Informer-based human lower limb gait phase personalized prediction method

Abstract

The invention relates to an Informer-based human lower limb gait phase personalized prediction method, and relates to the technical field of exoskeleton robots. The method comprises the following steps: acquiring human lower limb gait motion information through a data acquisition device and forming a data set; training the pre-model; storing the trained pre-model; collecting individual lower limb data of the user; performing personalized model training based on the pre-model; storing the personalized model; acquiring a human lower limb right thigh swing angle through a data acquisition device; and performing model use based on the real-time data of the wearer. Different from a traditional gait phase prediction method, the method increases the model complexity, avoids the use of a large number of sensors, and increases the model precision on the basis of reducing the use of hardware equipment, so that a corresponding controller structure and control parameters are more accurately set according to specific prior information such as man-machine coupling characteristics and the like; and the wearing comfort of the exoskeleton user in the man-machine coupling cooperative walking movement is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of exoskeleton robots, and in particular to an Informer-based personalized prediction method for the gait phase of human lower limbs. Background Art

[0002] The lower limb exoskeleton rehabilitation robot is a new type of wearable bionic device, which is mainly used to enable patients with lower limb motor dysfunction to stand and walk. It is currently a hot research topic in rehabilitation medicine. Gait phase refers to the relationship between the movement states of different limbs of the human body during walking. The lower limb exoskeleton rehabilitation robot needs to control the movement of the robot according to the gait phase to achieve an effect similar to the human gait.

[0003] At present, there are mainly the following methods to achieve gait phase control. The first is sensor control, that is, by installing sensors such as pressure sensors, gyroscopes, accelerometers, etc. on the robot to obtain information about the robot and the environment, so as to control the movement of the robot. The second is EMG signal control, that is, controlling the movement of the robot by measuring electromyographic signals. This method requires electrodes to be attached to the muscles and the robot is controlled by measuring the electrical signals generated when the muscles contract. The last is model predictive control, that is, predicting the movement of the robot by establishing a mathematical model and controlling the movement of the robot according to the predicted results. This method requires modeling the robot and the environment, and using optimization algorithms to solve the optimal control strategy. Although the above methods can realize the gait phase estimation of the lower limb exoskeleton rehabilitation robot, there are still many shortcomings, such as: the perception of human gait information relies on a large number of sensors and the gait phase prediction model has poor generalization ability among different wearers. Summary of the invention

[0004] The present invention aims to solve the technical problems in the prior art and provide an Informer-based personalized prediction method for the gait phase of human lower limbs.

[0005] In order to solve the above technical problems, the technical solutions of the present invention are as follows:

[0006] A personalized prediction method for human lower limb gait phase based on Informer, comprising the following steps:

[0007] Step S1, collecting human lower limb gait motion information through a data acquisition device and forming a data set;

[0008] Step S2, training the pre-model;

[0009] Data standardization of human lower limb gait data;

[0010] The plantar force information is marked with heel strike events, and the gait phase is proportionally mapped to the angle interval;

[0011] Window sliding interception processing: Perform sliding window interception on the standard data matrix to obtain several training data matrices;

[0012] Design the loss function and learning rate of the Informer model;

[0013] Use training sets and validation sets for training neural network models;

[0014] The test set evaluates the generalization ability of the model;

[0015] Step S3, save the trained Informer pre-model;

[0016] Step S4, collecting user's individual lower limb data;

[0017] Collect the thigh swing angle and plantar force information of the wearer's right lower limb under different speeds and movements;

[0018] Step S5: Perform personalized model training based on the Informer pre-model;

[0019] Step S6, saving the Informer personalized model;

[0020] Step S7, collecting the swing angle of the right thigh of the human lower limb through a data acquisition device;

[0021] Collect the wearer's real-time user data to obtain the user's real-time thigh swing angle on the right side of the lower limb under different speeds and movements;

[0022] Step S8: using the model based on the real-time data of the wearer;

[0023] The user's real-time data is standardized to obtain a standardized standard user real-time data vector; the standard user real-time data vector is intercepted by a sliding window to obtain several thigh swing angle vectors on the right side of the user's real-time lower limb; the user's real-time thigh swing angle vector on the right side of the lower limb is input into the Informer personalized model using the Informer personalized model to obtain the user's real-time gait phase sine and cosine values, and the inverse trigonometric function calculation is performed to obtain the user's real-time gait phase prediction value.

[0024] In the above technical solution, in step S2, a sliding window is performed on the standard data matrix to obtain several training data matrices, specifically:

[0025] Obtain the truncation width of the training data matrix to be generated, which is smaller than the width of the standard data matrix; select multiple adjacent columns of data in the standard data matrix of the truncation width to obtain the training data matrix; construct the training set, validation set and test set; design the Informer model network parameters, loss function, optimizer and learning rate; input the training data matrix into the Informer model to obtain the Informer pre-model, and calculate the actual gait phase and the estimated gait phase corresponding to the training data matrix to determine its loss value, and then optimize the Informer model based on the loss value.

[0026] In the above technical solution, step S5 is specifically as follows:

[0027] According to the user's plantar force information, mark the heel touchdown event and mark the gait phase at this moment as 0%, gradually increase linearly to 100% until the next heel touchdown event occurs and is re-marked as 0%, and the gait phase is proportionally mapped to the angle interval, and the corresponding sine and cosine values ​​are calculated;

[0028] The thigh swing angle of the right side of the user's lower limb is normalized to obtain a standardized standard user data matrix;

[0029] A sliding window is performed on the standard user data matrix to obtain several training matrices, including: obtaining a truncation width of the training user data matrix to be generated, the truncation width is smaller than the width of the standard user data matrix; selecting multiple adjacent columns of data in the standard user data matrix of the truncation width to obtain the training user data matrix; and constructing a user training set, a user verification set and a user test set; loading the Informer pre-model, and freezing all parameters of other layers except the fully connected layer, inputting the training user data matrix into the Informer pre-model to obtain the Informer personalized model, and calculating the real gait phase and the estimated gait phase corresponding to the training data matrix to determine its loss value, and then optimizing the Informer model based on the loss value.

[0030] In the above technical solution, in step S5, when constructing the user training set, the user verification set and the user test set, the data set is divided into the user training set, the user verification set and the user test set in a ratio of 7:2:1.

[0031] In the above technical solution, in step S2 and / or step S5, the gait phase is proportionally mapped to the angle interval, specifically: 0%-100% of the gait phase is proportionally mapped to the angle interval of 0-2π.

[0032] In the above technical solution, in step S4 and / or step S7, IMU and plantar force measuring shoes are used to respectively measure the unilateral thigh swing angle of the human lower limb and / or the plantar force information.

[0033] In the above technical solution, in step S4 and / or step S7, an elastic nylon strap is used to tie the IMU sensor 4CM above the knee joint of the right thigh in the coronal plane of the human body; the sampling frequency is 200Hz, and the communication rate is 2MB / s.

[0034] In the above technical solution, the data acquisition device in step S1 and / or step S7 is: an inertial measurement unit and a thin film pressure sensor.

[0035] In the above technical solution, step S3 specifically includes: saving the parameters and structure of the Informer pre-model, including the loss function and optimizer used, the dimension of the input data, and the dimension of the output data.

[0036] In the above technical solution, step S6 specifically includes: saving the network structure, network parameters, and optimizer of the Informer personalized model.

[0037] The present invention has the following beneficial effects:

[0038] The personalized prediction method of the gait phase of the human lower limb based on Informer of the present invention uses an inertial measurement unit and a thin film pressure sensor to collect the unilateral thigh swing angle and plantar force information of the lower limbs of different people of different ages and body shapes under different movements and different walking speeds, and divides the gait according to the plantar force information to obtain the gait phase variable, and trains the Informer model according to the collected data to obtain the Informer pre-model. Then, collect user data and train the Informer pre-model to obtain the Informer personalized model. Finally, when using the Informer personalized model, by collecting the user's real-time data and inputting it into the Informer personalized model, the user's real-time gait phase prediction value is obtained. This method is different from the traditional gait phase prediction method. While increasing the complexity of the model, it avoids the use of a large number of sensors, and increases the accuracy of the model on the basis of reducing the use of hardware equipment, so as to more accurately set the corresponding controller structure and control parameters according to the prior information such as specific human-machine coupling characteristics, and effectively improve the wearing comfort of the exoskeleton user in the human-machine coupling collaborative walking movement. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0040] Figure 1It is a schematic diagram of the structure of the Informer neural network model of the Informer-based personalized prediction method of the lower limb gait phase of the human body of the present invention.

[0041] Figure 2 It is a schematic diagram of the steps of Informer pre-model training in the Informer-based personalized prediction method of human lower limb gait phase of the present invention.

[0042] Figure 3 It is a schematic diagram of the steps of training the Informer personalized model in the Informer-based personalized prediction method of the lower limb gait phase of the human body of the present invention.

[0043] Figure 4 It is a schematic diagram of the steps of the method for using the Informer model in the Informer-based personalized prediction method of the lower limb gait phase of the human body of the present invention.

[0044] Figure 5 It is a schematic diagram of a data acquisition device involved in the Informer-based personalized prediction method of the lower limb gait phase of the human body of the present invention. DETAILED DESCRIPTION

[0045] The inventive concept of the present invention is:

[0046] The Informer-based personalized prediction method for the gait phase of human lower limbs of the present invention trains a human gait phase estimation model through the swing angle of the right thigh of the human lower limb and the plantar force information measured in real time by two sensors, and then uses only one sensor to measure the swing angle of the right thigh of the human lower limb in real time to predict the future gait phase of the exoskeleton robot, and trains a personalized network model according to the specific user individual, so as to achieve personalized gait phase estimation and prediction.

[0047] The present invention is described in detail below with reference to the accompanying drawings.

[0048] like Figure 1-5 As shown, the personalized prediction method of human lower limb gait phase based on Informer of the present invention comprises the following steps:

[0049] Step S1, collecting human lower limb gait motion information through a data acquisition device and forming a data set;

[0050] The inertial measurement unit and thin-film pressure sensor are used to collect the thigh swing angle and plantar force information of the right side of the lower limbs of different people with different body shapes and different movements at different speeds;

[0051] In order to ensure the estimation performance of the trained Informer neural network model, the subjects of the data collection experiment are healthy subjects of different body types (height, weight, gender, and age, etc.). The activity patterns tested are basic daily activities such as walking, uphill, downhill, going upstairs and downstairs. Multiple experiments are conducted at different movement speeds in each activity pattern. For pre-model data collection, volunteers need to be collected and wear collection equipment to complete specific walking patterns and complete the collection of a large amount of lower limb movement data to train the Informer pre-model.

[0052] The present invention has no special restrictions on the acquisition device for measuring the unilateral thigh swing angle and plantar force information of the human lower limb, as long as it is a device in the field that can achieve the aforementioned measurement function. In the experimental process of the present invention, IMU (Intertial Measurement Unit) and plantar force shoes are used to measure the unilateral thigh swing angle and plantar force information of the human lower limb respectively.

[0053] In order to accurately measure the unilateral thigh swing angle and plantar force information of the human lower limb during walking, an elastic nylon strap is used to tie the IMU sensor about 4CM above the right thigh knee joint in the coronal plane of the human body. The sampling frequency is 200Hz and the communication rate is 2MB / s. In order to ensure that the IMU and the plantar force shoe are collected at the same time, a trigger switch is set on the hardware controller (STM32F103) and connected to the external interrupt on the controller. When the switch is turned on, each sensor starts working, and when the switch is turned off, each sensor stops working. The schematic diagram of the collection device is shown in the figure. Figure 5 shown.

[0054] Step S2: training the Informer pre-model. The specific steps include:

[0055] Firstly, the data of human lower limb gait data is standardized; the thigh swing angle of the right side of the lower limb is standardized to obtain a standardized standard data matrix.

[0056] Specifically, the human lower limb gait data is Z-score standardized to obtain standardized data. In this embodiment, the mean and standard deviation of the n time sampling points of each variable are first calculated, and then the historical data is Z-score standardized, and the standardized data is used as the standardized data X, where X is an n×m matrix. The Z-score standardization process is calculated according to the following formula:

[0057]

[0058] Among them, X m represents the value of the mth variable in the standardized data X, D mrepresents the mth variable in historical data D, mean(D m ) represents the average value of the mth variable in the historical data D, std(D m ) represents the standard deviation of the mth variable in the historical data D.

[0059] Then, the plantar force information is marked with heel strike events, and the gait phase is proportionally mapped to the angle interval;

[0060] The plantar force information is used to mark the heel touchdown event, and the phase variable P (0-100%) is mapped to the angle range (0-2π). The reason is that the daily movements of the human body, such as walking, running, going up and down stairs, are all periodic movements. Therefore, in a single cycle, the phase variable P is used to represent the gait process at the current moment, that is, the range of P is 0-100%. However, if the phase variable P is directly used as the output of the Informer neural network model, the phase variable will directly enter the next gait phase cycle before reaching 100% in the current cycle, resulting in discontinuity in the gait phase. Therefore, the phase variable P is converted from rectangular coordinates to polar coordinates and represented by the angle α.

[0061] In this embodiment, the phase variable and the angle are first mapped, and then the angle is substituted into the sine and cosine functions and marked as y 1 and 2 , as the output of the Informer neural network model.

[0062]

[0063] y 1 = sin(α)

[0064] y 2 =cos(α)

[0065] Then, the window sliding interception process is performed; the standard data matrix is ​​subjected to sliding window interception to obtain several training data matrices.

[0066] In the above embodiment, the historical data matrix is ​​intercepted to obtain several training data matrices. The interception width of the training matrix to be generated can be obtained first, and the interception width is smaller than the width of the historical data matrix. Then, multiple adjacent columns of data in the historical data matrix of the interception width are selected to obtain the training data matrix. It should be noted that the interception width is generally much smaller than the width of the historical data matrix, so as to ensure that enough training data matrices are generated. For example, the standardized data X of size m×n can be organized into training data T of size m×d×(n-d+1). T contains (n-d+1) m×d training data matrices as input data. The data set is divided into training set, validation set and test set in a ratio of 7:2:1.

[0067] Then, the loss function and learning rate of the Informer model are designed; the loss function samples the RMSE loss for training, and the optimization algorithm uses the Adam algorithm to optimize the model parameters.

[0068] Depend on Figure 1 As shown in the figure, it is an Informer model architecture diagram. For the Informer model, the model includes an Encoder module and a Decoder module. The training data matrix is ​​input into the Encoder module to obtain the encoding features.

[0069] In this embodiment, the Encoder can generate a feature matrix based on the input training data matrix. Figure 1 As shown in Figure 1, the Encoder contains K modules stacked for feature extraction. Each submodule includes Multi-head Prob Sparse Self-attention and Self-attention distilling. In the Self-attention structure, it contains W Q , W K , W V Matrix, let the input data be X, then:

[0070] Q=W Q X

[0071] K=W K X

[0072] V=W V X

[0073] Q, K, and V represent query, key, and value respectively. Then use the following formula to extract features:

[0074]

[0075] In Multi-head ProbSarse and Self-attention, the sparse matrix Q is used for feature extraction calculation. For each vector q in Q i , calculate the sparse metric M(q i , K):

[0076]

[0077] Select the n q with the largest sparseness metric i Reassemble into Then perform feature extraction calculation. Only the selected n q i , other q i Directly take mean(V), and then S1 and S 0 The results are combined to get the final output S.

[0078]

[0079] In the self-attention distilling operation, higher weights are given to the dominant features with main characteristics, and a focused self-attention feature matrix is ​​generated in the next layer. Self-attention distilling includes one-dimensional convolution (Conv1d), ELU activation function, and maximum pooling (Maxpool) operations.

[0080] I i+1 =MaxPool(ELU(Conv1d[S i ]))

[0081] It should be noted that in this embodiment, in order to predict the gait phase of the future prediction time, the training data matrix needs to be filled. For example, the length of the historical time is d, and the length of the prediction time is t. In addition, the input of the Decoder uses "generative reasoning", and the predicted future phase sequence is filled with 0 values ​​of length t, and the input vector is (d+t). Figure 1 As shown in the figure, the input vector of the Decoder passes through a Masked Multi-head Prob Sparse Self-attention, which can prevent each position from paying attention to the future position, thereby avoiding autoregression. The output of the Masked Multi-headProb Sparse Self-attention is used as the query, and the output feature map of the Encoder is used as the key and value to input a Multi-head Attention. The output of the Decoder passes through a fully connected layer and outputs the gait phase prediction result of the future time length t.

[0082] In this embodiment, the input historical human lower limb data corresponding to all matrix elements of the training data matrix is ​​first obtained, and then the training data matrix is ​​input into the Encoder module to obtain the feature matrix, and then the feature matrix and the input historical gait phase are input into the Decoder module to generate the predicted gait phase, and finally the Informer model is trained based on the predicted gait phase. Thus, by inputting the historical gait phase and the training data matrix into the Informer model, through continuous iteration and optimization, an Informer model that can accurately predict the gait phase of the lower limbs can be obtained.

[0083] Then, use the training set and validation set to train the Informer neural network model, and the test set to test the generalization ability of the model. If the generalization ability does not meet the index requirements, the model will be retrained until its generalization ability meets the index requirements, and the trained model will be saved as the Informer pre-model.

[0084] It is understandable that the training of the model is an iterative process, and the training is performed by continuously adjusting the network parameters of the model until the overall loss function value of the model is less than the preset value, or the overall loss function value of the model no longer changes or changes slowly, the model converges, and a trained model is obtained. In this embodiment, the preset value can be changed according to actual design needs, and no limitation is made here.

[0085] Step S3, save the trained Informer pre-model;

[0086] After training is completed, save the parameters and structure of the Informer pre-model, including the loss function and optimizer used, the dimensions of the input data, and the dimensions of the output data.

[0087] Step S4: The designed collection device collects the wearer's right thigh swing angle and plantar force signals.

[0088] In order to ensure the prediction performance of the personalized model of Informer, the object of the data collection experiment is the personalized data of the wearer user. The activity mode tested is walking, uphill, downhill, upstairs and downstairs and other basic daily activities. In each activity mode, an experiment is conducted at different movement speeds. For personalized model data collection, it is necessary to collect users and wear collection equipment to complete specific walking patterns and complete the collection of lower limb movement data to train the personalized model of Informer.

[0089] During the experiment, IMU (Intertial Measurement Unit) and plantar force shoes were used to measure the unilateral thigh swing angle and plantar force information of the human lower limbs. In order to accurately measure the unilateral thigh swing angle and plantar force information of the human lower limbs during the user's walking, the IMU sensor was tied to the right thigh knee joint of the human coronal plane about 4CM with an elastic nylon strap. Its sampling frequency is 200Hz and the communication rate is 2MB / s.

[0090] Step S5: training the Informer personalized model based on user data.

[0091] In this embodiment, first, the average value and standard deviation of n_user time sampling points of each user data variable are calculated, and then the historical data is Z-score standardized, and the standardized data is used as the standardized data X user , X user It is a matrix of n_use×m. The Z-score standardization process is calculated according to the following formula:

[0092]

[0093] Among them, X m Represents the value of the mth variable in the standardized data X, U m represents the mth variable in the user's historical data U, mean(U m ) represents the average value of the mth variable in the historical data U, std(U m ) represents the standard deviation of the mth variable in the historical data U.

[0094] Then, the user's foot force information is marked with the heel touchdown event, and the phase variable P user (0-100%) is mapped to the angle interval α user (0-2π), because the daily movements of the human body, such as walking, running, going up and down stairs, are all periodic movements. Therefore, in a single cycle, the phase variable P is used user Represents the gait process at the current moment, that is, P user The range is 0-100%. However, if the phase variable P user As the output of the Informer neural network model, the phase variable will directly enter the next gait phase cycle before reaching 100% in the current cycle, resulting in discontinuity of the gait phase. user Convert from rectangular coordinates to polar coordinates and use angle α user express,

[0095] In this embodiment, the phase variable and the angle are first mapped, and then the angle is substituted into the sine and cosine functions and marked as y u1 and u2 , as the output of the Informer personalized model.

[0096]

[0097] y u1 = sin(α user )

[0098] y u2 = cos(α user )

[0099] Window sliding interception processing, in the above embodiment, the user history data matrix is ​​intercepted to obtain several training data matrices, and the interception width of the training data matrix to be generated can be first obtained, the interception width is smaller than the width of the user history data matrix, and then multiple adjacent columns of data in the user history data matrix of the interception width are selected to obtain the user training matrix. And the data set is divided into a user training set, a user verification set, and a user test set in a ratio of 7:2:1.

[0100] In this embodiment, when modifying the network parameters, only the network parameters of the fully connected layer are modified, that is, the transfer learning method is used to freeze the parameters of Informer except the fully connected layer, so that the parameters remain unchanged during the training process, ensuring that the learning ability of the model is not destroyed, and only changing the output result of the model, so that the model can more accurately predict the user's personalized gait phase, thereby ensuring the real-time and flexibility in the human-machine coupling process.

[0101] Step S6: Save the training parameters, network structure and optimizer of the Informer personalized model;

[0102] The Informer personalized model training is completed, and its network structure, network parameters, and optimizer are saved.

[0103] Step S7: The designed acquisition device collects the swing angle of the wearer's right thigh in real time.

[0104] After the training of the Informer personalized model is completed, the saved Informer personalized model can be directly used for the gait phase prediction of the lower limbs of the human body. Before real-time prediction, it is necessary to collect the swing angle of the right thigh of the user's lower limb in real time. The object of the data collection experiment is the real-time data of the wearer's user, and it is collected during basic daily activities such as walking, uphill, downhill, going upstairs and going downstairs. It is required to collect the user's real-time data and wear the collection equipment to complete a specific walking pattern, complete the real-time collection of lower limb motion data, and predict the user's gait phase in real time. Let the collected data be represented by R, where R is n real_time ×1 vector.

[0105] During the collection process, an IMU (Intertial Measurement Unit) is used to collect the user's real-time lower limb unilateral thigh swing angle. In order to accurately collect the lower limb right thigh swing angle of the user during walking, an elastic nylon strap is used to tie the IMU sensor about 4CM above the right thigh knee joint in the coronal plane of the human body. Its sampling frequency is 200Hz and the communication rate is 2MB / s.

[0106] After the training of the Informer personalized model is completed, the saved Informer personalized model can be directly used for the gait phase prediction of the lower limbs of the human body. Before real-time prediction, the swing angle of the right thigh of the lower limb of the human body needs to be collected in real time.

[0107] Step S8: Use the model based on the wearer's real-time data.

[0108] In this embodiment, the user's real-time lower limb thigh swing angle uses the standard deviation and average value of the user's thigh swing angle to perform user real-time data Z-score normalization, and the normalized data is used as the user's real-time normalized data R Z , R Z n real_time × 1 vector. The Z-score normalization process of user real-time data is calculated according to the following formula:

[0109]

[0110] Among them, R Z Represents user real-time standardized data, U m Indicates user data. In this embodiment, m is the first variable, namely U m Indicates the user's thigh swing angle. mean(U m ) represents the average value of the user's thigh swing angle, std(U m ) represents the standard deviation of the user’s thigh swing angle.

[0111] Window sliding interception processing, in the above embodiment, the user real-time data matrix is ​​intercepted to obtain a number of user input data vectors, the interception width is consistent with the interception width of the user historical data, and then the user real-time data of the interception width is obtained.

[0112] After data processing is completed, it is input into the Informer personalized model, and the predicted gait phase representation variable y is obtained at the same time 1 ,y 2 , and then perform inverse trigonometric function calculation to obtain the gait phase variable P.

[0113] In this embodiment, when the actual model is deployed, the corresponding equivalent gait phase variable is calculated according to the following formula:

[0114]

[0115] The training end condition of the deep neural network model is set as the effect of the current model on the validation set meets the requirements or the number of training rounds reaches the set training upper limit.

[0116] The personalized prediction method of the gait phase of the human lower limb based on Informer of the present invention uses an inertial measurement unit and a thin film pressure sensor to collect the unilateral thigh swing angle and plantar force information of the lower limbs of different people of different ages and body shapes under different movements and different walking speeds, and divides the gait according to the plantar force information to obtain the gait phase variable, and trains the Informer model according to the collected data to obtain the Informer pre-model. Then, collect user data and train the Informer pre-model to obtain the Informer personalized model. Finally, when using the Informer personalized model, by collecting the user's real-time data and inputting it into the Informer personalized model, the user's real-time gait phase prediction value is obtained. This method is different from the traditional gait phase prediction method. While increasing the complexity of the model, it avoids the use of a large number of sensors, and increases the accuracy of the model on the basis of reducing the use of hardware equipment, so as to more accurately set the corresponding controller structure and control parameters according to the prior information such as specific human-machine coupling characteristics, and effectively improve the wearing comfort of the exoskeleton user in the human-machine coupling collaborative walking movement.

[0117] Obviously, the above embodiments are merely examples for the purpose of clear explanation, and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the invention.

Claims

1. A personalized prediction method for human lower limb gait phase based on Informer, characterized in that: The following steps are involved: Step S1, collecting human lower limb gait motion information through a data acquisition device and forming a data set; Step S2, training the pre-model; Data standardization of human lower limb gait data; The plantar force information is marked with heel strike events, and the gait phase is proportionally mapped to the angle interval; Window sliding interception processing; Perform sliding window interception on the standard data matrix to obtain several training data matrices; Design the loss function and learning rate of the Informer model; Use training sets and validation sets for training neural network models; The test set evaluates the generalization ability of the model; Step S3, save the trained Informer pre-model; Step S4, collecting user's individual lower limb data; Collect the thigh swing angle and plantar force information of the wearer's right lower limb under different speeds and movements; Step S5: Perform personalized model training based on the Informer pre-model; Step S6, saving the Informer personalized model; Step S7, collecting the swing angle of the right thigh of the human lower limb through a data acquisition device; Collect the real-time data of the wearer and obtain the real-time thigh swing angle of the right side of the lower limb of the user at different speeds and different movements; Step S8: using the model based on the real-time data of the wearer; The user's real-time data is standardized to obtain a standardized standard user real-time data vector; the standard user real-time data vector is intercepted by a sliding window to obtain several thigh swing angle vectors on the right side of the user's real-time lower limb; the user's real-time thigh swing angle vector on the right side of the lower limb is input into the Informer personalized model using the Informer personalized model to obtain the user's real-time gait phase sine and cosine values, and the inverse trigonometric function calculation is performed to obtain the user's real-time gait phase prediction value.

2. The method for personalized prediction of human lower limb gait phase based on Informer according to claim 1, characterized in that: In step S2, a sliding window is performed on the standard data matrix to obtain several training data matrices, specifically: Obtain the truncation width of the training data matrix to be generated, which is smaller than the width of the standard data matrix; select multiple adjacent columns of data in the standard data matrix of the truncation width to obtain the training data matrix; construct the training set, validation set and test set; design the Informer model network parameters, loss function, optimizer and learning rate; input the training data matrix into the Informer model to obtain the Informer pre-model, and calculate the actual gait phase and the estimated gait phase corresponding to the training data matrix to determine its loss value, and then optimize the Informer model based on the loss value.

3. The method for personalized prediction of human lower limb gait phase based on Informer according to claim 1, characterized in that: Step S5 is specifically as follows: According to the user's plantar force information, mark the heel touchdown event and mark the gait phase at this moment as 0%, gradually increase linearly to 100% until the next heel touchdown event occurs and is re-marked as 0%, and the gait phase is proportionally mapped to the angle interval, and the corresponding sine and cosine values ​​are calculated; The thigh swing angle of the right side of the user's lower limb is normalized to obtain a standardized standard user data matrix; Performing sliding window interception on the standard user data matrix to obtain a plurality of training matrices, including: obtaining an interception width of the training user data matrix to be generated, the interception width being smaller than a width of the standard user data matrix; Select multiple adjacent columns of data in the standard user data matrix of the truncated width to obtain the training user data matrix; and construct the user training set, user verification set and user test set; load the Informer pre-model, and freeze all the parameters of other layers except the fully connected layer, input the training user data matrix into the Informer pre-model to obtain the Informer personalized model, and calculate the actual gait phase and the estimated gait phase corresponding to the training data matrix to determine its loss value, and then optimize the Informer model based on the loss value.

4. The method for personalized prediction of human lower limb gait phase based on Informer according to claim 3 is characterized in that: In step S5, when constructing the user training set, the user verification set and the user test set, the data set is divided into the user training set, the user verification set and the user test set in a ratio of 7:2:

1.

5. The method for personalized prediction of human lower limb gait phase based on Informer according to claim 1 or 3, characterized in that: In step S2 and / or step S5, the gait phase is proportionally mapped to the angle interval, specifically: 0%-100% of the gait phase is proportionally mapped to the angle interval of 0-2π.

6. The method for personalized prediction of human lower limb gait phase based on Informer according to claim 1, characterized in that: In step S4 and / or step S7, an IMU and a plantar force measuring shoe are used to respectively measure the unilateral thigh swing angle of a human lower limb and / or plantar force information.

7. The method for personalized prediction of human lower limb gait phase based on Informer according to claim 6, characterized in that: In step S4 and / or step S7, an elastic nylon strap is used to tie the IMU sensor 4CM above the knee joint of the right thigh in the coronal plane of the human body; the sampling frequency is 200Hz, and the communication rate is 2MB / s.

8. The Informer-based personalized prediction method for human lower limb gait phase according to claim 6, characterized in that: The data acquisition device in step S1 and / or step S7 is: an inertial measurement unit and a thin film pressure sensor.

9. The method for personalized prediction of human lower limb gait phase based on Informer according to claim 1, characterized in that: Step S3 is specifically as follows: saving the parameters and structure of the Informer pre-model, including the loss function and optimizer used, the dimension of the input data, and the dimension of the output data.

10. The method for personalized prediction of human lower limb gait phase based on Informer according to claim 1, characterized in that: Step S6 specifically includes: saving the network structure, network parameters, and optimizer of the Informer personalized model.

Images