Smart ski instructor plantar pressure acquisition system and data processing method

Abstract

This invention discloses an intelligent foot pressure acquisition system and data processing method for ski instructors, belonging to the field of intelligent sports equipment and data processing technology. Its purpose is to solve the problems of large measurement errors, low processing efficiency, and insufficient posture recognition accuracy in existing technologies. The invention includes a flexible array pressure sensor module, a microcontroller information processing terminal, a Bluetooth communication module, and a host computer. The microcontroller information processing terminal, with a microcontroller as its core, integrates a voltage conversion module and a signal acquisition and amplification circuit. The pressure analog signal output from the flexible array pressure sensor module is input to the signal acquisition and amplification circuit, amplified, and then output to the microcontroller's ADC acquisition channel. The microcontroller performs analog-to-digital conversion and data preprocessing on the analog signal, and then outputs it to the Bluetooth communication module via a UART interface. The host computer communicates with the microcontroller information processing terminal via the Bluetooth communication module.

Description

Technical Field

[0001] This invention belongs to the field of intelligent sports equipment and data processing technology, specifically relating to an intelligent ski instructor plantar pressure acquisition system and data processing method. Background Technology

[0002] With the popularization of skiing, the number of skiers in China continues to grow. In traditional ski instruction, instructor guidance is significantly delayed, failing to capture and provide feedback on skiers' posture deviations in real time. Existing intelligent ski instructor systems are mostly limited to offline services and have not effectively solved core problems such as delayed information feedback and insufficient posture recognition accuracy. Plantar pressure data, as a key indicator reflecting the distribution of a skier's center of gravity and posture characteristics, directly determines the reliability of posture recognition based on its acquisition accuracy and processing efficiency. However, existing technologies suffer from three major problems: first, cross-coupling interference from flexible array pressure sensors leads to large measurement errors; second, high-dimensional plantar pressure data is redundant, resulting in low processing efficiency; and third, traditional classification algorithms are prone to getting trapped in local optima when processing nonlinear, high-dimensional data, making it difficult to meet practical needs in terms of posture recognition accuracy. Summary of the Invention

[0003] The purpose of this invention is to propose an intelligent foot pressure acquisition system and data processing method for ski instructors, which solves the problems of large measurement errors, low processing efficiency, and insufficient posture recognition accuracy in existing technologies.

[0004] To achieve the above objectives, the intelligent ski instructor plantar pressure acquisition system of the present invention includes a flexible array pressure sensor module, a single-chip microcomputer information processing terminal, a Bluetooth communication module, and a host computer. The flexible array pressure sensor module adopts an ultra-thin flexible sensor array based on the piezoresistive effect, and the flexible array pressure sensor module acquires pressure through a secondary measurement method; The microcontroller-based information processing terminal integrates a voltage conversion module and a signal acquisition and amplification circuit, with a sampling frequency of 500Hz. The pressure analog signal output from the flexible array pressure sensor module is input to the signal acquisition and amplification circuit, amplified, and then output to the microcontroller's ADC acquisition channel. The microcontroller performs analog-to-digital conversion and data preprocessing on the analog signal before outputting it to the Bluetooth communication module via the UART interface. The voltage conversion module converts the lithium battery input voltage into a stable voltage, powering the microcontroller, the signal acquisition and amplification circuit, and the Bluetooth communication module. The host computer communicates with the microcontroller information processing terminal via a Bluetooth communication module. The host computer is configured with a data display, storage, and interactive interface to monitor pressure data in real time.

[0005] The flexible array pressure sensor module is a 6-row × 4-column array.

[0006] The ultra-thin flexible sensor based on the piezoresistive effect is the FlexiForce plantar pressure sensor.

[0007] The aforementioned secondary measurement method includes two scanning processes: A resistor array with n rows and m columns is set up, and the measurement point is placed in front of the row line selection switch; During the first scan, the common terminal of the line selection switch is connected to V. cc The column line selection switch is grounded (i.e. connected to CND), and the row line selection switch is switched n times. During the second scan, the common terminal of the row line selector switch is grounded (i.e., connected to CND), and the common terminal of the column line selector switch is connected to V. cc For each column line gating switch that toggles once, and the row line gating switch that toggles n times, the total number of transitions is: n×m; Where: n is the number of rows, and m is the number of columns; The resistance value R to be measured is calculated by two voltage scans. 22 for: R 22 =Vout1 / Vout2×R0; Where: Vout1 is the output voltage of the first scan; Vout2 is the output voltage of the second scan; R0 is the reference resistor.

[0008] The microcontroller is an STM32F103C8T6.

[0009] The Bluetooth communication module uses version 4.0 with a baud rate of 115200 baud.

[0010] The data processing method based on the intelligent ski instructor plantar pressure acquisition system includes the following steps: Step 1: Data Acquisition. Foot pressure data for five skiing postures—snowball downhill, snowball left turn, snowball right turn, snowball braking, and parallel turn—are collected using a flexible array pressure sensor module. At least 2.2 seconds of valid data are captured for each posture. Step 2: Data Preprocessing: Missing values ​​are handled and the collected data is normalized. The normalization formula is: ; Where: X' I The data is after normalization; X i This is the original data; μ is the mean of the data; σ is the standard deviation of the data; Step 3: Data dimensionality reduction. Weighted principal component analysis is used. First, representative data for each feature are determined by averaging multiple measurements. Then, the weights are calculated based on the total ground pressure. The weight calculation formula is as follows: ; in: Representative data for features; F represents the total pressure on the ground; The data is reduced to 5 dimensions by using an error threshold. Step 4: Attitude recognition. Based on the adaptive particle swarm optimization algorithm with dynamic Lévy flight perturbation, the BP neural network is improved to classify the dimensionality-reduced data and output the skiing attitude recognition results.

[0011] The data processing method further includes: Step 5: Result verification. The results are compared and verified with four optimization algorithms: LM, Bayesian regularization, SCG, and standard PSO, using four metrics: accuracy, precision, recall, and F1 score.

[0012] The 2.2 seconds of valid data is discontinuous.

[0013] Step three of the weighted principal component analysis method specifically includes the following steps: Step 1: Construct an m×n original data matrix X, where m is the number of samples and n is the number of features. The data corresponding to each variable can be denoted as X1, X2, X3, X4, ... X n The original data matrix is ​​denoted as X, specifically as follows: ; Step 2: Calculate representative data for each feature. Specifically: ; Step 3: Measure the total ground pressure F of the test subject in the corresponding posture, and calculate the weights P of each feature. L Specifically: ; Step 4: Construct a weighted data matrix Specifically: ; Step 5: [Regarding...] Perform mean normalization to 0 and calculate the covariance matrix; Step 6: Solve for the eigenvalues ​​and eigenvectors of the covariance matrix, select the eigenvectors corresponding to the first k largest eigenvalues ​​to construct the projection matrix, and reduce the data dimension to k dimensions, where k=5.

[0014] The implementation steps of the adaptive particle swarm optimization algorithm based on dynamic Lévy flight perturbation described in step four are as follows: Step 1: Initialization Phase Parameter Settings: Determine the particle swarm size N and the maximum number of iterations T. max Search space upper and lower bounds [Lb,μb], dimension dim, initial inertia weight ω, learning factors c1, c2; Chaotic mapping initialization: Initial particle positions are generated using chaotic sequences such as Logistic mapping; Velocity initialization: Randomly generate the initial velocity of the particles; Step 2: Fitness Calculation and Optimal Update Calculate fitness: Evaluate the fitness value of each particle based on the objective function; Update individual best (pBest): If the current fitness is better than the historical best, then update the individual's best position; Update Global Best (gBest): Records the best position among all particles at present; Step 3: Dynamic Lévy flight perturbation Lévy step size generation: Generate random step size based on Lévy distribution L ( beta ), L ( beta The value is usually taken as 1.5, and the formula is: Dynamic adjustment mechanism: The step size is dynamically scaled based on the number of iterations. Where: α is the initial step size factor, which decreases with each iteration to balance exploration and development; Position Update: Apply Levy perturbation to some poorly fit individual particles: In the formula: x rand The particle position is randomly selected; Step 4: Adaptive parameter adjustment. The inertia weight ω is adjusted either linearly decreasing or based on population diversity. The learning factors c1 and c2 are dynamically adjusted according to the convergence state. For example, increasing c1 in the early stage enhances exploration, and increasing c2 in the later stage accelerates convergence. Step 5: Speed ​​and Position Update Standard (PSO) Update: Boundary handling: If a particle goes out of bounds, a reflection or random reset strategy is used; Step 6: Chaotic Perturbation Stagnation Detection: If gBest fails to update for k consecutive times, a chaotic perturbation is triggered. Perturbation Method: A chaotic sequence is generated using a Logistic mapping, and the positions of some particles are reinitialized. Step 7: Termination condition reaches the maximum number of iterations T max Or the fitness value changes less than the threshold ε.

[0015] The BP neural network described in step four has a "5-12-5" structure, with 5 neurons in the input layer, 12 neurons in the hidden layer, and 5 neurons in the output layer. The optimization parameters include: 30 particles, 50 maximum iterations, inertia weight range of 0.4-0.9, initial learning factor values ​​of c1=1.5 and c2=2.5, diversity trigger threshold of 0.05, Sigmoid activation function, and Softmax function for the output layer.

[0016] Step three also includes comparing and verifying four dimensionality reduction methods: PCA, SPCA, MDE, and SPE. It analyzes feature correlation through heatmaps and data classification clustering through scatter matrix plots to ensure that the weighted PCA method retains ≥95% of the original data features.

[0017] The beneficial effects of this invention are as follows: The intelligent ski instructor plantar pressure acquisition system and data processing method of this invention have clear power supply, signal processing and control logic of the internal components of the single-chip microcomputer information processing terminal. The secondary measurement method effectively eliminates sensor cross-coupling interference, eliminates the need for a negative power supply, simplifies the circuit structure while reducing power consumption, and significantly improves measurement accuracy, making it suitable for long-term skiing monitoring. The data preprocessing stage improves data quality through outlier removal and normalization, laying the foundation for subsequent dimensionality reduction and classification. The weighted PCA dimensionality reduction method allocates weights according to the importance of pressure distribution, retaining more key features compared to traditional PCA, SPCA and other methods, thus improving data processing efficiency. The DLF-PSO improved BP neural network optimizes population diversity through Lévy flight perturbation, achieving high classification accuracy and precision, significantly outperforming traditional algorithms such as LM and Bayesian regularization, and providing accurate and reliable posture recognition. The system achieves real-time ski posture recognition and feedback with a response time of ≤0.5 seconds, helping skiers correct their posture without professional coaching guidance, greatly reducing the cost of learning to ski, filling the gap in domestic real-time feedback intelligent ski instructor systems, and has broad application prospects. Attached Figure Description

[0018] Figure 1 This is a structural diagram of an intelligent ski instructor plantar pressure acquisition system according to the present invention; Figure 2 This is a schematic diagram of a secondary measurement method for an intelligent ski instructor plantar pressure acquisition system according to the present invention. Figure 3 This is a flowchart of a data processing method based on an intelligent ski instructor plantar pressure acquisition system according to the present invention. Detailed Implementation

[0019] The embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0020] See Figure 1 and Figure 2 The intelligent ski instructor plantar pressure acquisition system of the present invention includes a flexible array pressure sensor module, a single-chip microcomputer information processing terminal, a Bluetooth communication module and a host computer; The flexible array pressure sensor module adopts an ultra-thin flexible sensor array based on the piezoresistive effect. The flexible array pressure sensor module acquires pressure through a secondary measurement method to eliminate cross-coupling interference. The microcontroller-based information processing terminal integrates a voltage conversion module and a signal acquisition and amplification circuit, with a sampling frequency of 500Hz. The pressure analog signal output from the flexible array pressure sensor module is input to the signal acquisition and amplification circuit, amplified, and then output to the microcontroller's ADC acquisition channel. The microcontroller performs analog-to-digital conversion and data preprocessing on the analog signal before outputting it to the Bluetooth communication module via the UART interface. The voltage conversion module converts the lithium battery input voltage into a stable voltage, powering the microcontroller, the signal acquisition and amplification circuit, and the Bluetooth communication module. The microcontroller is based on an STM32F103C8T6, and a minimum system (power supply circuit, reset circuit, 8MHz crystal oscillator circuit) is built. Programming and debugging are completed through the JTAG interface. The voltage conversion module: the input terminal of the AMS1117-3.3V chip is connected to a 3.7V lithium battery, and the measured output voltage is 3.302V, with ripple voltage <5mV, and the power supply stability meets the standards. Signal acquisition and amplification circuit: The AD8221 instrumentation amplifier has an input of 0-1V analog signal and an output voltage of 0-10.02V. The amplification factor error is <0.2%, and the signal amplification effect meets the standard. The host computer communicates with the microcontroller information processing terminal via a Bluetooth communication module. The host computer is configured with a data display, storage, and interactive interface to monitor pressure data in real time.

[0021] The flexible array pressure sensor module is a 6-row × 4-column array. FlexiForce piezoresistive sensors, model A201, are selected, with a sensor spacing of 1cm. Silver wires are led out and soldered to the input terminal of the signal acquisition and amplification circuit. Applying pressure from 0-500kPa using a standard pressure source verifies the sensor output signal: 0.02V output voltage at 0kPa, 0.98V output voltage at 500kPa, and linearity error <2%, meeting design requirements.

[0022] The ultra-thin flexible sensor based on the piezoresistive effect is the FlexiForce plantar pressure sensor. It consists of a sensitive layer, a protective layer, encapsulation, and an interface and circuitry. The sensitive layer is composed of two layers of polyester fiber film (polyimide is used in high-temperature models), each coated with conductive silver material and pressure-sensitive ink, bonded together with an adhesive to form the sensing area. The sensing area is a circular region with a diameter of 0.95 cm, from which silver wires are led to transmit signals. The protective layer and encapsulation surface are covered with wear-resistant and waterproof materials to ensure the sensor's durability in complex environments. For example, in wearable insoles, the sensor is embedded in a layered structure; the top layer uses soft cowhide to conform to the shape of the foot, while the bottom layer is for abrasion protection. The interface and circuitry are equipped with solderable connectors or integrated signal processing circuitry to convert resistance changes into voltage or digital signals.

[0023] The aforementioned secondary measurement method includes two scanning processes: A resistor array with n rows and m columns is set up, and the measurement point is placed in front of the row line selection switch; During the first scan, the common terminal of the line selection switch is connected to V. cc The column line selection switch is grounded (i.e. connected to CND), and the row line selection switch is switched n times. During the second scan, the common terminal of the row line selector switch is grounded (i.e., connected to CND), and the common terminal of the column line selector switch is connected to V. cc For each column line gating switch that toggles once, and the row line gating switch that toggles n times, the total number of transitions is: n×m; Where: n is the number of rows, and m is the number of columns; The resistance value R to be measured is calculated by two voltage scans. 22 for: R 22 =Vout1 / Vout2×R0; Where: Vout1 is the output voltage of the first scan; Vout2 is the output voltage of the second scan; R0 is the reference resistor.

[0024] The microcontroller is an STM32F103C8T6.

[0025] The Bluetooth communication module uses version 4.0. The HC-05 Bluetooth communication module connects to the STM32F103C8T6 via a UART interface, configured with a baud rate of 115200. The host computer, developed based on LabVIEW 2018, receives Bluetooth signals via a USB-to-TTL module and displays pressure data and heatmaps in real time, with data stored in CSV format.

[0026] See Figure 3The data processing method based on the intelligent ski instructor plantar pressure acquisition system includes the following steps: Step 1: Data Acquisition. Foot pressure data for five skiing postures—snowball downhill, snowball left turn, snowball right turn, snowball braking, and parallel turn—are collected using a flexible array pressure sensor module. At least 2.2 seconds of valid data are captured for each posture. Data collection process: (1) Volunteers wear ski boots and flexible insoles with integrated sensors, adapt to the pressure test platform for 5 minutes, and become familiar with the key points of the five skiing postures; (2) Start the host computer and the data collection system, send the "start measurement" command, and the system starts collecting data; (3) Volunteers complete the five postures in sequence: snowplow straight descent, snowplow left turn, snowplow right turn, snowplow braking, and parallel turn. Each posture is held for 3 seconds (the sensor value is stable for 3 seconds). Each posture is collected 10 times, with a 1-minute interval between two collections to avoid fatigue affecting the standardization of the movements; (4) After the data collection is completed, the host computer automatically stores the data, extracts 2.2 seconds of valid data for each posture (including 1100 sampling points), and generates the original data file; Step 2: Data Preprocessing: Missing values ​​are handled and the collected data is normalized. The normalization formula is: ; Where: X' I The data is after normalization; X i This is the original data; μ is the mean of the data; σ is the standard deviation of the data; Step 3: Data dimensionality reduction. Weighted principal component analysis (PCA) is used. First, representative data for each feature are determined by averaging multiple measurements. Then, the weights are calculated based on the total ground pressure. The weight calculation formula is as follows: ; in: Representative data for features; F represents the total pressure on the ground; The data is reduced to 5 dimensions by using an error threshold. Step 4: Attitude recognition. Based on the adaptive particle swarm optimization algorithm with dynamic Lévy flight perturbation, the BP neural network is improved to classify the dimensionality-reduced data and output the skiing attitude recognition results.

[0027] The data processing method further includes: Step 5: Result verification. The results are compared and verified with four optimization algorithms: LM, Bayesian regularization, SCG, and standard PSO, using four metrics: accuracy, precision, recall, and F1 score.

[0028] The 2.2 seconds of valid data is discontinuous.

[0029] Step three of the weighted principal component analysis method specifically includes the following steps: Step 1: Construct an m×n original data matrix X, where m is the number of samples and n is the number of features. The data corresponding to each variable can be denoted as X1, X2, X3, X4, ... X n The original data matrix is ​​denoted as X, specifically as follows: ; Step 2: Calculate representative data for each feature. Specifically: ; Step 3: Measure the total ground pressure F of the test subject in the corresponding posture, and calculate the weights P of each feature. L Specifically: ; Step 4: Construct a weighted data matrix Specifically: ; Step 5: [Regarding...] Perform mean normalization to 0 and calculate the covariance matrix; Step 6: Solve for the eigenvalues ​​and eigenvectors of the covariance matrix, select the eigenvectors corresponding to the first k largest eigenvalues ​​to construct the projection matrix, and reduce the data dimension to k dimensions, where k=5.

[0030] The implementation steps of the adaptive particle swarm optimization algorithm based on dynamic Lévy flight perturbation described in step four are as follows: Step 1: Initialization Phase Parameter Settings: Determine the particle swarm size N and the maximum number of iterations T. max Search space upper and lower bounds [Lb,μb], dimension dim, initial inertia weight ω, learning factors c1, c2; Chaotic mapping initialization: Initial particle positions are generated using chaotic sequences such as Logistic mapping to enhance population diversity; Velocity initialization: Randomly generate the initial velocity of the particles; Step 2: Fitness Calculation and Optimal Update Calculate fitness: Evaluate the fitness value of each particle based on the objective function; Update individual best (pBest): If the current fitness is better than the historical best, then update the individual's best position; Update Global Best (gBest): Records the best position among all particles at present; Step 3: Dynamic Lévy flight perturbation Lévy step size generation: Generate random step size based on Lévy distribution L ( beta ), L ( beta The value is usually taken as 1.5, and the formula is:

[0031] Dynamic adjustment mechanism: The step size is dynamically scaled based on the number of iterations.

[0032] Where: α is the initial step size factor, which decreases with each iteration to balance exploration and development; Position Update: Apply Levy perturbation to some poorly fit individual particles:

[0033] In the formula: x rand The particle position is randomly selected; Step 4: Adaptive parameter adjustment. The inertia weight ω is adjusted either linearly decreasing or based on population diversity.

[0034] The learning factors c1 and c2 are dynamically adjusted according to the convergence state. For example, increasing c1 in the early stage enhances exploration, and increasing c2 in the later stage accelerates convergence. Step 5: Speed ​​and Position Update Standard (PSO) Update:

[0035] Boundary handling: If a particle goes out of bounds, a reflection or random reset strategy is used; Step 6: Chaotic Perturbation Stagnation Detection: If gBest fails to update for k consecutive times, a chaotic perturbation is triggered. Perturbation Method: A chaotic sequence is generated using a Logistic mapping to reinitialize the positions of some particles, avoiding local optima; Step 7: Termination condition reaches the maximum number of iterations T max Or the fitness value changes less than the threshold ε.

[0036] The BP neural network described in step four has a "5-12-5" structure, with 5 neurons in the input layer, 12 neurons in the hidden layer, and 5 neurons in the output layer. The optimization parameters include: 30 particles, 50 maximum iterations, inertia weight range of 0.4-0.9, initial learning factor values ​​of c1=1.5 and c2=2.5, diversity trigger threshold of 0.05, Sigmoid activation function, and Softmax function for the output layer.

[0037] Step three also includes comparing and verifying four dimensionality reduction methods: PCA, SPCA, MDE, and SPE. It analyzes feature correlation through heatmaps and data classification clustering through scatter matrix plots to ensure that the weighted PCA method retains ≥95% of the original data features.

Claims

1. An intelligent plantar pressure data acquisition system for ski instructors, characterized in that, It includes a flexible array pressure sensor module, a microcontroller information processing terminal, a Bluetooth communication module, and a host computer; The flexible array pressure sensor module adopts an ultra-thin flexible sensor array based on the piezoresistive effect, and the flexible array pressure sensor module acquires pressure through a secondary measurement method; The microcontroller-based information processing terminal integrates a voltage conversion module and a signal acquisition and amplification circuit, with a sampling frequency of 500Hz. The pressure analog signal output from the flexible array pressure sensor module is input to the signal acquisition and amplification circuit, amplified, and then output to the microcontroller's ADC acquisition channel. The microcontroller performs analog-to-digital conversion and data preprocessing on the analog signal before outputting it to the Bluetooth communication module via the UART interface. The voltage conversion module converts the lithium battery input voltage into a stable voltage, powering the microcontroller, the signal acquisition and amplification circuit, and the Bluetooth communication module. The host computer communicates with the microcontroller information processing terminal via a Bluetooth communication module. The host computer is configured with a data display, storage, and interactive interface to monitor pressure data in real time.

2. The intelligent ski instructor plantar pressure acquisition system according to claim 1, characterized in that, The flexible array pressure sensor module is a 6-row × 4-column array.

3. The intelligent ski instructor plantar pressure acquisition system according to claim 1, characterized in that, The ultra-thin flexible sensor based on the piezoresistive effect is the FlexiForce plantar pressure sensor.

4. The intelligent ski instructor plantar pressure acquisition system according to claim 1, characterized in that, The aforementioned secondary measurement method includes two scanning processes: A resistor array with n rows and m columns is set up, and the measurement point is placed in front of the row line selection switch; During the first scan, the common terminal of the line selection switch is connected to V. cc The column line selection switch is grounded at its common terminal, and the row line selection switch is switched n times. During the second scan, the common terminal of the row line selector switch is grounded, and the common terminal of the column line selector switch is connected to V. cc For each column line gating switch that toggles once, and the row line gating switch that toggles n times, the total number of transitions is: n×m; Where: n is the number of rows, and m is the number of columns; The resistance value R to be measured is calculated by two voltage scans. 22 for: R 22 =Vout1 / Vout2×R0; Where: Vout1 is the output voltage of the first scan; Vout2 is the output voltage of the second scan; R0 is the reference resistor.

5. The data processing method based on the intelligent ski instructor plantar pressure acquisition system according to claim 1, characterized in that, Includes the following steps: Step 1: Data Acquisition. Foot pressure data for five skiing postures—snowball downhill, snowball left turn, snowball right turn, snowball braking, and parallel turn—are collected using a flexible array pressure sensor module. At least 2.2 seconds of valid data are captured for each posture. Step 2: Data Preprocessing: Missing values ​​are handled and the collected data is normalized. The normalization formula is: ； Where: X' I The data is after normalization; X i This is the original data; μ is the mean of the data; σ is the standard deviation of the data; Step 3: Data dimensionality reduction. Weighted principal component analysis is used. First, representative data for each feature are determined by averaging multiple measurements. Then, the weights are calculated based on the total ground pressure. The weight calculation formula is as follows: ； in: Representative data for features; F represents the total pressure on the ground; The data is reduced to 5 dimensions by using an error threshold. Step 4: Attitude recognition. Based on the adaptive particle swarm optimization algorithm with dynamic Lévy flight perturbation, the BP neural network is improved to classify the dimensionality-reduced data and output the skiing attitude recognition results.

6. The data processing method according to claim 5, characterized in that, The data processing method further includes: Step 5: Result verification. The results are compared and verified with four optimization algorithms: LM, Bayesian regularization, SCG, and standard PSO, using four metrics: accuracy, precision, recall, and F1 score.

7. The data processing method according to claim 5, characterized in that, The 2.2 seconds of valid data is discontinuous.

8. The data processing method according to claim 5, characterized in that, Step three of the weighted principal component analysis method specifically includes the following steps: Step 1: Construct an m×n original data matrix X, where m is the number of samples and n is the number of features. The data corresponding to each variable can be denoted as X1, X2, X3, X4, ... X n The original data matrix is ​​denoted as X, specifically as follows: ； Step 2: Calculate representative data for each feature. Specifically: ； Step 3: Measure the total ground pressure F of the test subject in the corresponding posture, and calculate the weights P of each feature. L Specifically: ； Step 4: Construct a weighted data matrix Specifically: ； Step 5: [Regarding...] Perform mean normalization to 0 and calculate the covariance matrix; Step 6: Solve for the eigenvalues ​​and eigenvectors of the covariance matrix, select the eigenvectors corresponding to the first k largest eigenvalues ​​to construct the projection matrix, and reduce the data dimension to k dimensions, where k=5.

9. The data processing method according to claim 5, characterized in that, The implementation steps of the adaptive particle swarm optimization algorithm based on dynamic Lévy flight perturbation described in step four are as follows: Step 1: Initialization Phase Parameter Settings: Determine the particle swarm size N and the maximum number of iterations T. max Search space upper and lower bounds [Lb,μb], dimension dim, initial inertia weight ω, learning factors c1, c2; Chaotic mapping initialization: Initial particle positions are generated using chaotic sequences such as Logistic mapping; Velocity initialization: Randomly generate the initial velocity of the particles; Step 2: Fitness Calculation and Optimal Update Calculate fitness: Evaluate the fitness value of each particle based on the objective function; Update individual best (pBest): If the current fitness is better than the historical best, then update the individual's best position; Update Global Best (gBest): Records the best position among all particles at present; Step 3: Dynamic Lévy flight perturbation Lévy step size generation: Generate random step size based on Lévy distribution L ( beta ), L ( beta The value is usually taken as 1.5, and the formula is: Dynamic adjustment mechanism: The step size is dynamically scaled based on the number of iterations. Where: α is the initial step size factor, which decreases with each iteration to balance exploration and development; Position Update: Apply Levy perturbation to some poorly fit individual particles: In the formula: x rand The particle position is randomly selected; Step 4: Adaptive parameter adjustment. The inertia weight ω is adjusted either linearly decreasing or based on population diversity. The learning factors c1 and c2 are dynamically adjusted according to the convergence state. For example, increasing c1 in the early stage enhances exploration, and increasing c2 in the later stage accelerates convergence. Step 5: Speed ​​and Position Update Standard (PSO) Update: Boundary handling: If a particle goes out of bounds, a reflection or random reset strategy is used; Step 6: Chaotic Perturbation Stagnation Detection: If gBest fails to update for k consecutive times, a chaotic perturbation is triggered. Perturbation Method: A chaotic sequence is generated using a Logistic mapping, and the positions of some particles are reinitialized. Step 7: Termination condition reaches the maximum number of iterations T max Or the fitness value changes less than the threshold ε.

10. The data processing method according to claim 5, characterized in that, The BP neural network described in step four has a "5-12-5" structure, with 5 neurons in the input layer, 12 neurons in the hidden layer, and 5 neurons in the output layer. The optimization parameters include: 30 particles, 50 maximum iterations, inertia weight range of 0.4-0.9, initial learning factor values ​​of c1=1.5 and c2=2.5, diversity trigger threshold of 0.05, Sigmoid activation function, and Softmax function for the output layer.

Images