A golf cart dynamic path planning method fusing golf course road conditions

Abstract

This invention provides a dynamic path planning method for golf carts that integrates golf course road conditions, relating to the field of road vehicle traffic control technology. By constructing a comprehensive system encompassing eco-mechanical coupling deviation perception, dynamic gain tensor control, and Bayesian parameter evolution mechanisms, this invention solves the technical challenges of perception blind spots and control rigidity in unstructured road conditions. Its core value lies in not only improving transient driving stability and anti-skid capability on concealed soft surfaces, but also achieving a fundamental shift in path planning from a simple "efficiency-first" approach to a "dual optimization of ecological protection and operational safety" through long-term adaptive calibration of the environmental evolution model, thereby enhancing the robustness and environmental adaptability of intelligent golf carts in complex and ever-changing environments.

Description

Technical Field

[0001] This invention relates to the field of traffic control technology for road vehicles, specifically to a dynamic path planning method for golf carts that integrates golf course road conditions. Background Technology

[0002] With the increasing application of intelligent transportation and autonomous driving technologies in specific closed areas (including golf courses), machine learning-based unstructured road environment perception and path planning have become crucial for improving vehicle operating efficiency and safety. Achieving accurate identification of road surface physical properties and adaptive control of vehicle dynamics under complex and varied terrain conditions is a cutting-edge technology that urgently needs breakthroughs in the field of intelligent transportation control.

[0003] Despite advancements in existing technologies for route planning, such as the method proposed in patent application CN120932488A based on dynamic road condition prediction, the following challenges remain when dealing with unstructured and complex road conditions: Existing technologies rely heavily on visual or meteorological data for macroscopic predictions, making it difficult to penetrate the surface and perceive the physical and mechanical resistance at the microscopic level (such as soil compaction). This leads to misjudgments in visually deceptive scenarios (such as grass-covered mud pits), indicating a limitation in the dimensions of perception. The lack of a closed-loop feedback mechanism for the dynamic evolution of the environment over time (such as lawn restoration) leads to the long-term solidification of the planned path, which can easily cause local ecological degradation and result in the static rigidity of the model. The lack of a nonlinear compensation mechanism based on multidimensional states (such as energy and trends) makes it difficult to balance transient safety and long-term energy efficiency under sudden road conditions, resulting in a single control strategy. Summary of the Invention

[0004] The purpose of this invention is to provide a dynamic path planning method for golf carts that integrates golf course conditions, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A method for dynamic path planning of golf carts that incorporates golf course road conditions, comprising the following steps: S1: Acquire the first data characterizing the real-time impedance characteristics; the first data is a quantitative value characterizing the physical and mechanical impedance characteristics of the current driving interface, calculated by a frequency domain feature transformation algorithm based on the electrical fluctuation attributes of the vehicle drive actuator and the vertical oscillation attributes of the inertial sensing end; and simultaneously acquire the kinematic parameters characterizing the current motion state of the vehicle. S2: Obtain the second data representing the expected carrying capacity index; the second data is a scalarized index representing the physical carrying capacity potential of the road surface at the current geographic coordinate point under theoretical conditions, generated based on a preset environmental growth and evolution model and historical traffic load records at the current location; S3: Using the ecomechanical coupling deviation calculation model, execute the heterogeneous data same-domain mapping procedure; specifically, first map the first data and the second data to the preset dimensionless reference traffic state space, then calculate the ecomechanical coupling deviation measure of the two in the dimensionless reference traffic state space, and output the ecomechanical coupling deviation measure as an ecomechanical coupling deviation signal used to characterize the degree of deviation of the true physical value of the road surface from the theoretical prediction value. S4: Based on the aforementioned ecomechanical coupling deviation signal, generate a dynamic control instruction set that includes dynamic response correction instructions and model parameter evolution instructions; S5: Execute the dynamic control instruction set to adjust the dynamic traffic response strategy of the vehicle passing through the current road segment in the first time domain window, and correct the environmental evolution weight parameters of the corresponding spatial nodes in the environmental growth and evolution model in the second time domain window, thereby realizing the closed-loop update of the subsequent path planning cost function.

[0006] Compared with existing technologies, the beneficial effects of this invention are: by using a heterogeneous data co-domain mapping program, the "real-time impedance characteristics" extracted based on vehicle electrical fluctuation attributes and the "expected load-bearing index" derived from cellular automata are mapped to the same dimensionless state space. By calculating the coupling deviation metric between the two, covariant environmental factors can be accurately isolated, and the non-statistical deviation of the perceived physical true value of the road surface from the theoretical prediction value can be quantified, thereby effectively identifying hidden risks.

[0007] A "three-dimensional dynamic gain tensor space" was constructed as the core engine for control decision-making. This mechanism constructs an orthogonal decision dimension by integrating the energy availability index, deviation trend factor, and deviation amplitude. Utilizing a trilinear interpolation algorithm, nonlinear composite attenuation coefficients can be dynamically generated based on the vehicle's current energy state and road condition deterioration rate, achieving a leap from passive response to predictive defense.

[0008] A "Bayesian confidence-weighted update mechanism" was established. Using measured data from a single passage and combining it with the confidence level of event observations, the state parameters in the environmental growth and evolution model were rationally adjusted. This closed-loop calibration mechanism ensures that long-term path planning can adapt to the dynamic evolution of the environment. Attached Figure Description

[0009] Figure 1 This is a schematic diagram illustrating the process principle of the dynamic path planning method for golf carts that integrates golf course road conditions, as described in this invention. Figure 2 This is a schematic diagram illustrating the technical principles of steps S1 to S3 of the present invention. Figure 3 A schematic diagram of the simulation experiment conducted to verify the technical effect of the dynamic instruction generation mechanism of the present invention; Figure 4 This is a schematic diagram illustrating the technical principles of steps S4 to S5 of the present invention. Figure 5 This is a schematic diagram of the simulation experiment conducted to verify the technical effectiveness of the dynamic control instruction set generation logic in this invention. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0011] It is understood that the terms “first,” “second,” etc., used in this application may be used herein to describe various elements, but unless otherwise stated, these elements are not limited by these terms. These terms are used only to distinguish one element from another.

[0012] Example 1: Please see Figures 1 to 5 The present invention provides a technical solution: A dynamic path planning method for golf carts that incorporates golf course road conditions includes the following steps: S1: Acquire the first data characterizing the real-time impedance characteristics; the first data is a quantitative value characterizing the physical and mechanical impedance characteristics of the current driving interface, calculated by a frequency domain feature transformation algorithm based on the electrical fluctuation attributes of the vehicle drive actuator and the vertical oscillation attributes of the inertial sensing end; and simultaneously acquire the kinematic parameters characterizing the current motion state of the vehicle. S2: Obtain the second data representing the expected carrying capacity index; the second data is a scalarized index representing the physical carrying capacity potential of the road surface at the current geographic coordinate point under theoretical conditions, generated based on a preset environmental growth and evolution model and historical traffic load records at the current location; S3: Using the ecomechanical coupling deviation calculation model, execute the heterogeneous data same-domain mapping procedure; specifically, first map the first data and the second data to the preset dimensionless reference traffic state space, then calculate the ecomechanical coupling deviation measure of the two in the dimensionless reference traffic state space, and output the ecomechanical coupling deviation measure as an ecomechanical coupling deviation signal used to characterize the degree of deviation of the true physical value of the road surface from the theoretical prediction value. S4: Based on the aforementioned ecomechanical coupling deviation signal, generate a dynamic control instruction set that includes dynamic response correction instructions and model parameter evolution instructions; S5: Execute the dynamic control instruction set to adjust the dynamic traffic response strategy of the vehicle passing through the current road segment in the first time domain window, and correct the environmental evolution weight parameters of the corresponding spatial nodes in the environmental growth and evolution model in the second time domain window, thereby realizing the closed-loop update of the subsequent path planning cost function.

[0013] 1) In step S1, the first data characterizing the real-time impedance features is obtained, specifically including: Acquire the high-frequency current ripple sequence of the drive unit within a preset time window; Perform a Fast Fourier Transform (FFT) on the high-frequency current ripple sequence to extract the energy spectral density of a specific frequency band; The energy spectral density is input into a pre-trained Long Short-Time Memory (LSTM) network, which outputs the real-time impedance features, wherein the real-time impedance features characterize the shear drag coefficient of the road medium on the tire.

[0014] In this embodiment, the three-phase current data of the motor controller is read via the CAN bus at a sampling rate of 1kHz. During data preprocessing, the DC component is filtered out, and a 50ms sliding window of data is extracted. A 512-point FFT transform is performed using an embedded DSP unit, focusing on the 200Hz-500Hz frequency band (this band mainly corresponds to the mechanical vibration frequency caused by the interaction between the tire and the non-rigid road surface). The power spectral density (PSD) vector of this frequency band is used as input features and fed into a lightweight LSTM model.

[0015] To clarify the "shear drag coefficient characterized by real-time impedance features" In the specific generation process of "X", the embedded DSP unit in this embodiment performs the following defined digital inference calculation steps: Feature vector construction and normalization: The system acquires energy spectral density data within the 200Hz-500Hz frequency band and constructs it into a one-dimensional feature vector X. The pre-stored global mean is then called. and global standard deviation Perform standardization on the one-dimensional eigenvector X, i.e. Generate standard input vector Call the pre-trained first weight matrix. (Dimensions are 128×512) and the first bias vector .calculate and Multiply the matrices and add the result. The ReLU activation function is performed on each element of the result vector. In this embodiment, we take... This compresses and maps the 512-dimensional spectral features into a 128-dimensional hidden layer feature vector. The hidden layer feature vectors are computed in the LSTM layer. The input is fed into a bidirectional LSTM computation unit. In this unit, the current input and the cell state vector from the previous time step are used. and hidden state vector Perform the following gating logic operations: The output value of the forget gate is calculated to determine how much of the cell state from the previous time step is retained; the input gate and candidate cell states are calculated to determine how much new information from the current input is written into the cell state; based on the above forgetting and input logic, the cell state vector at the current time step is calculated and updated. Based on the updated cell state and output gate, the current time-series output vector containing temporal context information is calculated. Its dimension is 128. The pre-trained regression weight matrix is ​​called. (Dimension 1×128) and regression bias scalar .calculate and The dot product, and the result with The summation yields a scalar result for the shear drag coefficient, which also represents the real-time impedance characteristic value. This value directly characterizes the shear resistance coefficient of the current road surface medium, with dimensionless or normalized mechanical units. In this embodiment, a calculation result of 0.2 characterizes "slippery mud with low shear resistance," and a result of 0.8 characterizes "dry grassland with high shear resistance." This embodiment has been trained offline using labeled data such as "mud," "sand," "dry grass," and "wet grass," enabling real-time inference of the real-time impedance characteristics represented by the shear resistance coefficient of the current road surface. This embodiment, by analyzing the microscopic ripples of the motor current, can accurately perceive the deep physical properties of the road surface (including soil looseness). This improves the robustness of perception under complex lighting or concealed road conditions and ensures the objectivity and authenticity of the data source.

[0016] 2) In step S2, the second data characterizing the expected carrying capacity index is obtained, specifically including: Query the road network cell state matrix stored in the local cache. Each cell in the road network cell state matrix corresponds to a discrete grid in the geospatial space. Read the cumulative compaction status value of the cell corresponding to the current vehicle location and the timestamp of the most recent update; Based on cellular automata rules and combined with current meteorological parameters, the soil recovery state of the cells at the current moment is deduced, and the expected carrying capacity index is calculated.

[0017] The environmental growth and evolution model in this embodiment is essentially a state transition calculation logic embedded in each cell. In this embodiment, the environmental growth and evolution model is concretized as a linear iterative calculation step that includes natural recovery and environmental depletion terms. The expected carrying capacity index at the current moment is calculated by performing the following steps: The system reads the previous state of the cell. The recorded cumulative compaction state value (denoted as ) (Values ​​range from 0 to 1). Calculate the natural recovery increment system at the current time. The time difference with the previous moment. Using the natural rate of reversion coefficient. Calculate the natural recovery of soil strength within this time difference. This step simulates the physical process of lawn soil gradually hardening and compacting over time under static conditions. Obtain the cumulative rainfall data for the current time period. Utilize the rainfall softening coefficient. Calculate the soil strength reduction caused by water infiltration. This step simulates the physical process of shear strength decrease due to increased soil moisture content. The cumulative compaction state value of the previous moment is added to the natural recovery increment and the environmental loss reduction is subtracted. The result is then truncated (limited to between 0 and 1) to obtain the expected bearing capacity index at the current moment.

[0018] This embodiment maintains a global digital twin in the cloud, employing a cellular automata architecture. Each 1m×1m cell contains state variables such as "soil health" and "moisture saturation." During vehicle operation, the onboard unit pre-downloads the cellular state matrix within a 500-meter radius. In step S2, the cell data corresponding to the current GPS coordinates is read, and the state transition equation is applied... Calculate the current theoretical bearing capacity of this location.

[0019] Indicates the current time The expected bearing index calculated by this cell is used to characterize the current cumulative compaction state value, and its value is limited to [0,1]. Characterization : Indicates the previous moment The system records the historical cumulative compaction state value of this cell. and These represent the timestamp of the current calculation time and the timestamp of the previous state record time, respectively. The difference between the two is the time interval used to calculate the natural recovery. Indicates from the previous moment up to the current moment Cumulative rainfall over the period of time (in mm).

[0020] This method treats road surfaces as organisms with dynamic "damage-recovery" characteristics. By quantitatively predicting the carrying capacity of road surfaces, high-intensity traffic can be avoided during the vulnerable period of lawns (initially after rain), thus achieving "ecological and efficiency dual optimization" in path planning.

[0021] "Environmental evolution model coefficients" in the calculation of the expected carrying capacity index This corresponds to the "natural recovery rate" mentioned in step S2. "and Rainfall softening coefficient" The following standardized experimental calibration procedure was used to determine the baseline value: a soil sample with the same soil quality as the target field was selected, and a 1m×1m experimental quadrat was constructed.

[0022] Regarding the natural recovery rate The calibration procedure was as follows: The sample plots were compacted using a standard roller until the soil density reached saturation, defined as a state value of 1.0. The sample plots were then left to stand in a constant temperature (25℃) environment with no rainfall. Soil hardness was measured every 12 hours using a penetration tester for 7 consecutive days. The measurement data were plotted as a time-hardness recovery curve. An exponential decay model was then used. The attenuation constant obtained by fitting the curve using the least squares method is the natural rate of recovery. .

[0023] Where H(t) represents the soil hardness value measured when the settling time is t. This represents the initial soil hardness value, which is the basic soil hardness value measured at the moment the sample plot is first compacted and begins to settle (t=0). This represents the soil saturation hardness limit, which is the theoretical maximum stable hardness value that this type of soil can achieve over time through natural compaction and hardening under the current experimental conditions. t represents the duration calculated from the start of the settling period, and is related to the natural recovery rate. The dimensions are consistent. e is the base of the natural logarithm.

[0024] For rainfall softening coefficient The calibration procedure is as follows: The quadrats are restored to their natural state. Using an artificial rainfall simulator, the quadrats are continuously sprayed with a fixed rainfall intensity (10 mm / h). Soil shear strength is measured every 10 minutes. A curve relating cumulative rainfall to shear strength is established. A linear regression model is then used. The absolute value of the slope obtained by fitting the linearly decreasing segment is the rainfall softening coefficient. Where S(r) represents the soil shear strength value measured when the cumulative rainfall reaches r. This represents the initial soil shear strength, i.e., before artificial rain spraying (…). (This refers to the baseline shear strength of the sample plot soil under natural conditions.) Indicates the value of the soil shear strength since the previous moment. up to the current moment The cumulative rainfall over a given period of time.

[0025] 3) In step S3, the generation of the ecomechanical coupling deviation signal using the ecomechanical coupling deviation calculation model specifically includes: By using a preset normalization mapping function, the real-time impedance characteristics and the expected load-bearing index are converted into dimensionless standard values ​​(i.e., the shear resistance coefficient of the road medium to the tire) with a value range of [0,1]. Calculate the ecomechanical coupling deviation measure (EMCDT) between two standard values; The ecomechanical coupling deviation metric includes an amplitude component and a trend component, wherein the amplitude component represents the degree of deviation between the measured value and the predicted value at the current moment, and the trend component represents the first derivative of the degree of deviation with time; The ecomechanical coupling deviation signal is output only when the amplitude component exceeds a preset confidence threshold.

[0026] Furthermore, in the step of converting the real-time impedance characteristics and the expected load-bearing index into dimensionless standard values ​​ranging from [0,1] using a preset normalization mapping function, a dynamic nonlinear normalization strategy is specifically adopted, which includes: Obtain the kinetic energy mode adjustment factor that characterizes the current motion intensity of the vehicle; the kinetic energy mode adjustment factor is a representation of the kinematic parameters; Construct a sigmoid activation function with a variable slope, the steepness of which is dynamically modulated by the kinetic mode adjustment factor; The real-time impedance characteristics and the expected load-bearing index are respectively input into the S-type activation function, and the output is the dimensionless standard value. The kinetic mode adjustment factor is positively correlated with the motion intensity of the vehicle, so that the linear response range of the S-shaped activation function is narrowed when the motion intensity is higher, in order to suppress the influence of high-frequency noise on the standard value.

[0027] Furthermore, in the step of calculating the ecomechanical coupling bias measure (EMCDT) between the two standard values, a covariance-weighted bias measure mechanism was specifically adopted, which includes: Maintain a sliding observation window that stores a historical sequence of dimensionless standard values; Based on the real-time impedance characteristics and the historical dimensionless standard value sequence of the expected bearing index recorded within the sliding observation window, a covariance matrix is ​​calculated to characterize the correlation of data fluctuations. Perform matrix inversion on the covariance matrix to obtain the information precision weight matrix; Calculate the original difference vector between the standard value of the real-time impedance characteristic and the standard value of the expected load index; Calculate the quadratic product of the original difference vector and the information precision weight matrix, and determine the arithmetic square root of the quadratic product as the amplitude component of the ecomechanical coupling deviation metric.

[0028] The following are specific implementation instructions for the above content: Before executing the heterogeneous data localization mapping procedure, the following spatiotemporal alignment logic is performed to eliminate the dimensional mismatch between the first data (temporal stream) and the second data (spatial grid); specifically, the high-frequency GPS trajectory sequence of vehicles within a 50ms time window corresponding to the generated first data (real-time impedance characteristics) is obtained. All road network cells covered by the GPS trajectory sequence are queried. For each covered cell, the proportion of the vehicle trajectory's dwell time within that cell is calculated. Using the dwell time proportion as a weighting coefficient, a weighted average of the "expected carrying capacity index" of all covered cells is calculated to obtain a "synthetic expected carrying capacity index" that is strictly aligned spatiotemporally with the first data. The "synthetic expected carrying capacity index" is used as the second data input for subsequent steps S3. This step ensures that the two physical quantities involved in the deviation calculation always correspond to the same physical space segment.

[0029] In existing golf course road condition sensing technologies, although measured "real-time impedance characteristics" and theoretical "expected load-bearing index" are obtained through steps S1 and S2 respectively, traditional methods employ static linear normalization and simple Euclidean distance difference when performing heterogeneous data fusion in step S3. In actual golf cart driving scenarios, a fundamental technical bottleneck exists: the problem of "confidence drift" in the data. For example, when a vehicle travels at high speed on undulating surfaces, the variance of the sensor data increases significantly due to intensified mechanical vibration (reducing the signal-to-noise ratio). If a fixed linear normalization is still used at this time, noise will be mistakenly amplified into a valid signal. Simultaneously, there is often a physical coupling correlation between the road surface's "impedance" and "load-bearing capacity." Simple linear subtraction cannot isolate this natural correlation, causing the system to easily misjudge normal "soft road surface deformation" as "road collapse risk."

[0030] To address the aforementioned issues, this preferred embodiment incorporates an "adaptive deviation calculation mechanism based on statistical manifold mapping." This mechanism dynamically adjusts the normalization sensitivity by introducing the vehicle's kinetic energy state and utilizes the covariance matrix to eliminate redundant correlations between variables, thereby achieving precise purification of the "degree of deviation from the physical truth" at the mathematical level.

[0031] It should be noted that in this embodiment, the LSTM network is configured as a lightweight three-layer cascaded structure to adapt to the computing power limitations of the in-vehicle embedded computing unit.

[0032] The first layer is the input projection layer: it contains 128 fully connected neurons with ReLU activation function. Its function is to reduce the dimensionality of the input 512-dimensional PSD vector (corresponding to the FFT output) and map it to a 128-dimensional feature latent space.

[0033] The second layer is the bidirectional LSTM core layer, containing 64 hidden units in each direction. The design consideration for the bidirectional structure is that the change in road impedance has temporal continuity, which is not only affected by the current moment, but also by the accumulation of past states (memory) and the potential constraints of future trends (context). The bidirectional structure can capture more complete temporal characteristics.

[0034] The third layer is the regression output layer, containing one linear neuron. Its function is to converge the high-dimensional state vector output by the LSTM layer into a scalar value, namely the real-time impedance feature.

[0035] The dataset used to train this model must meet the following key characteristics: the data originates from a dedicated test bench equipped with a high-precision force sensor (a six-dimensional force sensor with a measurement accuracy of no less than 0.1N). Controlled slip tests are conducted on four preset typical ground surfaces (dry grass, wet grass, sand pit, and mud) at speeds ranging from 0-20 km / h. The ratio of the "tangential reaction force" to the "normal load" measured in real-time by the force sensor is used as the "true value".

[0036] All input current ripple energy spectrum data underwent logarithmic transformation to compress the dynamic range and were mapped to a distribution interval with a mean of 0 and a variance of 1 using the Z-score normalization method. For the Z-score normalization step, the following parameter freezing and application logic was strictly implemented during the model deployment phase: Read two key statistical constants from the pre-configured model parameter file: the global mean vector (corresponding symbol...). ) and global standard deviation vector (corresponding symbol) These two vectors were obtained during the offline training phase by statistically calculating all samples in the entire training dataset. Their dimension is exactly the same as the input energy spectral density vector, 512 dimensions. For the real-time input vector acquired at the current moment and after logarithmic transformation... Calculate its relationship with the global mean vector. The difference vector; divide each element of this difference vector by the global standard deviation vector. The corresponding elements are then used to obtain the standardized feature vector that is finally input to the first layer of the network. This step ensures that the online inference data and the offline training data are in the same statistical distribution space.

[0037] The mean squared error (MSE) loss function was used during training, with an additional L2 regularization term (coefficient set to 0.001) to prevent overfitting. The Adam algorithm was used as the optimizer, with an initial learning rate of 0.001 and a dynamic decay strategy (decreasing to 0.9 times the original rate every 50 epochs). The training termination condition was set when the validation set loss value no longer decreased for 10 consecutive epochs.

[0038] In this embodiment, the following core technical parameters are defined to implement the above logic.

[0039] Kinetic mode adjustment factor, denoted as : A dimensionless coefficient characterizing the degree to which the vehicle's current motion state affects the stability of data acquisition. It reflects whether "the data is worth accepting with high sensitivity under the current operating conditions".

[0040] The system reads the "real-time displacement rate information" and "steering angular velocity information" output by the vehicle chassis control unit. The calculation logic is as follows: obtain the magnitude of the displacement rate; obtain the magnitude of the steering angular velocity; perform a weighted summation of the two; input the summation result into a preset logarithmic mapping function, and output the kinetic energy mode adjustment factor. The larger this value, the more intense the motion and the higher the risk of data noise. The specific calculation logic is defined as follows: Input the vehicle displacement rate v (unit: m / s); Input the vehicle steering angular velocity (Unit: rad / s). Linear and angular motion are unified in the energy dimension, introducing a reference characteristic length L, where L is the vehicle wheelbase (2.0 meters in this embodiment). The "equivalent angular velocity" is then calculated. Set linear velocity weights. The angular velocity weight is 0.6. The weight is 0.4, set based on the contribution of vehicle lateral / longitudinal dynamics to tire slip ratio. The composite kinetic energy index is calculated. To achieve passivation (avoiding hypersensitivity) in the low-speed range and activation (rapid response) in the high-speed range, a natural logarithm mapping model is adopted: Regarding the above sensitivity gain coefficient To determine this, the following standardized statistical process is executed in this embodiment: A standard golf course grass area was selected, and the vehicle was controlled to maintain a straight-line driving speed within a preset range (specifically 15 km / h to 18 km / h), with minor steering corrections (steering angle less than 5 degrees). Data collection was conducted continuously for at least 300 seconds. The arithmetic mean of all synthetic kinetic energy indices E within the above collection period was calculated to obtain the baseline kinetic energy mean. Predetermine a natural constant e, calculate the difference between this natural constant and 1; divide this difference by the mean baseline kinetic energy. The resulting quotient is then fixed as the sensitivity gain coefficient. This calculation logic aims to normalize and anchor the logarithmic mapping output value when the vehicle is in normal cruising mode to a value around 1.0. In this embodiment, the value range of this factor is limited to [0.1, 5.0].

[0041] Information precision weight matrix, denoted as It is a two-dimensional matrix that numerically represents the volatility of "real-time impedance characteristics" and "expected load index" within the current time window, as well as the degree of synchronous volatility between the two. Essentially, it is the inverse of the covariance matrix. It is determined by maintaining a historical data queue of length N. In each calculation cycle, the statistical covariance of the data in the queue is calculated to generate the covariance matrix; then, the covariance matrix is ​​inverted using a linear algebraic decomposition algorithm (Cholesky decomposition) to obtain the information precision weight matrix.

[0042] In this embodiment, the value of N is determined based on the matching relationship between Shannon's sampling theorem and road surface feature frequencies. Specifically, it is determined by the minimum attention frequency. In this scenario, the frequency of ground undulations that can cause significant changes in vehicle attitude is no less than 0.5Hz. Data sampling frequency is then determined. In this embodiment, the execution cycle of step S3 is 100ms, that is... To ensure a confidence level >95% for the covariance matrix estimation and to cover at least one complete lowest-frequency fluctuation period of 2 seconds, the window length must contain at least [missing information - likely a number]. Number of points. Set N=30. This value strikes the best balance between "capturing 2-3 complete fluctuation cycles to obtain stable statistical characteristics" and "avoiding excessively long windows that lead to a sluggish response to abrupt changes".

[0043] Statistical manifold deviation modulus, denoted as : This is the amplitude component of the preferred "ecomechanical coupling deviation metric" in this embodiment. It represents the "anomaly distance" of the current measured state relative to the theoretical state in the "probability density space." It is obtained based on logical operations of the Mahalanobis distance.

[0044] In this preferred embodiment, to suppress large-amplitude noise caused by high-speed vibration, an "adaptive S-shaped mapping" is used. Compared to a linear mapping, the S-shaped function is approximately linear in the central region, but exhibits "soft saturation" characteristics at both ends. By using the "kinetic energy mode adjustment factor" as the slope control parameter of the S-shaped function, it is possible to achieve: at low speeds ( At low speeds, the slope is steep and sensitive to minute impedance changes; at high speeds... When the slope is large, the slope is gentle, automatically suppressing violent data fluctuations caused by vibration.

[0045] Read the real-time impedance characteristic value output in step S1 And the expected load index value output in step S2 Subtract the historical mean from each of the two values ​​mentioned above to achieve zero-mean normalization. Obtain the current kinetic mode adjustment factor. For real-time impedance characteristic values Perform mapping operation: convert the real-time impedance characteristic values Multiply Calculate its exponential function value, add 1, and take the reciprocal. This step maps the original physical quantity to a dimensionless standard value within the interval (0,1). Similarly, the calculation yields... .

[0046] In this embodiment, to decouple the natural coupling between variables, a "statistical manifold projection method" is employed. "Impedance" and "carrying capacity" are not physically independent (both decrease simultaneously during rain). If only simple differences are calculated, this "synchronous decrease" would be considered a significant deviation, triggering false alarms. By introducing the inverse of the covariance matrix as weights, weights with high volatility dimensions can be automatically "punished," and components that change synchronously can be "offset," retaining only anomalous deviations that violate historical correlation patterns. The specific logic is as follows: calculate and The algebraic difference forms a column vector. Read the precision weight matrix from memory for the current moment. Perform matrix multiplication. Multiply the column vectors. The transpose vector and the information precision weight matrix Multiply them to obtain the intermediate row vector; then multiply this intermediate row vector with the original column vector. Multiplying these results in a scalar value. Performing the arithmetic square root operation on this scalar value yields the statistical manifold deviation modulus. .

[0047] Statistical manifold deviation modulus The magnitude component of the ecomechanical coupling deviation metric is assigned, and its time derivative (trend component) is combined to output the final signal. The following flowchart describes the entire process from data input to final decision: The current ripple is collected, processed by FFT and LSTM, and the real-time impedance characteristics are output. The road network cell matrix is ​​queried, and the expected load-bearing index is output through cellular automata derivation.

[0048] Initiate the dynamic normalization subroutine. Read the vehicle chassis bus speed and steering data in real time and calculate the kinetic mode adjustment factor. Use this kinetic mode adjustment factor to adjust the curvature of the S-shaped mapping function, mapping the "real-time impedance characteristics" and "expected load index" to dimensionless standard values ​​in the [0,1] interval. Read the historical standard values ​​of the most recent N time periods from the local cache, dynamically update the covariance matrix, and calculate the information precision weight matrix represented by its inverse matrix. Using this information precision weight matrix, perform covariance weighted deviation measurement logic on the two current dimensionless standard values, and output the statistical manifold deviation modulus.

[0049] The calculated statistical manifold deviation modulus is compared with a preset confidence threshold. Only when the modulus value is greater than the threshold is it determined that there is a substantial, non-statistical error deviation between the current road surface physical properties and the prediction of the digital twin model, including the situation of sudden hidden mud pits. In this case, a high-priority ecomechanical coupling deviation signal is generated and output, triggering subsequent path replanning actions.

[0050] Further elaboration on the implementation of steps S1 to S3 above: The core output of this technical solution is the amplitude component of the ecomechanical coupling deviation metric (EMCDT). This index is a scalar obtained after nonlinear normalization and statistical manifold projection, and its theoretical range is... When the output value At this point, the "ecological-mechanical fully synchronized state" indicates a precise consistency between the "physical reality perceived by the vehicle" and the "theoretical expectations derived from the digital twin." Specifically, electrical fluctuations at the vehicle's drive end fall entirely within the carrying potential range predicted by the environmental evolution model, and the minute fluctuations of both conform to historical statistical patterns, i.e., the covariance structure. At this time, the road surface is in an absolutely controllable state, with no hidden disasters or sudden anomalies. It is determined that no path intervention is necessary, and maintaining the current control strategy can achieve optimal energy efficiency.

[0051] When the output value When the confidence threshold is exceeded: "Non-statistical physical deviation state" indicates a structural break between the physical truth and theoretical prediction that cannot be explained by statistical error. Despite... Suppress vehicle motion noise and Even after filtering out environmental variables, the measured impedance characteristics still deviated from expectations. For example, in areas that should theoretically be hard grass, the vehicle detected hidden mud pits characterized by low impedance, similar to "fluid shearing." In such cases, an "out-of-model" sudden road condition was identified, and the original path planning, based on flawed prior information, had to be immediately re-planned to mitigate the physical risks.

[0052] To determine the final output Two core control parameters: kinetic energy mode adjustment factor With information accuracy weight matrix Execution logic tracing and effect demonstration.

[0053] For kinetic mode adjustment factor The impact on the normalized standard value is analyzed as follows: Kinetic mode adjustment factor It is negatively correlated with the bandwidth of the linear response region of the sigmoid activation function, i.e. As the linear region increases, the saturation region narrows and expands, thus exhibiting a nonlinear negative correlation with the system's sensitivity to small disturbances.

[0054] In a vehicle's high-dynamic state, specifically at high speed or during sharp steering, i.e. At high speeds, the mechanical vibration noise and high-frequency noise from road surface texture mixed into the raw sensor data (such as current ripple and IMU oscillation) increase exponentially. At this point, the signal-to-noise ratio of the data decreases significantly.

[0055] pass The steepness of the dynamically modulated S-shaped function. When As the amplitude increases, the S-curve function becomes steeper, and its linear transition interval is forcibly compressed and narrowed. This narrowing of the linear interval forces most of the input data towards the saturation region represented by the two ends of the S-curve function. Within the saturation region, the incremental output of any high-frequency input noise fluctuation approaches zero. This enables a gating mechanism based on kinetic energy state. During vigorous motion, it actively ignores subtle tactile inputs and only responds to large-amplitude impact signals, thus avoiding frequent false alarms due to oversensitivity. This ensures that when the vehicle switches from "stationary fine perception" to "high-speed cruising," it can automatically establish a "dynamic noise firewall," solving the fundamental deficiency of traditional linear normalization in distinguishing between "road features" and "vehicle vibration."

[0056] Information accuracy weight matrix right Impact Analysis: The strength of the cross-covariance of historical data is negatively correlated with the weight of the deviation calculation. If "real-time impedance" and "expected bearing capacity" have historically fluctuated synchronously (strongly positively correlated), their current difference will significantly weaken its contribution to the calculation of the total deviation. Road surface physical properties are often driven by common environmental factors (including rainfall and temperature). Rainfall leads to a decrease in the "expected bearing capacity index," at which point the soil softens, and the "real-time impedance characteristics" also decrease, reducing resistance. This synchronous change is a "normal phenomenon consistent with physical laws," not an "abnormal road condition." Traditional Euclidean distance calculations will treat the above "synchronous decrease" as a huge numerical difference, leading to misjudgment. This embodiment introduces the Mahalanobis distance mechanism, utilizing... As a metric operator. In quadratic product. middle, Differences in "highly correlated directions" will be automatically weighted less. Only when the direction of change between the measured and predicted values ​​deviates from historical statistical patterns will the difference be included with high weight in the final result. In this embodiment, "the direction of change violates historical statistical patterns" includes situations where the expected load remains unchanged, but the real-time impedance fluctuates drastically; that is, changes orthogonal to the principal component direction. This essentially constructs a "physical law filter." It can automatically remove systematic deviations caused by environmental background factors and retain only independent deviations truly caused by local road surface heterogeneity (potholes, foreign objects).

[0057] This mechanism enables the differentiation between "normal environmental evolution" and "sudden road disasters," solving the technical problem of the high false alarm rate of the single threshold discrimination method under complex weather conditions.

[0058] The following explanation is provided for steps S4 and S5: 4) In step S4, generating a dynamic control instruction set that includes dynamic response correction instructions and model parameter evolution instructions specifically includes: Based on the amplitude components of the ecomechanical coupling deviation signal, the dynamic response correction instruction is generated by looking up a preset safety-ecological gain mapping table; the dynamic response correction instruction includes a weakening coefficient for reshaping the torque-speed curve of the drive unit. Based on the spatial coordinates and polarity of the ecomechanical coupling deviation signal, the model parameter evolution instruction is generated; the model parameter evolution instruction includes a weighted update factor for correcting the recovery rate of a specific cell in the environmental growth evolution model.

[0059] 4.1) Further, the step of generating the dynamic response correction command by looking up a preset safety-ecological gain mapping table based on the amplitude component of the ecomechanical coupling deviation signal specifically includes: An energy availability index characterizing the current power output potential of the vehicle's energy storage unit is introduced, along with a deviation trend factor characterizing the degree of drastic change in the ecomechanical coupling deviation signal over time. A three-dimensional dynamic gain tensor space is constructed, wherein the three dimensions of the tensor space correspond to the amplitude component of the ecomechanical coupling deviation signal, the energy availability index, and the deviation trend factor, respectively. Using a trilinear interpolation algorithm, a unique composite attenuation coefficient is retrieved and calculated in the dynamic gain tensor space, and used as the core control parameter of the dynamic response correction command. The data structure of the dynamic gain tensor space is configured such that the value of the retrieved composite attenuation coefficient increases with the increase of the energy availability index and decreases with the increase of the deviation trend factor, thereby realizing dynamic nonlinear compensation based on vehicle energy state and deviation deterioration rate.

[0060] 5) The execution of the dynamic control instruction set in step S5 specifically achieves the following parallel and related technical effects: Effect 1: Within 100 milliseconds of detecting the sudden negative value of the ecomechanical coupling deviation signal, the shear stress of the drive unit on the road surface is forcibly reduced by applying the dynamic response correction command, thereby preventing wheel slippage and maintaining vehicle tracking stability. Effect 2: Within a preset time period after the completion of the current passage, the state parameters of the corresponding area in the environmental growth evolution model are permanently corrected by applying the model parameter evolution instructions, thereby reducing the probability weight of subsequent vehicle planning paths passing through the area. When the system detects that the first data is missing or the confidence level of the ecomechanical coupling deviation signal is lower than a preset lower limit, it automatically enters a degraded operation mode, shields the second effect, and locks the dynamic response correction command to a conservative preset value.

[0061] 5.1) Further, in the step of permanently correcting the state parameters of the corresponding region in the environmental growth evolution model by applying the model parameter evolution instructions, a Bayesian confidence-weighted update mechanism is specifically adopted, which includes: The event observation credibility of the current single passage event is calculated. The calculation specifically includes: obtaining the statistical dispersion of the first data within the current time window as a first evaluation index, obtaining the kinematic parameters characterizing the current attitude stability of the vehicle as a second evaluation index, and determining the event observation credibility based on the weighted combination result of the first evaluation index and the second evaluation index. Read the historical prior state values ​​and their associated prior confidence weights stored in the corresponding spatial nodes of the environmental growth and evolution model; Based on the event observation confidence and the prior confidence weight, calculate the posterior update gain used to balance the old and new data; Using the posterior update gain, a weighted iterative operation is performed on the historical prior state values ​​to generate posterior corrected state values, which are then updated into the environmental growth and evolution model, thereby achieving closed-loop calibration based on probabilistic statistical laws.

[0062] The following are specific implementation instructions for the above content: In the practical application of "dynamic path planning for golf carts integrating golf course road conditions," the traditional two-dimensional static lookup table method (relying solely on deviation amplitude) ignores the vehicle's own energy state (including remaining battery power) and the suddenness of deviation occurrence when generating power response correction commands. This can lead to overly conservative or aggressive torque reduction when the battery is low or the deviation worsens sharply, causing power interruption or slippage and loss of control. When correcting the environmental evolution model, directly overwriting or simply averaging based on the deviation value of a single passage is easily affected by accidental sensor noise or atypical driving behavior (sudden braking), causing "catastrophic forgetting" or parameter oscillations in the cloud-based digital twin model, thus compromising the long-term stability of the path planning cost function. This preferred embodiment introduces a "multi-dimensional state adaptive tensor mapping mechanism" and a "Bayesian confidence-weighted update logic" to achieve more refined transient safety control and more robust long-term ecological evolution.

[0063] In this embodiment, step S4 is a three-dimensional spatial optimization process involving multi-physics coupling, specifically including: Energy availability index, denoted as Simply relying on the remaining battery charge (SOC) for linear torque limiting ignores the nonlinear degradation of the battery's power throughput under extreme temperatures. To address this issue, this parameter constructs a composite evaluation model integrating SOC and temperature characteristics. It represents the normalized capability of the vehicle's current powertrain system to respond to sudden torque surges without triggering undervoltage protection or overheating limits. It reflects the "physiological health" of the actuators.

[0064] The logic unit reads the State of Charge (SOC) and real-time battery pack temperature output from the Battery Management System (BMS), as well as the bus voltage ripple rate of the motor controller. It obtains the real-time SOC value output from the vehicle energy management module. This value is processed using a preset piecewise linear function: if the SOC is higher than a preset high-capacity threshold (set to 0.3 in this embodiment), the base capacity factor is directly assigned a value of 1.0; if the SOC is lower than the high-capacity threshold, the SOC value is divided by the high-capacity threshold to obtain a linearly decaying base capacity factor.

[0065] Obtain the average temperature value output by the thermal management sensor inside the battery pack. Calculate the penalty coefficient using a trapezoidal membership function. If the temperature is within a preset optimal range (set to 20℃ to 45℃ in this embodiment), the penalty coefficient is set to 1.0. If the temperature is below the lower limit of the optimal range or above the upper limit, calculate the absolute difference between the current temperature and the boundary of the optimal range. Multiply this difference by a preset temperature sensitivity decay rate (0.05 decays per degree Celsius), and subtract this product from 1.0. Compare the calculated result with a minimum safety margin of 0.1, and take the larger of the two as the final temperature penalty coefficient. Calculate the arithmetic product of the basic energy factor and the temperature penalty coefficient; the output result is the energy availability index. A result of 1.0 indicates abundant energy; a result of 0.1 indicates that the energy level is on the verge of "limping mode".

[0066] Deviation trend factor, denoted as ; To enable the system to not only respond to the current deviation amplitude but also to predict the rate of deviation deterioration, this factor needs to be introduced. The core challenge lies in mapping the unbounded derivative to a bounded scaling factor. This factor characterizes the rate of deterioration of the ecomechanical coupling deviation signal (EMCDT) within the most recent time window. In this embodiment, the deviation trend factor is determined using the following logical steps: The logic unit reads the first derivative of the ecomechanical coupling deviation signal, i.e., the trend component, and calculates its absolute value, denoted as . Because of absolute value operations, this input is always a non-negative real number.

[0067] Perform a weighted nonlinear mapping. The logic unit retrieves the pre-stored "sensitivity gain coefficient" (denoted as...). (Typical value is 5.0). Calculation and The product of and . This product is used as the input variable to the hyperbolic tangent function. Since the independent variable is non-negative, the output range of the standard tanh function is naturally limited to the interval [0,1). The value of determines the sensitivity to the rate of deviation deterioration. A higher ______ A value of 1 will cause the tanh function to enter the saturation region more quickly (the output approaches 1), indicating a strong response to even small deterioration trends; conversely, a lower value will cause the tanh function to enter the saturation region more quickly (the output approaches 1), indicating a strong response to even small deterioration trends; The value makes the response curve flatter.

[0068] The logic unit retrieves the pre-stored "maximum amplification increment" (denoted as...). (Typical value is 1.0). Calculate the output value of the tanh function obtained in the previous step and... The product of.

[0069] Add the product from the previous step to the baseline constant 1.0 to obtain the final deviation trend factor. This calculation logic ensures that when the deviation remains unchanged (derivative is 0), the factor is 1.0 (baseline state); when the deviation deteriorates sharply (derivative approaches infinity), the deviation trend factor... Smooth tends to The maximum enhancement state represented.

[0070] The composite weakening coefficient is denoted as It represents the core scalar ultimately included in the dynamic response correction instruction, used to directly multiply the target torque request value of the motor controller.

[0071] Obtained through interpolation calculations on the three-dimensional dynamic gain tensor space. Three-dimensional dynamic gain tensor space. This is the "data source" for the table lookup operation in step S4. Each "composite weakening coefficient" stored internally represents the optimal balance between the conflicting indicators of "vehicle safety (anti-skid)" and "passability (avoiding getting stuck)". It must be calibrated through controlled experiments, specifically using the following experimental calibration scheme: A hardware-in-the-loop (HIL) simulation platform was constructed, comprising a high-precision chassis dynamometer and an environmental simulation chamber. A coordinate grid corresponding to the three dimensions (deviation amplitude, energy availability index, and deviation trend factor) defined in step S4 was established within the simulation environment. For each discrete node in the three-dimensional space... Execute the following test loop: Lock the road surface impedance parameters of the simulation environment to match the coordinates. Lock the virtual battery state to match the coordinates and apply the corresponding The dynamic rate of change of road surface impedance. Within the range of [0.1, 1.0], the "tentative attenuation coefficient" is scanned incrementally in steps of 0.05. For each tentative value, two indicators are measured: wheel slip ratio (target: less than 20%) and vehicle longitudinal acceleration (target: maximize).

[0072] Define evaluation function The trial value that maximizes J is selected and solidified as the "composite weakening coefficient" at that node. Here, Ghy is the normalized slip ratio; GJa is the normalized acceleration. and These are the weighting coefficients of the corresponding parameters, and the values ​​of the weighting coefficients are all within the range of 0 to 1. The sum of the weighting coefficients is 1. This is the initial setting in this embodiment. After calibrating all nodes, a three-dimensional Gaussian filter is applied to the three-dimensional data array to eliminate experimental noise, ultimately generating a three-dimensional dynamic gain tensor space for vehicle deployment. Data file.

[0073] This step uses a "three-dimensional dynamic gain tensor space" as the core decision container, and this tensor space is pre-defined as a... The three-dimensional data matrix is ​​stored in non-volatile memory.

[0074] Dimension X (deviation amplitude axis): Corresponds to the amplitude component of the ecomechanical coupling deviation signal, subdivided into 20 quantization levels. Dimension Y (energy state axis): Corresponds to the energy availability index. It is further subdivided into 10 quantitative levels. Dimension Z (Trend Dynamic Axis): corresponds to the deviation trend factor. It is further divided into 5 quantitative levels.

[0075] The real-time acquired deviation amplitude, energy availability index, and deviation trend factor are mapped to continuous floating-point coordinates (x, y, z) in tensor space. A linear normalization mapping algorithm with boundary clamping is used during "coordinate positioning." This embodiment takes mapping the energy availability index (physical value P, range [0, 1]) to the tensor Y-axis (index range [0, 9]) as an example. The specific calculation logic is as follows: determine whether P exceeds the [0, 1] interval; if it does, forcibly truncate it to the boundary value to obtain the energy availability index after boundary clamping. That is, the effective value of the original energy availability index P after being restricted to the interval [0, 1] by the judgment logic (taking 0 when P is less than 0, taking 1 when P is greater than 1, and taking the original value when P is between 0 and 1), to ensure that the input value meets the normalization mapping requirements. Perform the calculation. ,in The total number of discrete nodes along the Y-axis is 10 in this embodiment. This represents a continuous floating-point coordinate value mapped onto the Y-axis of the tensor space. This value serves as the basis for subsequent trilinear interpolation operations; its integer part is used to determine the indices of adjacent grid nodes participating in the interpolation, while the fractional part is used as the interpolation weighting coefficient.

[0076] The calculation result is retained as a floating-point number. The integer part and the carry-over integer part of this floating-point coordinate determine the indices of adjacent grid nodes participating in trilinear interpolation, while the fractional part serves as the interpolation weight coefficient. The same mapping logic is performed on the X and Z axes. The eight discrete grid nodes surrounding the coordinate point are identified. Cubic linear interpolation is performed. Four intermediate values ​​are interpolated in the X-axis direction, followed by two intermediate values ​​in the Y-axis direction. Finally, these two intermediate values ​​are interpolated in the Z-axis direction to calculate a unique, high-precision composite attenuation coefficient. By using trilinear interpolation instead of piecewise table lookup, the mathematical smoothness of the control command is ensured under continuously changing operating conditions, eliminating the step vibration of motor torque caused by table lookup jumps and achieving flexible intervention.

[0077] The overall calculation logic of step S4 is as follows: Synchronously receive the ecomechanical coupling deviation signal (including amplitude and trend) output from step S3, as well as the energy state data uploaded from the underlying bus. Calculate the energy availability index. Sum of deviation trend factors The normalization process for all input quantities is completed. The normalized input quantities are then input to a pre-defined three-dimensional dynamic gain tensor space retrieval module, where trilinear interpolation is performed to calculate the composite attenuation coefficient. This composite attenuation coefficient is then encapsulated into a "dynamic response correction instruction" data packet conforming to the CAN bus protocol. Simultaneously, spatial coordinates are extracted from the deviation signal, and based on the polarity (positive / negative) of the deviation, a "model parameter evolution instruction" containing a "weighted update factor" is generated. At this point, the weighted update factor is not yet finalized but is marked as "pending Bayesian correction."

[0078] Furthermore, in step S5, Bayesian probability update theory is used to perform long-term ecological calibration (effect 2) to ensure that every evolution of the cloud model is rational and reversible. The reliability of event observation is denoted as The core of step S5 lies in how to update the existing prior model using uncertain new observations. Through Bayesian inference logic, the "stability" of the physical world is transformed into mathematical "credibility." Event observation credibility is used to characterize the "credibility" of the detected road condition anomaly data. It is used to determine whether this is a real physical geological change or an accidental sensor disturbance. The reliability of event observations is determined by comprehensively considering the consistency of data and the smoothness of vehicle movement. Specifically, within the current observation window of 50ms, the standard deviation of the "real-time impedance characteristic" sequence is calculated. This standard deviation is then divided by a preset reference noise constant to obtain the normalized dispersion. In the calculation of the "normalized dispersion," the "reference noise constant" (symbol denoted as...) The constant is an important normalization factor. Its physical meaning is: the inherent background noise standard deviation of the current sensor measurement data when the vehicle is stationary and the drive motor is in standby mode (i.e., no road excitation and no motor active torque ripple). This embodiment determines this constant through the following offline calibration procedure: The vehicle is placed in a stationary state, the system is powered on but the motor does not output torque; current ripple data is continuously collected for 10 seconds; the statistical standard deviation of this data segment is calculated, and 1.2 times this statistical value is set as the standard deviation. In this embodiment, the parameter is set to 0.05. Specifically, a safety margin of 1.2 times is introduced to avoid excessively large calculated normalized dispersion due to small random fluctuations in an ideal, extremely quiet system, which could lead to a misjudgment of high uncertainty. This setting ensures that the dispersion index only begins to have a substantial impact on reliability when the fluctuations caused by road surface excitation are significantly higher than the background noise level.

[0079] Obtain the absolute values ​​of lateral acceleration and yaw rate output from the vehicle's inertial measurement unit. Then, perform a weighted summation of the two values ​​(with weights of 0.6 and 0.4 respectively) to obtain the attitude disturbance index.

[0080] Adding the normalized dispersion to the attitude perturbation exponent yields the total uncertainty. Calculating the reciprocal of this total uncertainty plus 1 gives the event observation confidence level. . The closer the value is to 1, the more reliable the data is.

[0081] Prior confidence weights, denoted as It is stored in the cloud-based cell attributes and represents the cumulative reliability of the historical state values ​​(soil permeability) stored in the current cell. The more times a value has been verified, the higher its weight. This value increases logarithmically with the number of historical updates, with an upper limit locked at 0.95.

[0082] A posteriori update gain, denoted as The weighting coefficients used to correct model parameters determine the "magnitude" of how the new data alters the old model. Specifically, the prior confidence weights of the current cell are read from the cloud database. .Will Multiply This represents the degree to which "new data is reliable and old models are unreliable." Adding a normalized balance term to the numerator value results in a balance term equal to... and The average value. Divide the numerator by the denominator; the quotient is the a posteriori update gain. This logic ensures that model parameters are only modified when the new data is of extremely high quality and the confidence level of the old model is low; otherwise, the model's stability is maintained.

[0083] Obtain historical prior state values ​​in the current model Obtain the derived measured inversion state values. Before performing state fusion, a "co-domain mapping preprocessing" step is performed: a preset "impedance-compaction inversion model" is invoked (this model is the inverse function of the forward inference model described in step S1 or a lookup table mapping), and the currently measured "real-time impedance characteristic values" are... Convert to "equivalent measured compaction degree" This step ensures that the observed data and model parameters are within the same physical dimensions and domain. Calculation Compared with the "historical prior state values" stored in the current model The algebraic difference between them is denoted as the state residual. Update the gain using the obtained posterior. For state residuals Perform weighted scaling, i.e. The weighted and scaled residual values ​​are accumulated to... In this process, the final "posterior corrected state value" is obtained. This process is equivalent to applying a "soft correction" to the model parameters in accordance with the confidence weights, aligning them with the direction indicated by the experimental evidence.

[0084] Further update the prior confidence weight of this cell; the execution logic is to adjust the current... With this Perform probability synthesis with logical "OR" properties to ensure that the confidence level increases monotonically with the number of observations.

[0085] The overall calculation logic for step S5 is as follows: open the "fast channel" and the "slow channel" in parallel.

[0086] The fast track is for transient safety intervention: the onboard embedded unit directly reads the power response correction command (including composite attenuation coefficient) generated in step S4. In the next 10ms motor control cycle, the upper limit of the motor output torque is set to... This operation is completed within 100ms, forcibly reducing shear stress and achieving "tactile anti-slip". This represents the rated output torque (or calibrated peak torque) of the vehicle's drive motor. This is a fixed constant determined by the physical characteristics of the motor's hardware, representing the theoretically maximum power output of the motor without any limiting measures.

[0087] The slow channel serves as a long-term ecological calibration: it checks data integrity. If data is missing, the slow channel is immediately interrupted. If the data is valid, the confidence level of the event observations is calculated. and posterior update gain . use The environmental evolution weight parameters (soil recovery rate factor) at the corresponding coordinate points in the cloud-based environmental growth and evolution model are weighted and corrected. The updated parameters are written back to the cloud database as the benchmark for calculating the cost function in the next path planning iteration.

[0088] To achieve complete decoupling between the core algorithm logic and specific application scenarios, and to ensure the broad applicability of the technical solution in different golf course environments (mountainous courses, coastal courses), all parameters such as thresholds, weight coefficients, gain factors, and grid resolutions involved in the above embodiments are not hard-coded into the executable program. Instead, they adopt the following standardized configuration scheme: a non-transient computer-readable storage medium (EEPROM of the vehicle controller or an encrypted SD card partition) is configured, which stores a structured "system configuration description file". This file organizes data in key-value pair format. During the vehicle power-on initialization phase, the initialization bootloader is configured to read this configuration file. After the program performs integrity verification (including CRC check), it loads these parameter values ​​into the global variable area in memory. During operation, the core algorithm only accesses these global variables to obtain specific parameter values.

[0089] The following further elaboration addresses the above: The core output command parameter of this technical solution is the composite weakening coefficient. Composite weakening coefficient As a dimensionless scalar, its value is strictly limited to the closed interval [0.1, 1.0].

[0090] when At this time, the system is in the "full power response state". Approaching low values, and with the energy availability index at a high level. The vehicle travels on stable road surfaces highly consistent with the digital twin model, including dry fairways, and the battery pack is within its optimal discharge window. Determining there is no risk of slippage or collapse, the drive unit is authorized to respond to the driver's acceleration requests at 100% of its rated capacity, thereby maximizing traffic efficiency.

[0091] when At this time, the vehicle is in a "deep safety clamp" limp mode. This is triggered by a high deviation amplitude, a drastic trend of deviation deterioration, or a severe lack of battery power output. The vehicle may be encountering unforeseen extreme soft substrate (hidden mud) or is on the verge of high dynamic instability. The motor output torque is forcibly limited to 10% of its rated value. This torque level is physically calibrated as a critical value "sufficient to maintain vehicle creep and extrication without generating destructive stress sufficient to damage the soil shear strength," thus achieving the highest level of safety with "zero slippage."

[0092] Regarding the decision The three key input dimensions and their impact logic are as follows: With output It is a nonlinear, negatively correlated S-shaped decay. This characterizes the degree to which physical road conditions deviate from theoretical expectations. As the deviation increases (the road surface suddenly softens), the risk of slippage rises exponentially. The S-shaped attenuation design remains insensitive in the low deviation zone (filtering noise) and quickly cuts off torque in the medium to high deviation zone (responding to risk), conforming to the mechanical characteristics of tire-soil interaction. This directly supports the technical effect of "transient safety intervention," ensuring that within a millisecond window of detecting abnormal road conditions, the driving force can be actively weakened before wheel slippage occurs.

[0093] Energy Availability Index With output It is a monotonically positive correlation. This maps the battery's current power throughput capability. When the battery is at low temperature or low charge, forcibly requesting high torque can cause a voltage drop, triggering undervoltage protection (BMS shutdown), which in turn poses a risk of power interruption. The positive correlation design forces the torque request to always be within the physical capacity limits of the battery. This solves the power interruption problem caused by the "hard torque limitation" of traditional solutions, supporting the robustness of this invention throughout its entire life cycle and under all climatic conditions.

[0094] Deviation trend factor With output It is a highly sensitive negative correlation, with hyperbolic tangent accelerating the decay. It is a mapping of the first derivative. If the deviation is rapidly worsening (large derivative), it indicates that the vehicle is rapidly entering a danger zone. In this case, a larger torque reduction must be implemented before the deviation amplitude reaches its peak, i.e., the "derivative-first" control strategy. This design provides "predictive defense" capability, effectively eliminating control lag compared to existing technologies that rely solely on amplitude, and improving dynamic safety at high speeds.

[0095] Furthermore, to verify the effectiveness of the "dynamic instruction generation mechanism based on multidimensional tensor mapping" proposed in this invention under complex road conditions, this embodiment establishes a high-fidelity hardware-in-the-Loop (HIL) test environment. The simulation platform integrates a seven-degree-of-freedom (7-DOF) vehicle dynamics model and a magic-formula tire model to accurately reproduce the tire adhesion characteristics of non-rigid road surfaces. The power system maps the dynamic response characteristics of a 48V AC asynchronous motor. In this simulation configuration, the key control parameters are set as follows: Sensitivity gain coefficient Set to 5.0 to capture impedance change rates in the microsecond range.

[0096] Trend maximum magnification increment Set to 4.0. This represents the deviation trend factor. The dynamic value range has been extended to the [1.0, 5.0] interval to ensure a strong response to extremely rapid road surface collapses or sudden skidding risks.

[0097] The control group uses the "two-dimensional static lookup table method" commonly used in existing technologies, which determines the torque reduction coefficient by linearly looking up the table based solely on the deviation amplitude, ignoring the energy state and trend terms, and its maximum reduction coefficient is conservatively limited to 0.4.

[0098] This experiment selected six representative golf course driving scenarios to focus on examining the differences and advantages of the present invention in control command output compared with existing technologies after introducing energy constraints and trend prediction, as shown in Table 1 below: Table 1: Performance Comparison and Verification Data of Multidimensional Tensor Mapping Mechanism and Existing Technologies Dynamic risk exposure, denoted as DRE; this parameter is an intermediate variable generated internally by this invention, and its value is obtained by multiplying the current degree with the deviation trend factor, i.e. This parameter quantitatively characterizes the "momentum" of the road condition risk currently faced by the vehicle. It integrates the spatial breadth (amplitude) and temporal intensity (rate) of the risk, serving as a comprehensive indicator to measure the probability of irreversible instability of the system.

[0099] Based on the simulation data shown in Table 1, the key technical effects of the present invention are analyzed as follows: Verification of predictive defense against deviation trend factors; comparative analysis was conducted using scenarios S-02 and S-03. In both experiments, the instantaneous deviation amplitude detected by the vehicles was completely identical. This means that the "surface softness" is the same at the current moment.

[0100] Due to a lack of awareness of the time dimension, the existing technology judges solely based on amplitude, outputting a moderate attenuation coefficient of 0.55 in both scenarios. This indicates that in the S-03 (sudden trap) scenario, the vehicle may be driven into the mud by this torque due to response lag, resulting in sinking.

[0101] This invention captures a deviation trend factor of 4.5 in S-03, indicating that the deviation is worsening at a high acceleration. This causes the calculated Dynamic Risk Exposure (DRE) to surge to 2.70. Based on this, nonlinear compensation is performed, forcefully reducing the output command to 0.25. Compared to existing technologies, this invention improves the response depth under sudden emergencies by 54.5%; this data strongly demonstrates the technical feature of this invention regarding "dynamic nonlinear compensation based on deviation trend factor," which can eliminate control lag and achieve proactive risk avoidance.

[0102] Adaptive robustness verification for energy state: In scenario S-04, although the road surface condition is good However, the energy availability index is only 0.30, simulating low temperature or low battery conditions. Existing technologies ignore the energy bottleneck and still demand a high torque of 0.90. In a physical vehicle, this would trigger the undervoltage protection of the battery management system (BMS), causing the vehicle to lose power instantaneously. This invention, through multi-dimensional tensor mapping, actively limits the torque to 0.28. Although it sacrifices instantaneous acceleration, it ensures the continuous operability of the vehicle under harsh conditions. This result directly translates to "Reshaping the drive curve based on the energy availability index," demonstrating all-weather robustness.

[0103] In scenario S-05, facing the triple constraints of high deviation, low battery, and rapid deterioration, this invention outputs a command of 0.12, closely matching the set theoretical minimum value of 0.10. This verifies the convergence of the algorithm under complex boundary conditions, ensuring that the system can still maintain a safety baseline under the "worst-case" scenario, preventing loss of control due to logical conflicts.

[0104] Furthermore, the verifiable performance metric is defined as the longitudinal slip ratio, denoted as . This is the most direct physical quantity for measuring tire adhesion. This is the linear stable region. This is the region of nonlinear instability.

[0105] Simulations show that the composite weakening coefficient There is a negatively correlated exponential decay relationship with the longitudinal slip ratio. As the shear stress on the road surface decreases, the longitudinal slip ratio converges accordingly.

[0106] Inflection point of dynamic saturation: when When the longitudinal slip ratio changes by approximately 0, it indicates that limiting torque provides no significant safety benefit and only reduces efficiency.

[0107] Critical instability inflection point: when At this point, the longitudinal slip ratio rapidly decreases from the unstable region (>20%) to the deep attachment region (<10%). Through OTD methodology derivation, the following practical application range with a solid physical basis was determined: Zone 1: Free Movement Zone Its boundary is determined by the power saturation inflection point. Within this range, the road impedance characteristics are highly consistent with theoretical predictions, and the energy is abundant, with the tires operating in the linear initial segment of the slip ratio curve. The drive unit executes a "torque transparency mode," without attenuating the driver's acceleration requests, only performing conventional smoothing filtering to prioritize traffic efficiency and driving pleasure.

[0108] Interval Two: Dynamic Flexible Intervention Zone Its boundary lies between the saturation inflection point and the instability inflection point. This range corresponds to a moderate degree of unexpected softness in the road surface or a vehicle operating at medium to high energy consumption. The drive unit performs "proportional reduction control." It adjusts the motor torque-speed external characteristic curve in real time, "shaving off" high-frequency, high-torque demands, and suppresses the increasing trend of slip ratio by fine-tuning the output without the driver's notice.

[0109] Section 3: Forced Safety Clamping Zone Its boundary is determined by the critical instability inflection point. Below this value, it indicates that a structural break has been detected between the physical truth and the theoretical prediction. The drive unit enters a "high-priority protection mode". Regardless of the driver's pedal depth, the output is forcibly locked within a low threshold; simultaneously, the "slow channel" update mechanism in step S5 is triggered to maximize the posterior update gain. The environmental evolution parameters of the area are marked and permanently corrected to prevent subsequent vehicles from repeating the same mistakes.

[0110] Figure 3 The simulation results aim to simulate the vehicle's dynamic response strategy under six different typical stadium conditions (S-01 to S-06). In the figure, the horizontal axis represents the different simulation test scenario numbers, and the vertical axis represents the composite attenuation coefficient of the control system output. The lower the composite weakening coefficient, the greater the restriction on the driving torque. The blue bars represent the calculation output using the multidimensional tensor mapping mechanism proposed in this invention, while the gray bars represent the calculation output using existing technology (static lookup table method).

[0111] like Figure 3 As shown, the simulation results clearly reveal the significant advantages of this invention in handling high dynamic risks. In conventional scenarios such as baseline cruise (S-01) and sensor noise (S-06), the outputs of both are highly consistent, indicating that this invention does not introduce unnecessary intervention. However, in the S-03 (sudden trap) scenario, the output of the prior art remains at 0.55, while the output of this invention is significantly reduced to 0.25. This indicates that, thanks to the deviation trend factor… With the introduction of this technology, the present invention can theoretically identify the "momentum" of deteriorating road conditions and reduce torque output by 54.5% earlier and more deeply than existing technologies. Similarly, in the S-05 (complex hazard) scenario, the present invention further responds to energy state constraints, reducing the composite attenuation coefficient to a limit safety value of 0.12.

[0112] Figure 5The simulation results aim to simulate the vehicle's dynamic response strategy under different energy states and road condition trends. In the figure, the horizontal axis represents the energy availability index, reflecting the vehicle's current power output potential; the Y-axis (vertical axis) represents the deviation trend factor, reflecting the severity of road condition deterioration; the Z-axis (vertical axis) represents the calculated composite attenuation coefficient, the lower the value, the stronger the system's restriction on driving torque (i.e., the greater the safety intervention). The colored surface represents the output of the three-dimensional dynamic gain tensor algorithm used in this invention, and the semi-transparent gray plane represents the output of the fixed threshold static mapping algorithm used in the prior art.

[0113] Figure 5 Simulation results clearly reveal the fundamental difference between this invention and existing technologies in terms of control strategy. The gray plane shows that existing technologies do not adjust their strategies according to changes in energy and trend, maintaining a single response level. In contrast, the colored surface of this invention exhibits significant nonlinear characteristics: in the region of "high trend factor - low energy index" (the deep red valley area where the surface dips in the figure), the composite attenuation coefficient drops sharply. This indicates that the method proposed in this invention can theoretically identify high-risk operating condition combinations and automatically implement more aggressive torque limiting to prevent slippage or power interruption; while in the "low trend - high energy" region, the surface remains at a high level, ensuring smooth passage. This figure provides direct, quantitative simulation evidence for the technical effect of the "dynamic nonlinear compensation based on vehicle energy state and deviation deterioration rate" claimed in this invention.

[0114] like Figure 1 As shown, this invention demonstrates the complete technical route of the system from physical perception to logical control. The golf cart, illustrated on the left, serves as the execution entity of the method, and is connected to the technical route flowchart on the right via data flow lines. The flowchart on the right contains four core steps: The first step is the "Multi-source Data Perception" module (corresponding to steps S1 and S2), used to simultaneously acquire real-time impedance characteristics representing electrical fluctuations and vertical oscillations at the vehicle's drive end, as well as the expected load-bearing index derived from the environmental evolution model. The second step is the "Eco-mechanical Coupling Calculation" module (corresponding to step S3), which utilizes a heterogeneous data homo-domain mapping program to map the above two types of data to the same dimensionless space, calculates and outputs the eco-mechanical coupling deviation signal. The third step is the "Dynamic Command Generation" module (corresponding to step S4), which generates a dynamic control command set based on the deviation signal, including dynamic response correction (for the motor) and model parameter evolution (for the environmental model). The fourth step is the "Closed-Loop Execution and Update" module (corresponding to step S5), which executes the commands to adjust the vehicle's current dynamic response strategy and corrects the environmental evolution weight parameters, thereby achieving a closed-loop update of the subsequent path planning cost function.

[0115] The computational logic involved in this application can be constructed using algorithms such as regression analysis in machine learning, establishing a mathematical model by analyzing the inherent trends and interrelationships of the collected parameters. This process can be implemented using specialized computational tools (such as Python's Scikit-learn library or the R language environment). Throughout all calculations, to eliminate the influence of different physical dimensions and ensure that data is compared and analyzed on the same scale, the input parameters in each formula are dimensionless. The dimensionless techniques used include, but are not limited to, max-min normalization or Z-score standardization.

[0116] The algorithm of this invention is implemented as a Python script. Before executing the core logic, the program first executes a data loading module (e.g., using the widely used pandas library in Python) configured to read the aforementioned spreadsheet file and load its contents into the program's working memory (e.g., a DataFrame data structure). Subsequent algorithm steps will directly query and retrieve the required configuration parameters from this in-memory data structure.

[0117] It should be emphasized that the foregoing embodiments are merely illustrative of preferred implementations of the present invention and are not intended to limit the scope of protection of the present invention. This application also provides a computer-readable storage medium having computer program instructions stored thereon.

Claims

1. A method for dynamic path planning of golf carts that integrates golf course road conditions, characterized in that, The specific steps include: S1: Acquire the first data characterizing the real-time impedance characteristics; the first data is a quantitative value characterizing the physical and mechanical impedance characteristics of the current driving interface, calculated by a frequency domain feature transformation algorithm based on the electrical fluctuation attributes of the vehicle drive actuator and the vertical oscillation attributes of the inertial sensing end; and simultaneously acquire the kinematic parameters characterizing the current motion state of the vehicle. S2: Obtain the second data representing the expected carrying capacity index; the second data is a scalarized index representing the physical carrying capacity potential of the road surface at the current geographic coordinate point under theoretical conditions, generated based on a preset environmental growth and evolution model and historical traffic load records at the current location; S3: Using the ecomechanical coupling deviation calculation model, execute the heterogeneous data same-domain mapping procedure; specifically, first map the first data and the second data to the preset dimensionless reference traffic state space, then calculate the ecomechanical coupling deviation measure of the two in the dimensionless reference traffic state space, and output the ecomechanical coupling deviation measure as an ecomechanical coupling deviation signal used to characterize the degree of deviation of the true physical value of the road surface from the theoretical prediction value. S4: Based on the aforementioned ecomechanical coupling deviation signal, generate a dynamic control instruction set that includes dynamic response correction instructions and model parameter evolution instructions; S5: Execute the dynamic control instruction set to adjust the dynamic traffic response strategy of the vehicle passing through the current road segment in the first time domain window, and correct the environmental evolution weight parameters of the corresponding spatial nodes in the environmental growth and evolution model in the second time domain window, thereby realizing the closed-loop update of the subsequent path planning cost function.

2. The method for dynamic path planning of golf carts based on golf course conditions according to claim 1, characterized in that: In step S1, the first data characterizing the real-time impedance features is obtained, specifically including: Acquire the high-frequency current ripple sequence of the drive unit within a preset time window; Perform a fast Fourier transform on the high-frequency current ripple sequence to extract the energy spectral density of a specific frequency band; The energy spectral density is input into a pre-trained long short-term memory network, which outputs the real-time impedance characteristics, wherein the real-time impedance characteristics characterize the shear drag coefficient of the road medium on the tire.

3. The method for dynamic path planning of golf carts based on golf course conditions according to claim 2, characterized in that: In step S2, second data characterizing the expected carrying capacity index is obtained, specifically including: Query the road network cell state matrix stored in the local cache. Each cell in the road network cell state matrix corresponds to a discrete grid in the geospatial space. Read the cumulative compaction status value of the cell corresponding to the current vehicle location and the timestamp of the most recent update; Based on cellular automata rules and combined with current meteorological parameters, the soil recovery state of the cells at the current moment is deduced, and the expected carrying capacity index is calculated.

4. The method for dynamic path planning of golf carts based on golf course conditions according to claim 3, characterized in that: In step S3, the generation of the ecomechanical coupling deviation signal using the ecomechanical coupling deviation calculation model specifically includes: The real-time impedance characteristics and the expected carrying capacity index are converted into dimensionless standard values ​​using a preset normalization mapping function; the ecomechanical coupling deviation between the two standard values ​​is calculated. The ecomechanical coupling deviation metric includes an amplitude component and a trend component, wherein the amplitude component represents the degree of deviation between the measured value and the predicted value at the current moment, and the trend component represents the first derivative of the degree of deviation with time; The ecomechanical coupling deviation signal is output only when the amplitude component exceeds a preset confidence threshold.

5. The method for dynamic path planning of golf carts based on golf course conditions according to claim 4, characterized in that: In the step of converting the real-time impedance characteristics and the expected load-bearing index into dimensionless standard values, a dynamic nonlinear normalization strategy is specifically adopted, which includes: Obtain the kinetic energy mode adjustment factor characterizing the current motion intensity of the vehicle; construct a sigmoid activation function with a variable slope, the steepness of which is dynamically modulated by the kinetic energy mode adjustment factor; The real-time impedance characteristics and the expected load-bearing index are respectively input into the S-type activation function, and the output is the dimensionless standard value. The kinetic mode adjustment factor is positively correlated with the vehicle's motion intensity, which makes the linear response range of the S-shaped activation function narrower as the motion intensity increases, thereby suppressing the influence of high-frequency noise on the standard value.

6. The method for dynamic path planning of golf carts based on golf course conditions according to claim 5, characterized in that: In the step of calculating the ecomechanical coupling deviation measure between two standard values, a covariance-weighted deviation measure mechanism is specifically adopted, which includes: Maintain a sliding observation window that stores a historical sequence of dimensionless standard values; based on the historical sequence of dimensionless standard values ​​of the real-time impedance characteristics and the expected bearing index recorded within the sliding observation window, calculate the covariance matrix used to characterize the correlation of data fluctuations; perform matrix inversion on the covariance matrix to obtain the information accuracy weight matrix; Calculate the original difference vector between the standard value of the real-time impedance characteristic and the standard value of the expected load index; Calculate the quadratic product of the original difference vector and the information precision weight matrix, and determine the arithmetic square root of the quadratic product as the amplitude component of the ecomechanical coupling deviation metric.

7. The method for dynamic path planning of golf carts based on golf course conditions according to claim 6, characterized in that: In step S4, generating a dynamic control instruction set that includes dynamic response correction instructions and model parameter evolution instructions specifically includes: Based on the amplitude components of the ecomechanical coupling deviation signal, the dynamic response correction instruction is generated by looking up a preset safety-ecological gain mapping table; the dynamic response correction instruction includes a composite attenuation coefficient for reshaping the torque-speed curve of the drive unit. Based on the spatial coordinates and polarity of the ecomechanical coupling deviation signal, the model parameter evolution instruction is generated; the model parameter evolution instruction includes a weighted update factor for correcting the recovery rate of a specific cell in the environmental growth evolution model.

8. The method for dynamic path planning of golf carts based on golf course conditions according to claim 7, characterized in that: The step of generating the dynamic response correction command based on the amplitude components of the ecomechanical coupling deviation signal by looking up a preset safety-ecological gain mapping table specifically includes: An energy availability index characterizing the current power output potential of the vehicle's energy storage unit is introduced, along with a deviation trend factor characterizing the degree of drastic change in the ecomechanical coupling deviation signal over time. A three-dimensional dynamic gain tensor space is constructed, wherein the three dimensions of the dynamic gain tensor space correspond to the amplitude component, energy availability index, and deviation trend factor of the ecomechanical coupling deviation signal, respectively. Using a trilinear interpolation algorithm, a unique composite attenuation coefficient is retrieved and calculated in the dynamic gain tensor space, and used as the core control parameter of the dynamic response correction command. The data structure of the dynamic gain tensor space is configured such that the value of the retrieved composite attenuation coefficient increases with the increase of the energy availability index and decreases with the increase of the deviation trend factor, thereby realizing dynamic nonlinear compensation based on vehicle energy state and deviation deterioration rate.

9. The method for dynamic path planning of golf carts based on golf course conditions according to claim 8, characterized in that: Step S5 executes the dynamic control instruction set, specifically achieving the following parallel and interconnected technical effects: Effect 1: Within 100 milliseconds of detecting the sudden negative value of the ecomechanical coupling deviation signal, the shear stress of the drive unit on the road surface is forcibly reduced by applying the dynamic response correction command, thereby preventing wheel slippage and maintaining vehicle tracking stability. Effect 2: Within a preset time period after the completion of the current passage, the state parameters of the corresponding area in the environmental growth evolution model are permanently corrected by applying the model parameter evolution instructions, thereby reducing the probability weight of subsequent vehicle planning paths passing through the area. When the first data is detected to be missing or the confidence level of the ecomechanical coupling deviation signal is lower than the preset lower limit, the system automatically enters the degraded operation mode, shields the second effect, and locks the dynamic response correction command to a conservative preset value.

10. The method for dynamic path planning of golf carts based on golf course conditions according to claim 9, characterized in that: In the step of permanently correcting the state parameters of the corresponding region in the environmental growth evolution model by applying the model parameter evolution instructions, a Bayesian confidence-weighted update mechanism is specifically adopted, which includes: The event observation credibility of the current single passage event is calculated. The calculation specifically includes: obtaining the statistical dispersion of the first data within the current time window as the first evaluation index, obtaining the kinematic parameters characterizing the current attitude stability of the vehicle as the second evaluation index, and determining the event observation credibility based on the weighted combination result of the first evaluation index and the second evaluation index. Read the historical prior state values ​​and their associated prior confidence weights stored in the corresponding spatial nodes of the environmental growth and evolution model; Based on the event observation confidence and the prior confidence weight, calculate the posterior update gain used to balance the old and new data; Using the posterior update gain, a weighted iterative operation is performed on the historical prior state values ​​to generate posterior corrected state values, and the posterior corrected state values ​​are updated into the environmental growth and evolution model, thereby achieving closed-loop calibration based on probabilistic statistical laws.

Images