Lower limb exoskeleton gait track prediction method based on LSTM-KAN fusion model

Abstract

The invention discloses a lower limb exoskeleton gait track prediction method based on an LSTM-KAN fusion model, and the method comprises the steps: collecting human motion data through a sensor assembly, and carrying out the filtering, missing value processing and normalization of the human motion data; then constructing an overall architecture of a prediction model based on an LSTM-KAN network, optimizing parameters of the prediction model by using a particle swarm optimization (PSO) algorithm, extracting key features in the preprocessed data as a training data set, and inputting the training data set into the prediction model for training; and finally, collecting current human body motion data, pre-processing the current human body motion data, inputting the pre-processed current human body motion data into the trained prediction model, and outputting future human body gaits and tracks by the prediction model. And the controller takes a future gait track generated by the prediction model as a reference track to generate a driving signal and sends the driving signal to the actuator so as to realize accurate control of the actuator. By adopting the gyroscope sensor, the acceleration sensor and the pressure sensor for gait estimation, exoskeleton motion control can be effectively improved, and man-machine interaction experience is effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent wearable devices and biomechanical control, and particularly to a lower limb exoskeleton gait trajectory prediction method based on an LSTM-KAN fusion model. Background Art

[0002] With the acceleration of population aging, the number of disabled and semi-disabled people is increasing, and the application of lower limb exoskeleton robots for the rehabilitation training of incomplete injury patients is becoming more and more extensive. In order to assist lower limb rehabilitation gait training, researchers have explored a variety of rehabilitation control strategies. Among them, one of the main control strategies is trajectory tracking control.

[0003] In trajectory tracking control, gait data and gait trajectories are usually used as reference inputs into a prediction model. The prediction model predicts the current joint angle value by learning the characteristics of historical gait data. The predicted value can be further converted into joint torque or acceleration as the reference input for the lower limb exoskeleton robot. In trajectory tracking control, a pre-set gait trajectory, that is, joint angle, angular velocity, angular acceleration, joint torque, etc., is usually used as the reference input for the controller. Therefore, the prediction of gait trajectories can provide a gait reference trajectory for the tracking controller, which helps to improve the timeliness of control.

[0004] For example, the patent application document with the application number 202310811633.X discloses a human gait trajectory prediction method based on an attention mechanism, which includes: collecting original gait data; preprocessing the original gait data; introducing two attention layers into the encoder-decoder architecture based on LSTM to construct a gait trajectory prediction model based on the attention mechanism; training the gait trajectory prediction model with the preprocessed original gait data to obtain a trained gait trajectory prediction model; and performing human gait trajectory prediction based on the trained gait trajectory prediction model.

[0005] The above technical solution can accurately predict the current value of the joint angle in the gait trajectory by introducing the attention mechanism during the prediction process, provide timely gait trajectory values for the control tracking strategy of the lower limb rehabilitation robot, and effectively improve the accuracy of trajectory prediction. However, it has the following disadvantages:

[0006] 1. High computational complexity: The time complexity of the self-attention mechanism increases with the square of the sequence length, making it difficult to meet the real-time requirements of gait prediction. 2. Overfitting prone with small data: It has a large number of parameters and depends on a large amount of data, while gait data usually has limited samples. 3. Weak local temporal modeling: Lack of a gating mechanism, insufficient capture of the microscopic dynamics of continuous joint movement features. 4. Difficult to optimize: When fused with KAN and PSO, the parameter space explodes and the convergence efficiency of particle swarm optimization drops sharply.

[0007] Therefore, it is necessary for us to improve the above-mentioned existing technologies to overcome the above-mentioned defects. Summary of the Invention

[0008] The purpose of the present invention is to provide a lower limb exoskeleton gait trajectory prediction method based on an LSTM-KAN fusion model. The method collects human motion data in real time through a multi-sensor module, combines a gait prediction algorithm to extract future gait trajectory information, and the system further transmits the estimated trajectory to a control system for guiding motion trajectory control, thereby realizing intelligent collaborative motion.

[0009] The above technical purpose of the present invention is achieved through the following technical solutions:

[0010] A lower limb exoskeleton gait trajectory prediction method based on an LSTM-KAN fusion model includes the following steps:

[0011] S1. Data collection and preprocessing: Collect human motion data through a sensor component, and perform filtering, missing value processing, and normalization on it;

[0012] S2. Construct and train a prediction model: Based on the LSTM-KAN network, construct the overall architecture of the prediction model, and use the particle swarm optimization algorithm PSO to optimize the parameters of the prediction model, extract the key features in the preprocessed data as the training data set, and input it into the prediction model for training;

[0013] S3. Gait and trajectory prediction: Collect the current human motion data, preprocess it, and input it into the trained prediction model. The prediction model outputs the future human gait and trajectory.

[0014] Furthermore, the architecture of the prediction model includes

[0015] An input layer, which is used to receive time series data of human gait and trajectory;

[0016] An LSTM layer, which is used to extract temporal features and output hidden states;

[0017] A KAN layer: which is used to perform a non-linear transformation on the hidden states output by the LSTM layer to enhance the model's expression ability;

[0018] A fusion output layer, which is used to combine the results of LSTM and KAN to generate multi-step predictions;

[0019] A PSO optimizer: which is used to globally optimize the parameters of the prediction model.

[0020] Furthermore, the specific prediction method of the prediction model is as follows:

[0021] S1. Preprocessing: The following methods are used to process outliers

[0022]

[0023] Among them, x i : The value of the i-th row. The value after replacing the outlier of the i-th row. n: The total number of rows of data. k: The window size.

[0024] S2. Input processing: Input the step time series data;

[0025] X = [x t-k , x t-k+1 ,..., x t ∈ R k×d (2)

[0026] Among them, d represents the feature dimension, and k represents the time window length;

[0027] S3. LSTM time series modeling: The LSTM layer is responsible for capturing time dependencies and outputting hidden states;

[0028] h t = LSTM(x t , h t-1 ) (3)

[0029] S4. KAN non-linear enhancement: Input the hidden state h t ∈ R n output by LSTM into the KAN layer to enhance the non-linear mapping ability:

[0030]

[0031] Among them, h t,p represents the p-th dimensional feature of the LSTM hidden state, φ q,p (·) represents a learnable univariate function, and Φ q (·) represents the outer composite function;

[0032] S5. Multi-step output prediction: Combine the outputs of LSTM and KAN to generate future m-step predictions;

[0033]

[0034] Among them, [h t ; z t represents the concatenation of the LSTM hidden state and KAN, and W out , b out represent the output layer weights and biases.

[0035] The optimization method of the PSO optimizer is as follows:

[0036]

[0037] Among them, ν i represents the particle velocity, p best , g best represent the individual and global optimal solutions, and W, C1, and C2 represent the inertia weight and acceleration constants.

[0038] A lower limb exoskeleton gait trajectory prediction method based on an LSTM-KAN fusion model, including

[0039] A multi-sensor unit, which is installed on the thighs, calves, and soles of the human body and is used to collect angular velocity, linear acceleration, and plantar force distribution data during human movement;

[0040] An MCU unit, which is used for the multi-sensor unit and stores and transmits the data;

[0041] A memory, which is used to store the angular velocity, linear acceleration, and plantar force distribution data received by the MCU unit;

[0042] A communication unit, which is used to realize two-way communication between the MCU unit and the prediction model;

[0043] A prediction model, which is constructed based on the LSTM-KAN network and combined with the particle swarm optimization algorithm, and is used to model the current and historical gait data to generate future gaits and trajectories;

[0044] A controller, which is used to transmit the predicted gait and trajectory to the actuator in real time;

[0045] An actuator, which is used to generate a drive signal according to the predicted gait and trajectory as reference data to achieve synchronous movement with the user's gait and trajectory.

[0046] A power supply module, which is used to supply power to the above modules.

[0047] Furthermore, the multi-sensor unit includes an IMU module and a pressure sensor module. The IMU module includes a gyroscope sensor and an acceleration sensor, and the pressure sensor module includes a pressure sensor.

[0048] Furthermore, the prediction model uses the data sent by the multi-sensor unit to extract the gait cycle, stride, and joint angles therein for model training.

[0049] Furthermore, the prediction model is based on current and historical data, and collects sensor data in real time to generate future gait trajectories, which include joint angle changes and gait change curves.

[0050] Furthermore, the controller uses the future gait trajectory generated by the prediction model as a reference trajectory to generate a drive signal and sends it to the actuator to achieve precise control of the actuator.

[0051] In summary, the present invention has the following beneficial effects:

[0052] Based on the gyroscope sensor, the acceleration sensor, and the pressure sensor, human motion data is collected; the future trajectory is predicted according to historical data to obtain a reference trajectory, and the exoskeleton trajectory control is performed with this as a reference to achieve compliant control and better human-machine interaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a schematic diagram of the lower limb exoskeleton gait trajectory prediction method based on the LSTM-KAN fusion model of the present invention.

[0054] Figure 2 It is a system schematic diagram of the lower limb exoskeleton gait trajectory prediction method based on the LSTM-KAN fusion model of the present invention.

[0055] Figure 3 It is a schematic diagram of the installation position of the IMU module of the present invention.

[0056] Figure 4 It is a schematic diagram of the installation position of the pressure sensor module of the present invention.

[0057] Figure 5 It is a joint trajectory diagram of three different persons collected by the collection device of the present invention.

[0058] Figure 6 It is a PSO optimization flowchart of the model of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0059] In order to make the technical means, creative features, achieved purposes, and functions of the present invention easy to understand, the present invention will be further described below in conjunction with the drawings and specific embodiments.

[0060] As Figures 1 to 6 shown, a lower limb exoskeleton gait trajectory prediction method based on the LSTM-KAN fusion model proposed by the present invention includes the following steps:

[0061] S1. Data collection and preprocessing: Collect human motion data through the sensor assembly, and perform filtering, missing value processing, and normalization on it;

[0062] S2. Construct and train the prediction model: Based on the network of the LSTM-KAN fusion model, construct the overall architecture of the prediction model, and use the particle swarm optimization algorithm PSO to optimize the parameters of the prediction model, extract the key features in the preprocessed data as the training data set and input it into the prediction model for training;

[0063] S3. Gait and Trajectory Prediction: Collect the current human motion data, preprocess it, and input it into the trained prediction model. The prediction model outputs the future human gait and trajectory.

[0064] The architecture of the prediction model includes

[0065] An input layer, which is used to receive the time series data of human gait and trajectory;

[0066] An LSTM layer, which is used to extract temporal features and output hidden states;

[0067] A KAN layer: which is used to perform a non - linear transformation on the hidden states output by the LSTM layer to enhance the model's expressive ability;

[0068] A fusion output layer, which is used to combine the results of LSTM and KAN to generate multi - step predictions;

[0069] A PSO optimizer: which is used to globally optimize the parameters of the prediction model.

[0070] The specific prediction method of the prediction model is as follows:

[0071] S1. Preprocessing: The following method is used to process outliers

[0072]

[0073] where, x i : The value of the i - th row. The value after replacing the outlier of the i - th row. n: The total number of rows of data. k: The window size.

[0074] S2. Input Processing: Input the step time series data;

[0075] X = [x t-k , x t-k+1 ,..., x t ∈ R k×d (2)

[0076] where, d represents the feature dimension, and k represents the time window length;

[0077] S3. LSTM Temporal Modeling: The LSTM layer is responsible for capturing time - dependent relationships and outputting hidden states;

[0078] h t = LSTM(x t , h t-1 ) (3)

[0079] S4. KAN Non - linear Enhancement: The hidden state h output by the LSTM t ∈ R nInput the KAN layer to enhance the non - linear mapping ability:

[0080]

[0081] Among them, h t,p represents the p - dimensional feature of the LSTM hidden state, φ q,p (·) represents a learnable univariate function, Φ q (·) represents the outer - layer combination function;

[0082] S5. Multi - step output prediction: Combine the outputs of LSTM and KAN to generate future m - step predictions;

[0083]

[0084] Among them, [h t ; z t represents the concatenation of the LSTM hidden state and KAN, W out , b out represent the output - layer weights and biases.

[0085] Use the particle swarm optimization algorithm to solve the optimal hyperparameters of the model. The PSO algorithm process is as Figure 6 shown. The optimization method of the PSO optimizer is as follows:

[0086]

[0087] Among them, ν i represents the particle velocity, p best , g best represent the individual and global optimal solutions, and W, C1, C2 represent the inertia weight and acceleration constants.

[0088] To enhance the search ability of the particle swarm optimization algorithm (PSO), Gaussian noise is introduced into the velocity update formula. The addition of Gaussian noise can effectively expand the search range of particles, help particles jump out of local optima, and thus improve the global search ability and convergence speed of the algorithm. The formula is as follows:

[0089]

[0090] Table 1 Particle swarm optimizer parameter settings

[0091]

[0092] A lower - limb exoskeleton gait trajectory prediction method based on the LSTM - KAN fusion model, including

[0093] Multi - sensor units, which are installed on the thighs, calves and soles of the human body, and are used to collect data on angular velocity, linear acceleration and plantar force distribution during human movement;

[0094] An MCU unit, which is used for a multi-sensor unit and stores the transmitted data;

[0095] A memory, which is used to store the angular velocity, linear acceleration, and plantar force distribution data received by the MCU unit;

[0096] A communication unit, which is used to realize two-way communication between the MCU unit and the prediction model;

[0097] A prediction model, which is constructed based on the LSTM-KAN fusion model network and combines the particle swarm optimization algorithm, and is used to model the current and historical gait data to generate future gaits and trajectories;

[0098] A controller, which is used to transmit the predicted gait and trajectory to the actuator in real time;

[0099] An actuator, which is used to generate a drive signal according to the predicted gait and trajectory as reference data to achieve synchronous movement with the user's gait and trajectory.

[0100] A power supply module, which is used to supply power to the above modules.

[0101] The multi-sensor unit includes an IMU module and a pressure sensor module. The IMU module A includes a gyroscope sensor and an acceleration sensor, and the pressure sensor module B includes a pressure sensor.

[0102] The prediction model uses the data sent by the multi-sensor unit, extracts the gait cycle, stride, and joint angle therein for model training. As Figure 5 shown, there are differences between individuals, and this model collects the gait data of multiple people, enhancing the robustness of the prediction model.

[0103] The prediction model is based on current and historical data, and collects sensor data in real time to generate future gait trajectories, which include joint angle changes and gait change curves.

[0104] The controller uses the future gait trajectory generated by the prediction model as a reference trajectory to generate a drive signal and sends it to the actuator to achieve precise control of the actuator.

[0105] To verify the rationality and effectiveness of the technical solution of the present invention, this embodiment uses the data of multiple people collected by this device for experiments, and uses R 2 , MSE, RMSE, MAE as objective evaluation indicators for the classification results. This embodiment is implemented based on the deep learning framework Pytorch and uses a graphics processing unit (GPU) to accelerate the operation, with 32GB of memory and an Nvidia GeForce RTX3080 SUPER graphics card.

[0106] Table 2 is a comparison table of data training, as follows:

[0107]

[0108] The experimental results in Table 1 prove that R 2 > 93%, MSE <0.06, and its performance is better than other models, achieving accurate prediction of the target. It can be seen from the above embodiments that the present invention can greatly improve the prediction of gait trajectories and enhance the precise control of exoskeleton devices.

[0109] In this article, the orientation or positional relationship indicated by the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", "vertical", "horizontal", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of clearly expressing the technical solution and description, and therefore cannot be construed as a limitation to the present invention.

[0110] In this article, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, and in addition to the listed elements, it may also include other elements not expressly listed.

[0111] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and the above embodiments and descriptions in the specification only illustrate the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements fall within the scope of the present invention claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A lower limb exoskeleton gait trajectory prediction method based on an LSTM-KAN fusion model, characterized in that, It includes the following steps: S1. Set up a data acquisition experimental device, and collect human lower limb gyroscope data, acceleration data, and pressure data through sensors; S2. Data collection and preprocessing: Collect human motion data through a sensor component, and perform filtering, missing value processing, and normalization on it; S3. Construct and train a prediction model: Based on the LSTM-KAN fusion model, construct the overall architecture of the prediction model, and use the particle swarm optimization algorithm PSO to optimize the parameters of the prediction model. Extract the key features from the preprocessed data as the training data set and input it into the prediction model for training; S4. Gait and trajectory prediction: Collect the current human motion data, preprocess it, and input it into the trained prediction model. The prediction model outputs the future human gait and trajectory.

2. The lower limb exoskeleton gait trajectory prediction method based on the LSTM-KAN fusion model according to claim 1, characterized in that The architecture of the prediction model includes an input layer, which is used to receive the time series data of human gait and trajectory; an LSTM layer, which is used to extract temporal features and output hidden states; a KAN layer: which is used to perform nonlinear transformation on the hidden states output by the LSTM layer to enhance the model's expression ability; a fusion output layer, which is used to combine the results of LSTM and KAN to generate multi-step predictions; a PSO optimizer: which is used to globally optimize the parameters of the prediction model.

3. The lower limb exoskeleton gait trajectory prediction method based on the LSTM-KAN fusion model according to claim 2, wherein, The specific prediction method of the prediction model is as follows: S1. Preprocessing: The following methods are used to process outliers where x i : the value of the i-th row. The value after replacing the outlier of the i-th row. n: the total number of rows of data. k: the window size. S2. Input processing: Input the step time series data; X = [x t-k , x t-k+1 ,..., x t ∈ R k×d (2) where d represents the feature dimension and k represents the time window length; S3. LSTM temporal modeling: The LSTM layer is responsible for capturing time dependencies and outputting hidden states; h t = LSTM(x t , h t-1 ) (3) S4. KAN Nonlinear Enhancement: The hidden state h output by the LSTM t ∈R n is input into the KAN layer to enhance the non - linear mapping ability: where h t,p represents the p-th dimensional feature of the LSTM hidden state, and φ q,p (·) represents a learnable univariate function, and Φ q (·) represents an outer composite function; S5. Multi-step output prediction: Combine the outputs of LSTM and KAN to generate future m-step predictions; Among them, [h t ; z t represents the concatenation of the LSTM hidden state and KAN, W out , b out represent the output layer weights and biases.

4. The lower limb exoskeleton gait trajectory prediction method based on the LSTM-KAN fusion model according to claim 3, characterized in that The optimization method of the PSO optimizer is as follows: Among them, ν i represents the particle velocity, p best , g best represent the individual and global optimal solutions, and W, C1, and C2 represent the inertia weight and acceleration constants.

5. A lower limb exoskeleton gait trajectory prediction method based on an LSTM-KAN fusion model, characterized in that, It includes a multi-sensor unit, which is installed on the thigh, calf, and sole of the human body and is used to collect angular velocity, linear acceleration, and plantar force distribution data during human motion; an MCU unit, which is used for the multi-sensor unit and stores and transmits the data; a memory, which is used to store the angular velocity, linear acceleration, and plantar force distribution data received by the MCU unit; a communication unit, which is used to realize two-way communication between the MCU unit and the prediction model; a prediction model, which is constructed based on LSTM and KAN networks and combines the particle swarm optimization algorithm, and is used to model the current and historical gait data to generate future gaits and trajectories; a controller, which is used to transmit the predicted gait and trajectory to the actuator in real time; an actuator, which is used to generate a drive signal according to the predicted gait and trajectory as reference data to achieve synchronous movement with the user's gait and trajectory. a power supply module, which is used to supply power to the above modules.

6. The lower limb exoskeleton gait trajectory prediction method based on the LSTM-KAN fusion model according to claim 5, wherein The multi-sensor unit includes an IMU module and a pressure sensor module. The IMU module includes a gyroscope sensor and an acceleration sensor, and the pressure sensor module includes a pressure sensor.

7. The lower limb exoskeleton gait trajectory prediction method based on the LSTM-KAN fusion model according to claim 5, characterized in that, The prediction model uses the data sent by the multi-sensor unit, extracts the gait cycle, stride, and joint angles from it for model training.

8. The lower limb exoskeleton gait trajectory prediction method based on the LSTM-KAN fusion model according to claim 7, characterized in that The prediction model is based on current and historical data, and real-time collects sensor data to generate future gait trajectories, which include joint angle changes and gait change curves.

9. The lower limb exoskeleton gait trajectory prediction method based on the LSTM-KAN fusion model according to claim 8, wherein The controller uses the future gait trajectory generated by the prediction model as a reference trajectory to generate a driving signal and sends it to the actuator to achieve precise control of the actuator.

Images