A digital-twin-based drive-by-wire chassis energy consumption simulation method and system

Abstract

The application discloses a kind of based on digital twinning linear control chassis energy consumption simulation method and system, including following steps: S1, constructs digital twinning model, generates state sequence by collecting and processing operation data;S2, using neural control differential equation extraction fusion state;S3, based on fourier neural operator simulation calculation power sequence;S4, through long-term memory network prediction trend, total energy consumption is generated using Simpson integral;S5, disturbance control parameter is generated disturbance energy consumption sequence using SASP algorithm;S6, compare two kinds of energy consumption, and control parameter is optimized using PPO algorithm;S7, according to the total energy consumption of optimization result regeneration, and with actual energy consumption error analysis, update digital twinning model.The application fuses neural control differential equation, fourier neural operator etc., with the advantages of high modeling precision, energy consumption prediction accuracy and high simulation efficiency.

Description

Technical Field

[0001] This invention relates to the field of energy consumption prediction technology, and in particular to a method and system for simulating energy consumption of a drive-by-wire chassis based on digital twins. Background Technology

[0002] In modern intelligent driving vehicles and automated vehicle systems, the drive-by-wire chassis, as a key actuator, undertakes core control tasks such as steering, braking, and driving. Its operating efficiency directly affects the vehicle's energy consumption performance and dynamic response capability. Traditional drive-by-wire chassis energy consumption analysis typically relies on real-vehicle testing and static energy assessment models, which are insufficient to meet the energy consumption prediction needs under multiple operating conditions and combinations of multi-parameter control strategies. Especially in complex dynamic scenarios, there is a lack of high-precision modeling and response evaluation mechanisms for the impact of different control strategies on energy consumption.

[0003] To enhance the flexibility and foresight of energy consumption modeling, recent research has introduced digital twin technology to achieve virtual-physical energy consumption simulation and prediction by constructing a virtual mapping model of the chassis system. These methods typically employ physical modeling or finite element analysis to simulate the actuator response process, then estimate energy consumption changes based on input commands and sensor feedback data. However, this technology still suffers from the following shortcomings: First, the model structure is highly rigid, making it difficult to adapt to the dynamic adjustment requirements of different vehicles or control strategies; second, simulation efficiency is limited by the accuracy of the physical solution, hindering real-time performance evaluation and strategy optimization; third, under complex control parameter perturbations, there is a lack of effective learning mechanisms for strategy evolution in the direction of energy consumption minimization, making it difficult for the model to form a closed-loop self-updating mechanism.

[0004] Furthermore, most existing studies employ rule-based parameter scanning or heuristic optimization methods, which fail to fully utilize the continuous time-dependent information in historical operational data and struggle to achieve efficient policy iteration and convergence in high-dimensional control spaces. The organic integration of fusion state modeling, nonlinear response feature extraction, energy consumption trend prediction, and reinforcement learning policy optimization remains a gap, limiting the practical deployment and engineering implementation of digital twins in the field of online control chassis energy consumption simulation.

[0005] Therefore, how to provide a method and system for simulating the energy consumption of a drive-by-wire chassis based on digital twins is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a method and system for simulating the energy consumption of a drive-by-wire chassis based on digital twins. This invention integrates neural control differential equations, Fourier neural operators, long short-term memory networks, SASP perturbation mechanisms, and PPO reinforcement learning algorithms to construct a digital twin model with continuous-time modeling capabilities, frequency domain response analysis capabilities, and control strategy self-optimization capabilities. It details the entire process from state feature extraction, energy consumption trend prediction, parameter perturbation control to model closed-loop update, and has the advantages of high modeling accuracy, accurate energy consumption prediction, fast control strategy convergence, and high simulation efficiency.

[0007] A method for simulating the energy consumption of a drive-by-wire chassis based on digital twins according to an embodiment of the present invention includes the following steps: S1. Construct a digital twin model of the drive-by-wire chassis, collect and preprocess the driving-by-wire chassis's operating data, and construct a state data sequence; S2. Perform feature extraction and fusion operations on the state data sequence using neural control differential equations, record the hidden states in continuous time, and obtain the fused state sequence. S3. In the digital twin model, Fourier neural operators are used to perform simulation based on the fused state sequence to calculate the steering, braking and driving power corresponding to each time step, forming a power sequence. S4. A long short-term memory network is used to predict the trend of power sequence changes. At the same time, the Simpson integral is used to perform time integration operation on the power sequence. The error compensation is performed on the integration result based on the prediction result to generate the total energy consumption sequence. S5. Initialize the control parameter set, use the SASP algorithm to perform perturbation processing on the control parameter set, and calculate the perturbation energy consumption sequence in the digital twin model based on the perturbation results; S6. Compare and analyze the disturbance energy consumption sequence with the total energy consumption sequence, construct a loss function based on the analysis results, and use the PPO algorithm to iteratively update the control parameter set until the loss function converges. S7. Load the updated control parameter set into the digital twin model to regenerate the total energy consumption sequence, calculate the error value between the actual energy consumption and the actual energy consumption, and update the digital twin model based on the error value.

[0008] Optionally, the digital twin model represents a simulation model that virtually maps the physical structure, control behavior, and energy consumption response process of the steer-by-wire chassis. The operating data includes sensor sampling values ​​of steering, braking, driving, and vehicle body status, such as steering angle, brake pedal displacement, drive motor speed, vehicle acceleration, vehicle speed, motor current, and motor voltage. The preprocessing includes time alignment, missing data completion, Z-Score normalization, noise filtering, and structured stitching operations. The control parameter set represents a set of parameters used to adjust the control behavior of the steer-by-wire chassis, including multiple numerical parameters affecting steering, braking, and driving behavior. The loss function represents a functional expression used to evaluate the degree of difference between the disturbance energy consumption sequence and the target energy consumption sequence.

[0009] Optionally, the construction and operation process of the digital twin model specifically includes: Based on the structural design information and control logic specifications of the drive-by-wire chassis, steering, braking and drive control rules are constructed, and the mapping relationship between control input, structural response and power output is defined to form control behavior units; Based on sensor deployment information and sampling rules, a correspondence between physical parameters and sampling data is established, and a state mapping matrix is ​​constructed. Energy consumption calculation logic is embedded in the control behavior unit, and Fourier neural operators are used to set power calculation paths for steering response value, braking response value and driving response value respectively to construct an energy consumption response mechanism. The state mapping matrix and energy consumption response mechanism are integrated, and the parameter set for subsequent perturbation optimization and error update is initialized to complete the construction of the digital twin model; The fused state sequence is received in the digital twin model. The Fourier neural operator is called to perform multi-level frequency domain transformation and nonlinear activation operation on the fused state sequence to generate steering power, braking power and driving power at each time step and splice them into a multi-dimensional power vector to construct a power sequence.

[0010] Optionally, S2 specifically includes: S21. Map the state data sequence to a continuous time axis according to the time step order, construct a driving path for the state data of each time step, and establish a dynamic equation structure based on neural control differential equations. S22. Perform hidden state update operation on the dynamic equation structure through vector field, calculate the change value of hidden state between each adjacent time step, and generate hidden state trajectory. S23. Perform multi-layer GELU function transformation and feature fusion operation on the hidden state trajectory to splice and reconstruct the hidden states in different time intervals to form a fused state representation that maintains temporal continuity. S24. Discretize the fusion state representation according to the original time sequence to generate a fusion state sequence corresponding to the state data sequence.

[0011] Optionally, S21 specifically includes: S211. Construct a continuous time axis based on the state data sequence and the preset data acquisition frequency. The data acquisition frequency represents the preset frequency of acquiring the driving data of the drive-by-wire chassis through the sensor. Each time step in the continuous time axis corresponds to a set of state data. S212. Based on the continuous time axis, perform time-domain interpolation on the state data sequence using the cubic spline interpolation method to construct the interpolation function, specifically including: Extract the time index and corresponding state data of each time step in the state data sequence to construct an initial sample set; Arrange the time indices in the initial sample set in order to determine the interpolation interval between adjacent time steps; A cubic spline interpolation function is constructed for each interpolation interval according to preset rules, and conditions for continuity of function values, continuity of the first derivative, and continuity of the second derivative are set. All coefficients in the cubic spline interpolation function are solved based on three types of continuity conditions, and all the solution results are concatenated to form the interpolation function. S213. Based on the neural control differential equation, a dynamic equation structure is constructed using the interpolation function as the control variable. The dynamic equation structure is used to define the differential evolution rules of the hidden state in continuous time.

[0012] Optionally, S3 specifically includes: S31. Receive the fused state sequence in the digital twin model, construct a multidimensional state tensor based on the fused state sequence in chronological order, and perform frequency domain mapping operation according to the preset mapping rules to generate frequency domain feature representation. S32. The frequency domain feature representation is subjected to multi-layer frequency domain convolution and frequency weight update operations using Fourier neural operators. The convolution results of each layer are then weighted and superimposed according to a preset weight set to form the frequency domain response features, specifically including: The entire frequency domain feature representation is divided into multiple subsets according to a preset frequency range, and a set of complex convolution kernels of a fixed size is configured for each subset; In each subset, the corresponding complex convolution kernel is used to perform complex convolution operation on the frequency domain feature representation within the subset to extract the real and imaginary features and construct a set of frequency responses. Based on the set frequency amplitude adjustment rules, frequency weighting parameters are configured for each frequency response set. Multiplication and scaling operations are performed on the real and imaginary features respectively according to the frequency weighting parameters to form a weighted frequency set. All weighted frequency sets are concatenated according to the frequency subset partitioning order and then aligned in dimensions. A linear weighted fusion operation is performed on the alignment results based on a preset set of weight coefficients to obtain the frequency domain response features; S33. According to the preset discrete Fourier inverse transform rules, perform inverse frequency domain transform operation on the frequency domain response features to convert the frequency domain response features into time domain response features, and perform Swish function activation operation to generate state response sequence. S34. In the digital twin model, the state response sequence of each time step is mapped to steering response value, braking response value and driving response value respectively. S35. Perform power calculation operations on the steering response value, braking response value and driving response value respectively to generate steering power, braking power and driving power at each time step; S36. The power calculation results are spliced ​​and rearranged in chronological order to construct a multidimensional power vector, and all multidimensional power vectors are summarized to form a power sequence.

[0013] Optionally, S4 specifically includes: S41. Convert the power sequence into a multi-dimensional time series tensor in chronological order, and extract the steering power, braking power and driving power of each time step as channel features. S42. Input the multidimensional temporal tensor into the long short-term memory network, and perform forward and backward propagation operations respectively. Extract the bidirectional hidden states at each time step and splice them to generate a power trend sequence. S43. Using the Simpson integral method, perform time integration on the power sequence to calculate steering energy consumption, braking energy consumption, and driving energy consumption respectively, forming a preliminary energy consumption sequence, specifically including: Take every three adjacent time steps in the power sequence as an integration unit, and extract the steering power, braking power and driving power corresponding to the start time step, middle time step and end time step in the integration unit; A fixed weight value is assigned to each of the three time steps in each integration unit. The steering power, braking power and driving power are weighted and summed. The product is calculated by combining the time step intervals to generate the steering energy value, braking energy value and driving energy value corresponding to the integration unit. The three types of energy values ​​obtained from all the integration units are summed up to form steering energy consumption, braking energy consumption and driving energy consumption, and then combined in a preset order to form a preliminary energy consumption sequence. S44. Calculate the difference between the power trend sequence and the power sequence, extract the prediction error value at each time step, construct a set of compensation factors based on the prediction error value, and use the set of compensation factors to perform a weighted correction operation on the preliminary energy consumption sequence step by step to generate a compensated energy consumption sequence. S45. Perform a channel-by-channel accumulation operation on the compensated energy consumption sequence according to the channel characteristics, concatenate the accumulation results into an energy consumption vector set in chronological order, and summarize all energy consumption vector sets to generate the total energy consumption sequence.

[0014] Optionally, S5 specifically includes: S51. Construct a set of control parameters based on the preset control dimensions, group all control parameters into three types: steering control, braking control and drive control, assign initial values ​​to each control parameter, and arrange them in a vectorized manner according to the set order to form a set of parameter vectors. S52. A pseudo-random number generator is used to determine the perturbation target parameters in the parameter vector set, and the SASP algorithm is used to assign a perturbation step size and perturbation direction to each perturbation target parameter to construct a perturbation index sequence, specifically including: In the parameter vector set, determine the index number corresponding to each control parameter, and according to the set disturbance probability threshold, call the pseudo-random number generator to generate a random value between zero and one for each index number; Each random value is compared with the disturbance probability threshold. If the random value is less than the disturbance probability threshold, the corresponding index number is marked as the disturbance target parameter, and the control type and initial value of the disturbance target parameter are recorded to form a disturbance candidate set. Based on the control type of each disturbance target parameter in the disturbance candidate set, set the disturbance direction selection rules; If the control type is steering control, then set the disturbance direction to the negative direction; If the control type is braking control, then the disturbance direction is set to the positive direction; If the control type is drive control, the disturbance direction is determined by the sign of the difference between the disturbance target parameter and the preset upper limit of the control threshold. If the difference is positive, it is set to the negative direction; if the difference is negative, it is set to the positive direction. Based on the current value, control type, and disturbance direction of each disturbance target parameter, and in conjunction with the set disturbance amplitude adjustment strategy, the SASP algorithm is used to calculate the corresponding disturbance step size, and the disturbance step size is assigned a positive or negative sign according to the disturbance direction. All the index numbers, perturbation step size and perturbation direction of the perturbation target parameters are combined into a perturbation index sequence according to the order in the parameter vector set; S53. Perform a time-step disturbance update operation on the parameter vector set according to the disturbance index sequence, and map the disturbance result to a disturbance control parameter set; S54. Load the disturbance control parameter set into the digital twin model, perform a running simulation operation under the condition that the fused state sequence remains unchanged, recalculate the steering power, braking power and driving power at each time step, and generate the corresponding disturbance energy consumption sequence.

[0015] Optionally, S6 specifically includes: S61. Perform time alignment processing on the disturbance energy consumption sequence and the total energy consumption sequence, and perform difference calculation to generate an energy consumption difference sequence. S62. Perform time-step squared sum addition on the energy consumption difference sequence, normalize the result according to the number of time steps, generate a mean square error sequence, and calculate the comprehensive energy consumption difference value based on the mean square error sequence. S63. Construct a loss function based on the comprehensive energy consumption difference value and parameter vector set; S64. Under the constraint of the loss function, the PPO algorithm is used to perform policy evaluation and update operations on the control parameter set, specifically including: The dominance estimation calculation is performed on the strategy probability distribution corresponding to the disturbance control parameter set to generate a dominance value sequence; The shearing probability ratio is constructed based on the dominance value sequence, and an interval restriction operation is performed on the shearing probability ratio to form a policy update factor. The control parameter set is then updated based on the policy update factor. S65. Reload the updated control parameter set into the digital twin model, recalculate the perturbation energy consumption sequence while keeping the fused state sequence consistent, and iteratively execute the gradient update operation until the loss function reaches the preset convergence threshold to obtain the converged control parameter set.

[0016] According to an embodiment of the present invention, a drive-by-wire chassis energy consumption simulation system based on digital twin includes: The data acquisition module is used to build a digital twin model of the drive-by-wire chassis, collect and preprocess the driving-by-wire chassis's operating data, and construct a status data sequence. The feature fusion module is used to perform feature extraction and fusion operations on the state data sequence using neural control differential equations, record the hidden states in continuous time, and obtain the fused state sequence. The simulation module is used to perform simulations in the digital twin model using Fourier neural operators based on the fused state sequence, and to calculate the steering, braking and driving power corresponding to each time step to form a power sequence. The energy consumption calculation module is used to predict the changing trend of the power sequence using a long short-term memory network, and at the same time, it performs time integration operation on the power sequence through Simpson integration, performs error compensation on the integration result based on the prediction result, and generates the total energy consumption sequence. The disturbance processing module is used to initialize the control parameter set, perform disturbance processing on the control parameter set using the SASP algorithm, and calculate the disturbance energy consumption sequence in the digital twin model based on the disturbance results. The comparative analysis module is used to compare and analyze the disturbance energy consumption sequence with the total energy consumption sequence, construct a loss function based on the analysis results, and use the PPO algorithm to iteratively update the control parameter set until the loss function converges. The error update module is used to load the updated control parameter set into the digital twin model to regenerate the total energy consumption sequence, calculate the error value between the actual energy consumption and the actual energy consumption, and update the digital twin model based on the error value.

[0017] The beneficial effects of this invention are: First, this invention overcomes the problem of weak ability to capture time dynamic features by traditional static sampling methods by introducing neural control differential equations to perform continuous-time modeling and hidden state fusion of the state data sequence of the drive-by-wire chassis. This achieves high-precision modeling of the control behavior response process and improves the adaptability and expressiveness of the simulation model to different working conditions.

[0018] Secondly, Fourier neural operators are used to model the fused state sequence in the frequency domain, and combined with long short-term memory networks and Simpson integral methods, energy consumption trends are predicted and error compensation is performed, which effectively improves the temporal consistency of energy consumption simulation and the numerical stability of energy estimation, making energy consumption calculation more accurate and engineering-ready.

[0019] Finally, by constructing a control parameter perturbation mechanism based on the SASP algorithm and combining it with the PPO reinforcement learning algorithm to perform gradient optimization and convergence determination of the control strategy, the self-evolutionary update of the control parameter set and the self-correction of the simulation model error were realized. This significantly enhanced the model's adaptive capability and energy consumption optimization capability in complex control scenarios, thereby improving the overall energy efficiency of the drive-by-wire chassis in multi-task dynamic working conditions. Attached Figure Description

[0020] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a method for simulating the energy consumption of a drive-by-wire chassis based on digital twins, as proposed in this invention. Figure 2 This is a flowchart of the disturbance control and reinforcement learning process for a digital twin-based energy consumption simulation method for a drive-by-wire chassis proposed in this invention. Figure 3 This is a module structure diagram of a drive-by-wire chassis energy consumption simulation system based on digital twin proposed in this invention. Detailed Implementation

[0021] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0022] refer to Figures 1-2 A method for simulating energy consumption of a drive-by-wire chassis based on digital twins includes the following steps: In this embodiment S1. Construct a digital twin model of the drive-by-wire chassis, collect and preprocess the driving-by-wire chassis's operating data, and construct a state data sequence; S2. Perform feature extraction and fusion operations on the state data sequence using neural control differential equations, record the hidden states in continuous time, and obtain the fused state sequence. S3. In the digital twin model, Fourier neural operators are used to perform simulation based on the fused state sequence to calculate the steering, braking and driving power corresponding to each time step, forming a power sequence. S4. A long short-term memory network is used to predict the trend of power sequence changes. At the same time, the Simpson integral is used to perform time integration operation on the power sequence. The error compensation is performed on the integration result based on the prediction result to generate the total energy consumption sequence. S5. Initialize the control parameter set, use the SASP algorithm to perform perturbation processing on the control parameter set, and calculate the perturbation energy consumption sequence in the digital twin model based on the perturbation results; S6. Compare and analyze the disturbance energy consumption sequence with the total energy consumption sequence, construct a loss function based on the analysis results, and use the PPO algorithm to iteratively update the control parameter set until the loss function converges. S7. Load the updated control parameter set into the digital twin model to regenerate the total energy consumption sequence, calculate the error value between the actual energy consumption and the actual energy consumption, and update the digital twin model based on the error value.

[0023] In this embodiment, the digital twin model represents a simulation model that virtually maps the physical structure, control behavior, and energy consumption response process of the drive-by-wire chassis. The operating data includes sensor sampling values ​​of steering, braking, driving, and vehicle body status, such as steering angle, brake pedal displacement, drive motor speed, vehicle acceleration, vehicle speed, motor current, and motor voltage. The preprocessing includes time alignment, missing data completion, Z-Score normalization, noise filtering, and structured stitching operations. The control parameter set represents the set of parameters used to adjust the control behavior of the drive-by-wire chassis, which includes multiple numerical parameters that affect steering, braking, and driving behavior. The loss function represents a functional expression used to evaluate the degree of difference between the disturbance energy consumption sequence and the target energy consumption sequence.

[0024] In this embodiment, the construction and operation process of the digital twin model specifically includes: Based on the structural design information and control logic specifications of the drive-by-wire chassis, steering, braking and drive control rules are constructed, and the mapping relationship between control input, structural response and power output is defined to form control behavior units; Based on sensor deployment information and sampling rules, a correspondence between physical parameters and sampling data is established, and a state mapping matrix is ​​constructed. Energy consumption calculation logic is embedded in the control behavior unit, and Fourier neural operators are used to set power calculation paths for steering response value, braking response value and driving response value respectively to construct an energy consumption response mechanism. The state mapping matrix and energy consumption response mechanism are integrated, and the parameter set for subsequent perturbation optimization and error update is initialized to complete the construction of the digital twin model; The fused state sequence is received in the digital twin model. The Fourier neural operator is called to perform multi-level frequency domain transformation and nonlinear activation operation on the fused state sequence to generate steering power, braking power and driving power at each time step and splice them into a multi-dimensional power vector to construct a power sequence.

[0025] In this embodiment, S2 specifically includes: S21. Map the state data sequence to a continuous time axis according to the time step order, construct a driving path for the state data of each time step, and establish a dynamic equation structure based on neural control differential equations. S22. Perform hidden state update operation on the dynamic equation structure through vector field, calculate the change value of hidden state between each adjacent time step, and generate hidden state trajectory. S23. Perform multi-layer GELU function transformation and feature fusion operation on the hidden state trajectory to splice and reconstruct the hidden states in different time intervals to form a fused state representation that maintains temporal continuity. S24. Discretize the fusion state representation according to the original time sequence to generate a fusion state sequence corresponding to the state data sequence.

[0026] In this embodiment, S21 specifically includes: S211. Construct a continuous time axis based on the state data sequence and the preset data acquisition frequency, wherein the data acquisition frequency represents the preset frequency of acquiring the drive-by-wire chassis operation data through the sensor, and each time step in the continuous time axis corresponds to a set of state data. S212. Based on the continuous time axis, perform time-domain interpolation on the state data sequence using the cubic spline interpolation method to construct the interpolation function, specifically including: Extract the time index and corresponding state data of each time step in the state data sequence to construct an initial sample set; Arrange the time indices in the initial sample set in order to determine the interpolation interval between adjacent time steps; A cubic spline interpolation function is constructed for each interpolation interval according to preset rules, and conditions for continuity of function values, continuity of the first derivative, and continuity of the second derivative are set. All coefficients in the cubic spline interpolation function are solved based on three types of continuity conditions, and all the solution results are concatenated to form the interpolation function. S213. Based on the neural control differential equation, construct a dynamic equation structure with the interpolation function as the control variable. The dynamic equation structure is used to define the differential evolution rules of the hidden state in continuous time.

[0027] In this embodiment, the process of constructing the dynamic equation structure specifically includes: Extract the state data of the starting time step from the state data sequence, and generate the initial hidden state through a preset encoding mapping function; Change rules are constructed based on the initial hidden state and the interpolation function values ​​at the corresponding time steps. First-order differential expressions are generated based on the change rules as the dynamic equation structure. The change rules are used to determine the change trend of the state data between adjacent time steps.

[0028] In this embodiment, the process of generating the hidden state trajectory specifically includes: According to the time order in the continuous time axis, the value of the interpolation function at each time step is obtained sequentially and concatenated with the current hidden state; Based on the change rules, the change in the splicing result is calculated at each time step using the Runge-Kutta method, and an update operation is performed on the current hidden state according to the change to obtain the hidden state of the next time step, specifically including: The slope of the current splicing result is calculated based on the change rules, and the intermediate state is estimated using the slope calculation result. The slope of the intermediate state is then calculated based on the change rules. Iteratively perform state estimation and slope calculation operations until the fourth slope is obtained; The four slopes are weighted and combined according to a preset ratio to obtain the change, and then added to the current hidden state to obtain the hidden state of the next time step. Arrange the hidden states of all time steps in chronological order to generate hidden state trajectories corresponding to continuous time axes.

[0029] In this embodiment, S3 specifically includes: S31. Receive the fused state sequence in the digital twin model, construct a multidimensional state tensor based on the fused state sequence in chronological order, and perform frequency domain mapping operation according to the preset mapping rules to generate frequency domain feature representation. S32. The frequency domain feature representation is subjected to multi-layer frequency domain convolution and frequency weight update operations using Fourier neural operators. The convolution results of each layer are then weighted and superimposed according to a preset weight set to form the frequency domain response features, specifically including: The entire frequency domain feature representation is divided into multiple subsets according to a preset frequency range, and a set of complex convolution kernels of a fixed size is configured for each subset; In each subset, the corresponding complex convolution kernel is used to perform complex convolution operation on the frequency domain feature representation within the subset to extract the real and imaginary features and construct a set of frequency responses. Based on the set frequency amplitude adjustment rules, frequency weighting parameters are configured for each frequency response set. Multiplication and scaling operations are performed on the real and imaginary features respectively according to the frequency weighting parameters to form a weighted frequency set. All weighted frequency sets are concatenated according to the frequency subset partitioning order and then aligned in dimensions. A linear weighted fusion operation is performed on the alignment results based on a preset set of weight coefficients to obtain the frequency domain response features; S33. According to the preset discrete Fourier inverse transform rules, perform inverse frequency domain transform operation on the frequency domain response features to convert the frequency domain response features into time domain response features, and perform Swish function activation operation to generate state response sequence. S34. In the digital twin model, the state response sequence of each time step is mapped to steering response value, braking response value and driving response value respectively. S35. Perform power calculation operations on the steering response value, braking response value and driving response value respectively to generate steering power, braking power and driving power at each time step; S36. The power calculation results are spliced ​​and rearranged in chronological order to construct a multidimensional power vector, and all multidimensional power vectors are summarized to form a power sequence.

[0030] In this embodiment, S4 specifically includes: S41. Convert the power sequence into a multi-dimensional time series tensor in chronological order, and extract the steering power, braking power and driving power of each time step as channel features. S42. Input the multidimensional temporal tensor into the long short-term memory network, and perform forward and backward propagation operations respectively. Extract the bidirectional hidden states at each time step and splice them to generate a power trend sequence. S43. Using the Simpson integral method, perform time integration on the power sequence to calculate steering energy consumption, braking energy consumption, and driving energy consumption respectively, forming a preliminary energy consumption sequence, specifically including: Take every three adjacent time steps in the power sequence as an integration unit, and extract the steering power, braking power and driving power corresponding to the start time step, middle time step and end time step in the integration unit; A fixed weight value is assigned to each of the three time steps in each integration unit. The steering power, braking power and driving power are weighted and summed. The product is calculated by combining the time step intervals to generate the steering energy value, braking energy value and driving energy value corresponding to the integration unit. The three types of energy values ​​obtained from all the integration units are summed up to form steering energy consumption, braking energy consumption and driving energy consumption, and then combined in a preset order to form a preliminary energy consumption sequence. S44. Calculate the difference between the power trend sequence and the power sequence, extract the prediction error value at each time step, construct a set of compensation factors based on the prediction error value, and use the set of compensation factors to perform a weighted correction operation on the preliminary energy consumption sequence step by step to generate a compensated energy consumption sequence. S45. Perform a channel-by-channel accumulation operation on the compensated energy consumption sequence according to the channel characteristics, concatenate the accumulation results into an energy consumption vector set in chronological order, and summarize all energy consumption vector sets to generate the total energy consumption sequence.

[0031] In this embodiment, S42 specifically includes: S421. Input the multidimensional temporal tensor into the feedforward structure of the long short-term memory network in chronological order, and perform the following operations sequentially at each time step: Based on the multidimensional temporal tensor of the current time step and the hidden state of the previous time step, calculate the activation values ​​of the forget gate, input gate and output gate. In the first time step, the hidden state of the previous time step is used as the preset initialization vector. The memory state of the previous time step is retained based on the forget gate activation value. In the first time step, the memory state of the previous time step is based on the preset initial zero vector. A weighted mapping operation is performed on the current multidimensional temporal tensor based on the input gate activation value to generate candidate memory content for the current time step; The candidate memory content is superimposed with the retained memory state to generate the memory state at the current time step; Perform an element-wise multiplication operation between the output gate activation value and the memory state of the current time step to generate the hidden state of the current time step; S422. Input the multidimensional temporal tensor into the backward structure of the long short-term memory network in reverse order of time steps, and perform the same operation as the forward structure in each time step to obtain the hidden state on the backward propagation path. S423. The hidden states generated by the forward and backward structures are spliced ​​together according to time steps, and a linear mapping operation is performed on the splicing result to generate a power trend sequence.

[0032] In this embodiment, S5 specifically includes: S51. Construct a set of control parameters based on the preset control dimensions, group all control parameters into three types: steering control, braking control and drive control, assign initial values ​​to each control parameter, and arrange them in a vectorized manner according to the set order to form a set of parameter vectors. S52. A pseudo-random number generator is used to determine the perturbation target parameters in the parameter vector set, and the SASP algorithm is used to assign a perturbation step size and perturbation direction to each perturbation target parameter to construct a perturbation index sequence, specifically including: In the parameter vector set, determine the index number corresponding to each control parameter, and according to the set disturbance probability threshold, call the pseudo-random number generator to generate a random value between zero and one for each index number; Each random value is compared with the disturbance probability threshold. If the random value is less than the disturbance probability threshold, the corresponding index number is marked as the disturbance target parameter, and the control type and initial value of the disturbance target parameter are recorded to form a disturbance candidate set. Based on the control type of each disturbance target parameter in the disturbance candidate set, set the disturbance direction selection rules; If the control type is steering control, then set the disturbance direction to the negative direction; If the control type is braking control, then the disturbance direction is set to the positive direction; If the control type is drive control, the disturbance direction is determined by the sign of the difference between the disturbance target parameter and the preset upper limit of the control threshold. If the difference is positive, it is set to the negative direction; if the difference is negative, it is set to the positive direction. Based on the current value, control type, and disturbance direction of each disturbance target parameter, and in conjunction with the set disturbance amplitude adjustment strategy, the SASP algorithm is used to calculate the corresponding disturbance step size, and the disturbance step size is assigned a positive or negative sign according to the disturbance direction. All the index numbers, perturbation step size and perturbation direction of the perturbation target parameters are combined into a perturbation index sequence according to the order in the parameter vector set; S53. Perform a time-step disturbance update operation on the parameter vector set according to the disturbance index sequence, and map the disturbance result to a disturbance control parameter set; S54. Load the disturbance control parameter set into the digital twin model, perform a running simulation operation under the condition that the fused state sequence remains unchanged, recalculate the steering power, braking power and driving power at each time step, and generate the corresponding disturbance energy consumption sequence.

[0033] In this embodiment, S6 specifically includes: S61. Perform time alignment processing on the disturbance energy consumption sequence and the total energy consumption sequence, and perform difference calculation to generate an energy consumption difference sequence. S62. Perform time-step squared sum addition on the energy consumption difference sequence, normalize the result according to the number of time steps, generate a mean square error sequence, and calculate the comprehensive energy consumption difference value based on the mean square error sequence. S63. Construct a loss function based on the comprehensive energy consumption difference value and parameter vector set; S64. Under the constraint of the loss function, the PPO algorithm is used to perform policy evaluation and update operations on the control parameter set, specifically including: The dominance estimation calculation is performed on the strategy probability distribution corresponding to the disturbance control parameter set to generate a dominance value sequence; The shearing probability ratio is constructed based on the dominance value sequence, and an interval restriction operation is performed on the shearing probability ratio to form a policy update factor. The control parameter set is then updated based on the policy update factor. S65. Reload the updated control parameter set into the digital twin model, recalculate the perturbation energy consumption sequence while keeping the fused state sequence consistent, and iteratively execute the gradient update operation until the loss function reaches the preset convergence threshold to obtain the converged control parameter set.

[0034] refer to Figure 3 A digital twin-based energy consumption simulation system for drive-by-wire chassis includes: The data acquisition module is used to build a digital twin model of the drive-by-wire chassis, collect and preprocess the driving-by-wire chassis's operating data, and construct a status data sequence. The feature fusion module is used to perform feature extraction and fusion operations on the state data sequence using neural control differential equations, record the hidden states in continuous time, and obtain the fused state sequence. The simulation module is used to perform simulations in the digital twin model using Fourier neural operators based on the fused state sequence, and to calculate the steering, braking and driving power corresponding to each time step to form a power sequence. The energy consumption calculation module is used to predict the changing trend of the power sequence using a long short-term memory network, and at the same time, it performs time integration operation on the power sequence through Simpson integration, performs error compensation on the integration result based on the prediction result, and generates the total energy consumption sequence. The disturbance processing module is used to initialize the control parameter set, perform disturbance processing on the control parameter set using the SASP algorithm, and calculate the disturbance energy consumption sequence in the digital twin model based on the disturbance results. The comparative analysis module is used to compare and analyze the disturbance energy consumption sequence with the total energy consumption sequence, construct a loss function based on the analysis results, and use the PPO algorithm to iteratively update the control parameter set until the loss function converges. The error update module is used to load the updated control parameter set into the digital twin model to regenerate the total energy consumption sequence, calculate the error value between the actual energy consumption and the actual energy consumption, and update the digital twin model based on the error value.

[0035] Example 1: To verify the feasibility of this invention in practice, it was applied to the energy management and control strategy optimization scenario of a certain intelligent chassis platform. This platform is equipped with a steer-by-wire system, a brake-by-wire system, and an electric drive system, possessing a typical multi-actuator coupling structure. In actual use, this chassis system suffers from problems such as inaccurate energy consumption prediction, reliance on experience for control parameter adjustment, and a lack of unified strategies for energy consumption optimization under different operating conditions, resulting in low operating efficiency and unstable energy consumption performance. Therefore, the energy consumption simulation method based on digital twins proposed in this invention was deployed for modeling analysis and strategy optimization verification.

[0036] First, a high-frequency sensor network is used to collect state data of the drive-by-wire chassis during actual operation. Data dimensions include: steering angle, steering wheel torque, brake pedal displacement, motor voltage, motor current, vehicle speed, and vehicle acceleration. Data is collected at a frequency of 100Hz per second for 120 seconds, yielding 12,000 time steps of high-dimensional state data. After collection, time alignment, missing value imputation, Z-score normalization, sliding window filtering, and structural stitching are performed sequentially to construct a state data sequence for subsequent modeling.

[0037] The preprocessed state data sequence is input into the neural control differential equation structure to complete continuous-time modeling and hidden state updates, resulting in a fused state sequence. This sequence preserves the evolution trend of the state data at small time scales and enhances the ability to model the dynamic behavior of the system. Based on this, Fourier neural operators are used to perform frequency domain modeling on the fused state sequence. Through multi-layer frequency convolution, weight updates, and activation function processing, a high-resolution state response sequence is generated and mapped to steering, braking, and driving power at each time step, forming a power sequence.

[0038] A long short-term memory (LSTM) network is used to predict the time-series trend of power sequences, capturing the evolution path of power. The Simpson integral method is combined to integrate the power sequences over time, and an error compensation mechanism is constructed to correct prediction errors, ultimately generating a basic total energy consumption sequence as the energy consumption performance under the current strategy. Next, a control parameter set is initialized, and the SASP algorithm is used to perform disturbance operations, generating multiple combinations of disturbance control parameters. Under the condition that the fused state sequence remains unchanged, each set of disturbance parameters is loaded into a digital twin model for simulation to obtain the corresponding disturbance energy consumption sequence.

[0039] By comparing the perturbation energy consumption sequence with the total energy consumption sequence, a mean squared error loss function is constructed, and the PPO algorithm is used to perform gradient optimization and convergence detection in the policy space. The control parameter set is updated after each iteration until the loss function converges to a preset threshold, ultimately obtaining the optimal control parameters. This parameter set is loaded into the digital twin model to regenerate the optimized total energy consumption sequence, and compared with historical energy consumption data from the real vehicle at each time step to calculate the error value. Finally, based on the error feedback, the dynamic mapping and energy consumption response mechanism in the digital twin model are corrected to achieve closed-loop model updates.

[0040] In the experimental setup, several typical driving scenarios were constructed: urban congestion driving (frequent start-stop), medium-speed turning, slow-moving downhill driving, hill start, cruise control, navigating complex intersections, continuous S-curves, high-speed braking obstacle avoidance, low-speed precise parking, and dynamic U-turns. For each scenario, the energy consumption before and after optimization, the adjustment range of strategy parameters, and the average prediction error between the simulation model and real vehicle data were recorded and compared with a conventional baseline model, as shown in the table below: Table 1. Comparison of Energy Consumption Simulation and Optimization Results for Controller-by-Wire Chassis under Multiple Operating Conditions

[0041] As shown in Table 1, the method of this invention achieved stable energy consumption optimization in ten typical operating conditions, with an average reduction of 8.62% in total energy consumption. The energy-saving effect was most significant in continuous S-curves, urban congestion driving, and high-speed braking obstacle avoidance conditions, reaching a maximum of 9.39%. The average prediction error decreased from 9.91% before optimization to 2.99% after optimization, indicating good error convergence. This demonstrates that the digital twin model, after optimization by PPO reinforcement learning, possesses high prediction accuracy and control adaptability.

[0042] Regarding strategy parameters, based on the control type, several key parameters are involved, including steering return delay time, braking response threshold, drive power limit, energy recovery entry point, and control dead zone width. After parameter adjustment, the model converges rapidly, typically reaching a stable value within 10-16 optimization rounds. Overall performance demonstrates that this invention not only possesses accurate energy consumption prediction capabilities but also enables adaptive optimization of control parameters, improving the energy utilization efficiency and system control intelligence of the drive-by-wire chassis under different dynamic operating conditions, exhibiting significant engineering application value and promotion potential.

[0043] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for simulating energy consumption of a drive-by-wire chassis based on digital twins, characterized in that, Includes the following steps: S1. Construct a digital twin model of the drive-by-wire chassis, collect and preprocess the driving-by-wire chassis's operating data, and construct a state data sequence; S2. The neural control differential equation is used to perform feature extraction and fusion operations on the state data sequence, and the hidden state in continuous time is recorded to obtain the fused state sequence. S3. In the digital twin model, Fourier neural operators are used to perform simulation based on the fused state sequence to calculate the steering, braking and driving power corresponding to each time step, forming a power sequence. S4. A long short-term memory network is used to predict the trend of power sequence changes. At the same time, the Simpson integral is used to perform time integration operation on the power sequence. The error compensation is performed on the integration result based on the prediction result to generate the total energy consumption sequence. S5. Initialize the control parameter set, use the SASP algorithm to perform perturbation processing on the control parameter set, and calculate the perturbation energy consumption sequence in the digital twin model based on the perturbation results; S6. Compare and analyze the disturbance energy consumption sequence with the total energy consumption sequence, construct a loss function based on the analysis results, and use the PPO algorithm to iteratively update the control parameter set until the loss function converges. S7. Load the updated control parameter set into the digital twin model to regenerate the total energy consumption sequence, calculate the error value between the actual energy consumption and the actual energy consumption, and update the digital twin model based on the error value.

2. The method for simulating energy consumption of a drive-by-wire chassis based on digital twins according to claim 1, characterized in that, The digital twin model represents a simulation model that virtually maps the physical structure, control behavior, and energy consumption response of the steer-by-wire chassis. The operational data includes sensor sampling values ​​for steering, braking, driving, and vehicle status, such as steering angle, brake pedal displacement, drive motor speed, vehicle acceleration, vehicle speed, motor current, and motor voltage. The preprocessing includes time alignment, missing data completion, Z-score normalization, noise filtering, and structured stitching operations. The control parameter set represents a set of parameters used to adjust the control behavior of the steer-by-wire chassis, including multiple numerical parameters affecting steering, braking, and driving behavior. The loss function represents a functional expression used to evaluate the degree of difference between the disturbance energy consumption sequence and the target energy consumption sequence.

3. The method for simulating energy consumption of a drive-by-wire chassis based on digital twins according to claim 1, characterized in that, The construction and operation process of the digital twin model specifically includes: Based on the structural design information and control logic specifications of the drive-by-wire chassis, steering, braking and drive control rules are constructed, and the mapping relationship between control input, structural response and power output is defined to form control behavior units; Based on sensor deployment information and sampling rules, a correspondence between physical parameters and sampling data is established, and a state mapping matrix is ​​constructed. Energy consumption calculation logic is embedded in the control behavior unit, and Fourier neural operators are used to set power calculation paths for steering response value, braking response value and driving response value respectively to construct an energy consumption response mechanism. The state mapping matrix and energy consumption response mechanism are integrated, and the parameter set for subsequent perturbation optimization and error update is initialized to complete the construction of the digital twin model; The fused state sequence is received in the digital twin model. The Fourier neural operator is called to perform multi-level frequency domain transformation and nonlinear activation operation on the fused state sequence to generate steering power, braking power and driving power at each time step and splice them into a multi-dimensional power vector to construct a power sequence.

4. The method for simulating energy consumption of a drive-by-wire chassis based on digital twins according to claim 1, characterized in that, S2 specifically includes: S21. Map the state data sequence to a continuous time axis according to the time step order, construct a driving path for the state data of each time step, and establish a dynamic equation structure based on neural control differential equations. S22. Perform hidden state update operation on the dynamic equation structure through vector field, calculate the change value of hidden state between each adjacent time step, and generate hidden state trajectory. S23. Perform multi-layer GELU function transformation and feature fusion operation on the hidden state trajectory to splice and reconstruct the hidden states in different time intervals to form a fused state representation that maintains temporal continuity. S24. Discretize the fusion state representation according to the original time sequence to generate a fusion state sequence corresponding to the state data sequence.

5. The method for simulating energy consumption of a drive-by-wire chassis based on digital twins according to claim 4, characterized in that, S21 specifically includes: S211. Construct a continuous time axis based on the state data sequence and the preset data acquisition frequency. The data acquisition frequency represents the preset frequency of acquiring the driving data of the drive-by-wire chassis through the sensor. Each time step in the continuous time axis corresponds to a set of state data. S212. Based on the continuous time axis, perform time-domain interpolation on the state data sequence using the cubic spline interpolation method, and construct the interpolation function, specifically including: Extract the time index and corresponding state data of each time step in the state data sequence to construct an initial sample set; Arrange the time indices in the initial sample set in order to determine the interpolation interval between adjacent time steps; A cubic spline interpolation function is constructed for each interpolation interval according to preset rules, and conditions for continuity of function values, continuity of the first derivative, and continuity of the second derivative are set. All coefficients in the cubic spline interpolation function are solved based on three types of continuity conditions, and all the solution results are concatenated to form the interpolation function. S213. Based on the neural control differential equation, a dynamic equation structure is constructed using the interpolation function as the control variable. The dynamic equation structure is used to define the differential evolution rules of the hidden state in continuous time.

6. The method for simulating energy consumption of a drive-by-wire chassis based on digital twins according to claim 1, characterized in that, S3 specifically includes: S31. Receive the fused state sequence in the digital twin model, construct a multidimensional state tensor based on the fused state sequence in chronological order, and perform frequency domain mapping operation according to the preset mapping rules to generate frequency domain feature representation. S32. The frequency domain feature representation is subjected to multi-layer frequency domain convolution and frequency weight update operations using Fourier neural operators. The convolution results of each layer are then weighted and superimposed according to a preset weight set to form the frequency domain response features, specifically including: The entire frequency domain feature representation is divided into multiple subsets according to a preset frequency range, and a set of complex convolution kernels of a fixed size is configured for each subset; In each subset, the corresponding complex convolution kernel is used to perform complex convolution operation on the frequency domain feature representation within the subset to extract the real and imaginary features and construct a frequency response set. Based on the set frequency amplitude adjustment rules, frequency weighting parameters are configured for each frequency response set. Multiplication and scaling operations are performed on the real and imaginary features respectively according to the frequency weighting parameters to form a weighted frequency set. The entire weighted frequency set is concatenated according to the frequency subset partitioning order and then aligned in dimensions. A linear weighted fusion operation is performed on the alignment results based on a preset set of weight coefficients to obtain the frequency domain response features; S33. According to the preset discrete Fourier inverse transform rules, perform inverse frequency domain transform operation on the frequency domain response features to convert the frequency domain response features into time domain response features, and perform Swish function activation operation to generate state response sequence. S34. In the digital twin model, the state response sequence of each time step is mapped to steering response value, braking response value and driving response value respectively. S35. Perform power calculation operations on the steering response value, braking response value and driving response value respectively to generate steering power, braking power and driving power at each time step; S36. The power calculation results are spliced ​​and rearranged in chronological order to construct a multidimensional power vector, and all multidimensional power vectors are summarized to form a power sequence.

7. The method for simulating energy consumption of a drive-by-wire chassis based on digital twins according to claim 1, characterized in that, S4 specifically includes: S41. Convert the power sequence into a multi-dimensional time series tensor in chronological order, and extract the steering power, braking power and driving power of each time step as channel features. S42. Input the multidimensional temporal tensor into the long short-term memory network, and perform forward and backward propagation operations respectively. Extract the bidirectional hidden states at each time step and splice them to generate a power trend sequence. S43. Using the Simpson integral method, perform time integration on the power sequence to calculate steering energy consumption, braking energy consumption, and driving energy consumption respectively, forming a preliminary energy consumption sequence, specifically including: Take every three adjacent time steps in the power sequence as an integration unit, and extract the steering power, braking power and driving power corresponding to the start time step, middle time step and end time step in the integration unit; A fixed weight value is assigned to each of the three time steps in each integration unit. The steering power, braking power and driving power are weighted and summed. The product is calculated by combining the time step intervals to generate the steering energy value, braking energy value and driving energy value corresponding to the integration unit. The three types of energy values ​​obtained from all the integration units are summed up to form steering energy consumption, braking energy consumption and driving energy consumption, and then combined in a preset order to form a preliminary energy consumption sequence. S44. Calculate the difference between the power trend sequence and the power sequence, extract the prediction error value at each time step, construct a set of compensation factors based on the prediction error value, and use the set of compensation factors to perform a weighted correction operation on the preliminary energy consumption sequence step by step to generate a compensated energy consumption sequence. S45. Perform a channel-by-channel accumulation operation on the compensated energy consumption sequence according to the channel characteristics, concatenate the accumulation results into an energy consumption vector set in chronological order, and summarize all energy consumption vector sets to generate the total energy consumption sequence.

8. The method for simulating energy consumption of a drive-by-wire chassis based on digital twins according to claim 1, characterized in that, S5 specifically includes: S51. Construct a set of control parameters based on the preset control dimensions, group all control parameters into three types: steering control, braking control and drive control, assign initial values ​​to each control parameter, and arrange them in a vectorized manner according to the set order to form a set of parameter vectors. S52. A pseudo-random number generator is used to determine the perturbation target parameters in the parameter vector set, and the SASP algorithm is used to assign a perturbation step size and perturbation direction to each perturbation target parameter to construct a perturbation index sequence, specifically including: In the parameter vector set, determine the index number corresponding to each control parameter, and according to the set disturbance probability threshold, call the pseudo-random number generator to generate a random value between zero and one for each index number. Each random value is compared with the disturbance probability threshold. If the random value is less than the disturbance probability threshold, the corresponding index number is marked as the disturbance target parameter, and the control type and initial value of the disturbance target parameter are recorded to form a disturbance candidate set. Based on the control type of each disturbance target parameter in the disturbance candidate set, set the disturbance direction selection rules; If the control type is steering control, then set the disturbance direction to the negative direction; If the control type is braking control, then the disturbance direction is set to the positive direction; If the control type is drive control, the disturbance direction is determined by the sign of the difference between the disturbance target parameter and the preset upper limit of the control threshold. If the difference is positive, it is set to the negative direction; if the difference is negative, it is set to the positive direction. Based on the current value, control type, and disturbance direction of each disturbance target parameter, and in conjunction with the set disturbance amplitude adjustment strategy, the SASP algorithm is used to calculate the corresponding disturbance step size, and the disturbance step size is assigned a positive or negative sign according to the disturbance direction. All the index numbers, perturbation step size and perturbation direction of the perturbation target parameters are combined into a perturbation index sequence according to the order in the parameter vector set; S53. Perform a time-step disturbance update operation on the parameter vector set according to the disturbance index sequence, and map the disturbance result to a disturbance control parameter set; S54. Load the disturbance control parameter set into the digital twin model, perform the running simulation operation under the condition that the fused state sequence remains unchanged, recalculate the steering power, braking power and driving power at each time step, and generate the corresponding disturbance energy consumption sequence.

9. The method for simulating energy consumption of a drive-by-wire chassis based on digital twins according to claim 1, characterized in that, S6 specifically includes: S61. Perform time alignment processing on the disturbance energy consumption sequence and the total energy consumption sequence, and perform difference calculation to generate an energy consumption difference sequence. S62. Perform time-step squared sum addition on the energy consumption difference sequence, normalize the result according to the number of time steps, generate a mean square error sequence, and calculate the comprehensive energy consumption difference value based on the mean square error sequence. S63. Construct a loss function based on the comprehensive energy consumption difference value and parameter vector set; S64. Under the constraint of the loss function, the PPO algorithm is used to perform policy evaluation and update operations on the control parameter set, specifically including: The dominance estimation calculation is performed on the strategy probability distribution corresponding to the disturbance control parameter set to generate a dominance value sequence; The shearing probability ratio is constructed based on the dominance value sequence, and an interval restriction operation is performed on the shearing probability ratio to form a policy update factor. The control parameter set is then updated based on the policy update factor. S65. Reload the updated control parameter set into the digital twin model, recalculate the perturbation energy consumption sequence while keeping the fused state sequence consistent, and iteratively execute the gradient update operation until the loss function reaches the preset convergence threshold to obtain the converged control parameter set.

10. A drive-by-wire chassis energy consumption simulation system based on digital twins, executing the drive-by-wire chassis energy consumption simulation method based on digital twins as described in any one of claims 1 to 9, characterized in that, include: The data acquisition module is used to build a digital twin model of the drive-by-wire chassis, collect and preprocess the driving-by-wire chassis's operating data, and construct a status data sequence. The feature fusion module is used to perform feature extraction and fusion operations on the state data sequence using neural control differential equations, record the hidden states in continuous time, and obtain the fused state sequence. The simulation module is used to perform simulations in the digital twin model using Fourier neural operators based on the fused state sequence, and to calculate the steering, braking and driving power corresponding to each time step to form a power sequence. The energy consumption calculation module is used to predict the changing trend of the power sequence using a long short-term memory network, and at the same time, it performs time integration operation on the power sequence through Simpson integration, performs error compensation on the integration result based on the prediction result, and generates the total energy consumption sequence. The disturbance processing module is used to initialize the control parameter set, perform disturbance processing on the control parameter set using the SASP algorithm, and calculate the disturbance energy consumption sequence in the digital twin model based on the disturbance results. The comparative analysis module is used to compare and analyze the disturbance energy consumption sequence with the total energy consumption sequence, construct a loss function based on the analysis results, and use the PPO algorithm to iteratively update the control parameter set until the loss function converges. The error update module is used to load the updated control parameter set into the digital twin model to regenerate the total energy consumption sequence, calculate the error value between the actual energy consumption and the actual energy consumption, and update the digital twin model based on the error value.

Images