagery EEG signal feature vector, q is an m-dimensional column vector whose value is -1, and parameter C is the penalty factor. Consider the quadratic programming problem constructed by motor imagery EEG signal data

[0071]

[0072] sty T α=0 (5)

[0073] 0≤α≤C (6)

[0074] where α∈R m is the Lagrange multiplier, W = A·B∈R m×m And A=(K(x i , x j )) m×m is the Gaussian kernel function matrix of the motor imagery EEG signal feature vector, B = yy T ∈R m×m , q = (-1, -1, ..., -1) T ∈R m , where x i 、x j Represents the feature vector of the i-th and j-th data, K(x i , x j ) is the Gaussian kernel function, B is the label matrix composed of the label vector y, A is the Gaussian kernel function matrix of the motor imagery EEG signal feature vector, q is an m-dimensional column vector, all of which are -1, and parameter C is the penalty factor. If this problem is solved, the normal vector of the prediction function of the motor imagery EEG signal is and displacement terms can be calculated from α, * represents the optimal solution of α, represents the i-th α * The specific value of SVS represents the set of support vectors in the data sample.

[0075] Then, converting the quadratic programming problem into a piecewise linear projection equation can be divided into two steps. The first step is to convert the quadratic programming problem into a linear variational inequality. The quadratic programming problem (4)-(6) is equivalent to the following equation:

[0076]

[0077] where u∈R is the dual decision variable corresponding to constraint (5), and the vector [α * ,u * ] T What needs to be found is the optimal solution of the Lagrange multiplier α of the quadratic programming problem and the optimal solution of the dual decision variable u. Then, equation (7) can be transformed into a linear variational inequality problem. Assume that the quadratic programming problem has an optimal solution α * , then we can find an optimal solution d for the following linear variational inequality * ={d|ζ - ≤d≤ζ +} makes

[0078] (dd * ) T (Hd * +p)≥0 (8)

[0079] The matrix and vector are simplified from equation (7) and are defined as follows:

[0080]

[0081]

[0082] d is the primal-dual decision vector, d * The optimal solution α that needs to be solved according to the known conditions * The second step is to transform the linear variational inequality into a piecewise linear projection equation. The linear variational inequality (8) can be equivalently transformed into the following piecewise linear projection equation:

[0083] P Ω (d-(Hd+p))-d=0 (11)

[0084] Where P Ω (d)=[[P Ω (d)]1,[P Ω (d)]2,…,[P Ω (d)] m+1 ] T ∈R m+1 is a piecewise projection function, which is defined as:

[0085]

[0086] in, Indicates d i Therefore, solving the QP problem (4)-(6) is equivalent to the piecewise linear projection equation problem (11), which is equivalent to finding d * So that the error function E(d) = P Ω (d-(Hd+p))-d converges to zero.

[0087] Again, in order to solve the piecewise linear projection equation problem, the present invention proposes a support vector neural network solver. The dynamic equation of the solver is:

[0088]

[0089] Where θ>0 is a design parameter used to measure the convergence rate of the original dual neural network. It means to take the derivative of vector d.

[0090] Finally, the prediction results of motor imagery EEG signals are obtained. (0) Iterate and finally get the optimal solution d according to the proposed solver and convergence theory. * , and the optimal solution of the quadratic programming problem is α * Yes * The first m elements of * and b * It can also be calculated. The prediction function f(x) = sgn(ω *T ·x)+b * ) is used to judge the category. The category of the corresponding EEG signal can be judged according to the positive and negative values ​​of f(x). If the output value is -1, it is judged as the imagined left hand category, and if the output value is 1, it is judged as the imagined right hand category.

[0091] Example 3

[0092] like Figure 3 As shown, the 64-channel acquisition electrode diagram of the motor imagery EEG signal in this embodiment. First, the user's motor imagery EEG signal data is preprocessed, and then a quadratic programming problem is established according to the characteristics of the data and the corresponding labels. Then, the quadratic programming problem is converted into a piecewise linear projection equation through Lagrange multipliers and equivalent transformations, and then the original dual neural network solver is used to calculate the optimal solution, and finally the new motor imagery EEG signal prediction result is obtained. The details are as follows:

[0093] First, given a motor imagery EEG signal dataset containing m samples, collected from Figure 3 The 64-channel acquisition electrode diagram shown in the figure defines the data set as S:

[0094] S={(x1,y1),(x2,y2),…,(x m ,y m )} (14)

[0095] Among them, x i ∈R n , i = 1, 2, ..., m is the feature vector of the i-th data sample, y i ∈{-1, 1}, i=1, 2, ..., m is the category label of the i-th data sample, m represents the number of data samples, R n Represents the field of n-dimensional real numbers.

[0096] The operations of preprocessing the data set include normalization and data partitioning. The maximum and minimum normalization adopted in the present invention is specifically expressed as Where x represents data, x min Indicates the minimum value of the data, x max Represents the maximum value of the data, so that the processed data falls between [0, 1].

[0097] Secondly, a quadratic programming problem is established for the motor imagery EEG signal dataset. The motor imagery EEG signal prediction decision function is:

[0098] f(x)=sgn(ω T ·x)+b)

[0099] Where b is the displacement term, x represents the data of motor imagery EEG signal, ω∈R n is the normal vector that determines the direction of the hyperplane.

[0100] Consider the quadratic programming problem constructed from motor imagery EEG signal data

[0101]

[0102] sty T α=0 (16)

[0103] 0≤α≤C (17)

[0104] where α∈R m represents the Lagrange multiplier, m is the number of data samples, R m represents the m-dimensional real number domain, W is the Gaussian kernel function matrix of the motor imagery EEG signal feature vector, q=(-1, -1, ..., -1) T ∈R m , y∈R m is the motor imagery EEG signal label vector, and parameter C is the penalty factor.

[0105] If this problem is solved, the optimal parameters of the decision function for motor imagery EEG signal prediction are and displacement terms can be calculated from α, where and represents the i-th and j-th optimal solutions of the Lagrange multiplier α of the quadratic programming problem, x i 、x j represents the feature vector of the i-th and j-th data, SVS represents the set of support vectors in the data sample, and y i ,y j represents the category label of the i-th and j-th data samples, and m represents the number of data samples.

[0106] Then, converting the quadratic programming problem into a piecewise linear projection equation can be divided into two steps. The first step is to convert the quadratic programming problem into a linear variational inequality. The quadratic programming problem (15)-(17) is equivalent to the following equation:

[0107]

[0108] Among them, u * represents the optimal solution of u. Then, equation (18) can be transformed into a linear variational inequality problem. Assume that the quadratic programming problem has an optimal solution α * , then we can find an optimal solution d for the following linear variational inequality * ={d|ζ - ≤d≤ζ +} makes

[0109] (dd * ) T (Hd * +p)≥0 (19)

[0110] The second step is to transform the linear variational inequality into a piecewise linear projection equation. The linear variational inequality (19) can be equivalently transformed into the following piecewise linear projection equation:

[0111] P Ω (d-(Hd+p))-d=0 (20)

[0112] Therefore, solving the QP problem (15)-(17) is equivalent to solving the piecewise linear projection equation problem (20), which is equivalent to finding d * So that the error function P Ω (d-(Hd+p))-d=0 converges to zero, P Ω (·)∈R m+1 represents the piecewise linear projection function; the matrix H = (W, y; -yT , 0)∈R (m+1)×(m+1) , W represents the Gaussian kernel function matrix of the motor imagery EEG signal feature vector, y∈R m represents the label vector of the motor imagery EEG signal; R m represents the m-dimensional real number field, d represents the primal-dual decision vector matrix; vector p=(q;0)∈R m+1 , q=(-1,-1,…,-1) T ∈R m .

[0113] Again, in order to solve the piecewise linear projection equation problem, the present invention proposes a support vector neural network solver. The dynamic equation of the solver is

[0114]

[0115] Where θ>0 is a design parameter used to measure the convergence rate of the original dual neural network, and I is the identity matrix;

[0116] Finally, the prediction results of motor imagery EEG signals are obtained. (0) Iterate and finally get the optimal solution d according to the proposed solver and convergence theory. * , and the optimal solution of the quadratic programming problem is α * Yes * The first m elements of the corresponding prediction function are the normal vector ω * and the displacement term b * It can also be calculated. The prediction function f(x) = sgn(ω *T ·x)+b * ) is used to judge the category. The category of the corresponding EEG signal can be judged according to the positive and negative values ​​of f(x). If the output value is -1, it is judged as the imagined left hand category, and if the output value is 1, it is judged as the imagined right hand category.

[0117] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for predicting motor imagery EEG signals, characterized in that: The steps include: 1) Preprocessing the collected user's motor imagery EEG signal data; 2) Obtain the features and labels of the preprocessed data, and establish a quadratic programming problem using the interval maximization criterion of the two types of samples; 3) The quadratic programming problem is transformed into a linear variational inequality through the Lagrange multiplier method; the quadratic programming problem is transformed into a piecewise linear projection equation: P Ω (d-(Hd+p))-d=0 Where P Ω (·)∈R m+1 represents the piecewise linear projection function, d represents the primal-dual decision vector matrix, and the matrix H = (W, y; -y T ,0)∈R (m+1)×(m+1) , W represents the Gaussian kernel function matrix of the motor imagery EEG signal feature vector, y∈R m represents the label vector of the motor imagery EEG signal; R m represents the m-dimensional real number field; vector p=(q;0)∈R m+1 , q=(-1,-1,…,-1) T ∈R m ; The piecewise linear projection equation is: Among them, d i represents the i-th element in d, Indicates d i The maximum and minimum boundary values ​​of 4) Convert the linear variational inequality into a piecewise linear projection equation; The support vector neural network solver is introduced to solve the piecewise linear projection equation. The dynamic equation of the solver is: Where θ>0 is a parameter used to measure the convergence rate of the original dual neural network, I is the identity matrix; the matrix H = (W, y; -y T ,0)∈R (m+1)×(m+1) , W is the Gaussian kernel function matrix of the motor imagery EEG signal feature vector, y∈R m is the motor imagery EEG signal label vector, d is the decision variable of the Lagrange multiplier α containing the quadratic programming problem, and the vector p=(q;0)∈R m+1 , represents the derivative of vector d, P Ω (·)∈R m+1 represents the piecewise linear projection equation; The result d is used in the support vector neural network solver * Get the first m Lagrange multipliers α * , and further obtain the normal vector ω in the prediction function * and the displacement term b * : Where m is the number of data samples, and represents the i-th and j-th optimal solutions of the Lagrange multiplier α of the quadratic programming problem, x i 、x j represents the eigenvector of the i-th and j-th data, the parameter C is the penalty factor, and SVS represents the set of support vectors in the data sample; 5) Design the primal-dual neural network based on the piecewise linear projection equation and solve the problem; 6) The solution results of the original dual neural network are used to obtain the weight vector and offset of the motor imagery EEG signal classification function; 7) passing the weight factor and the offset to the motor imagery EEG signal classification function to obtain the prediction result of the EEG signal; 8) The prediction results are output and displayed on the computer screen.

2. The method for predicting motor imagery EEG signals according to claim 1, characterized in that: The preprocessing is to perform z-score standardization on the collected user motor imagery EEG signal data in sequence.

3. The method for predicting motor imagery EEG signals according to claim 1, characterized in that: The quadratic programming problem is a minimization problem: a T Wα / 2+q T a The quadratic programming problem is subject to the constraints: T α=0,α≥0,0≤α≤C,where α∈R m represents the Lagrange multiplier, m is the number of data samples, R m represents the m-dimensional real number field, y∈R m is the motor imagery EEG signal label vector, W = A·B∈R m×m And A=(K(x i ,x j )) m×m The Gaussian kernel function matrix representing the motor imagery EEG signal feature vector, B = yy T ∈R m×m , q=(-1,-1,…,-1) T ∈R m , where x i 、x j Represents the feature vector of the i-th and j-th data, K(x i ,x j ) is the Gaussian kernel function, B is the label matrix composed of the label vector y, A is the Gaussian kernel function matrix of the motor imagery EEG signal feature vector, q is the m-dimensional column vector, where the value of q is -1, and the parameter C is the penalty factor.

4. The method for predicting motor imagery EEG signals according to claim 1, characterized in that: The quadratic programming problem in step 3) is transformed into a linear variational inequality: (d-d * ) T (Hd * +p)≥0 where d is the primal-dual decision vector, d * represents the optimal solution of the original dual decision vector; the matrix H = (W, y; -y T ,0)∈R (m +1)×(m+1) , W is the Gaussian kernel function matrix of the motor imagery EEG signal feature vector, y∈r m is the motor imagery EEG signal label vector, vector p=(q;0)∈R m+1 , q=(-1,-1,…,-1) T ∈R m , R m represents the m-dimensional real number field, α∈R m represents the Lagrange multiplier, u represents the Lagrange multiplier for the constraint condition, α * and u * represents the optimal solution of α and u.

5. The method for predicting motor imagery EEG signals according to claim 1, characterized in that: The prediction results of motor imagery EEG signals are: f(x)=sgn((ω T ·x)+b * ) Where x represents the characteristic data of motor imagery EEG signal, ω∈R n is the normal vector that determines the direction of the hyperplane.

6. The method for predicting motor imagery EEG signals according to claim 1, characterized in that: The classification result of the input sample is obtained according to the prediction function f(x), and the result is transmitted and displayed on the computer screen.

Citation Information

Patent Citations

  • Motor imagery electroencephalogram voting strategy sorting method based on extreme learning machines

    CN103413050A

  • Motor imagery electroencephalogram signal classification method of semi-supervised learning optimization SVM

    CN110399805A