A method, system, device and medium for dynamic compensation control of thermal degradation of a turbomachine

Abstract

The application provides a kind of vortex machine heat recession dynamic compensation control method, system, equipment and medium, belong to automobile environmental protection detection equipment control field, including the construction includes temperature, temperature change rate and real-time torque coefficient 3-dimensional feature vector;3-dimensional feature vector is input heat recession prediction model, extracts local time sequence feature of heat recession, and the key temperature interval feature in local time sequence feature is weighted and enhanced;The feature sequence after weighting is input GRU layer, and the torque coefficient prediction value of future 1 control period is output;According to torque coefficient prediction value, calculate normalized prediction deviation Δe;Based on Δe, piecewise self-adaptive adjustment PID parameter is carried out, and heat recession dynamic correction term is calculated;Heat recession dynamic correction term combines speed deviation to generate final excitation current instruction;Excitation current instruction is sent to excitation current regulating unit, and vortex machine heat recession dynamic compensation is realized.The application significantly improves the test accuracy and stability of automobile environmental protection dynamometer.

Description

Technical Field

[0001] This invention belongs to the field of automotive environmental protection testing equipment control technology, specifically involving a dynamic compensation control method, system, equipment and medium for thermal decay of an eddy current engine. Background Technology

[0002] The automotive environmental dynamometer is a core device for testing vehicle environmental performance. Its core loading actuator is an eddy current generator, whose core function is to accurately simulate the vehicle's road resistance and provide a stable braking torque for the testing system. From a working principle perspective, the eddy current generator is a braking device based on the eddy current effect. By applying an excitation current to the stator to establish a magnetic field, an induced electromotive force is generated when the metal rotor cuts the magnetic field lines, thus forming eddy currents inside the rotor. These eddy currents are subjected to the Ampere force in the magnetic field, generating a braking torque opposite to the rotor's rotation direction, realizing the conversion of the system's kinetic energy into electrical energy. According to Joule's law, the eddy currents generate heat when flowing through the rotor's resistance, directly causing the rotor temperature to rise.

[0003] Under actual continuous braking conditions, eddy current engines have an inherent performance degradation problem (i.e., thermal decay). The core physical mechanism can be summarized as follows: the heat generated during braking increases the rotor temperature. The resistivity of the ferromagnetic material that makes up the rotor has a nonlinear positive correlation with temperature. According to Ohm's law, the increase in resistivity will directly lead to a decrease in eddy current, which in turn will cause a decrease in braking torque. At the same time, thermal decay has a time-cumulative dependence. The historical temperature changes of the rotor will directly affect the current torque output characteristics, which further exacerbates the instability of braking torque. Therefore, dynamic compensation control of thermal decay of eddy current engines is particularly important.

[0004] Currently, dynamic compensation technologies for eddy current engine thermal decay in the automotive environmental dynamometer field mainly include two aspects: traditional linear PID control and conventional compensation methods. Traditional linear PID control adjusts the gain based solely on speed deviation. When thermal decay occurs, the control parameters cannot adaptively adjust, leading to a steady-state error exceeding ±5%. Furthermore, severe accuracy drift occurs after prolonged continuous operation. Essentially, the core flaw of traditional PID control lies in the fact that the controller parameters (Kp, Ki, Kd) are only tuned under cold conditions. Once tuned, the parameters remain fixed, failing to adapt to the drastic changes in system dynamics under hot conditions. This ultimately results in overshoot, oscillation, or increased steady-state error during continuous braking, significantly degrading control performance. Conventional compensation methods involve experimentally obtaining optimal PID parameters at different temperature points, then establishing a linear relationship between temperature and parameters or constructing a lookup table. Although this method can improve the control effect to some extent, it still has fundamental problems that cannot be solved: First, it cannot accurately describe the dynamic process, relying only on the instantaneous temperature value for parameter adjustment, ignoring that heat accumulation is a time-series process closely related to historical power, and lacking the ability to predict the temperature rise trend; Second, the model is too simplified, and the use of linear mapping relationship makes it difficult to accurately characterize the strong nonlinear characteristics in the system, such as the square relationship between heat generation and current, and the nonlinear response between resistance and temperature; Third, the hysteresis is difficult to overcome. Due to the thermal inertia of the system, when a temperature change is detected, the actual thermal state of the rotor has already changed significantly, and the control behavior has an inherent response hysteresis.

[0005] In summary, existing technologies have not fundamentally overcome the core challenges of timing, nonlinearity, and hysteresis caused by thermal decay in automotive environmental dynamometer eddy current mechanisms. The key bottlenecks lie in two aspects: first, the lack of forward-looking prediction capabilities of the system's thermal dynamics, making it impossible to predict the development trend of thermal decay in advance; and second, the lack of a nonlinear adaptive mechanism that can accurately describe the complex changes in thermal decay, making it difficult to achieve real-time and accurate compensation for braking torque decay. Summary of the Invention

[0006] To address the aforementioned issues, this invention provides a dynamic compensation control method for thermal decay of eddy current engines. This method resolves the thermal decay problem caused by the fundamental contradiction between the "post-event response" of traditional control schemes and the "pre-event accumulation" thermal inertia of the system in automotive environmental dynamometers, thereby achieving proactive prediction and dynamic offsetting of thermal decay.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A dynamic compensation control method for thermal decay of an eddy current engine includes the following steps: The temperature at the outer end of the top excitation coil of the eddy current generator, the excitation current of the excitation circuit of the eddy current generator, the measured speed of the drum, and the measured torque of the drum are collected in real time. The temperature change rate and the real-time torque coefficient are calculated based on the temperature, excitation current, and measured torque of the drum. A 3D feature vector containing temperature, temperature change rate, and real-time torque coefficient with multiple consecutive steps is constructed. A 3D feature vector is input into a thermal decay prediction model, which is based on a 1D CNN framework. An attention module, a GRU layer, and a fully connected layer are sequentially introduced after the 1D CNN. The 1D CNN extracts local temporal features representing thermal decay from the 3D feature vector. The attention module then weights and enhances the key temperature range features in the local temporal features. The weighted feature sequence is input into the GRU layer to obtain a high-dimensional hidden state vector containing all temporal patterns of thermal decay. Based on the high-dimensional hidden state vector, the fully connected layer outputs the predicted torque coefficient value for the next control cycle. The normalized prediction deviation Δe is calculated based on the predicted torque coefficient; the PID parameters are adaptively adjusted piecewise based on Δe, and a dynamic thermal decay correction term is calculated based on the adjusted PID parameters; the dynamic thermal decay correction term is combined with the speed deviation to generate an excitation current command; the excitation current command is sent to the excitation current adjustment unit of the eddy current machine to realize dynamic compensation for thermal decay of the eddy current machine; the speed deviation is the difference between the target speed of the drum and the measured speed of the drum.

[0008] Preferably, the attention module introduces a physics-based attention mechanism. This physics-based attention mechanism introduces a parallel and interpretable physics rule weight generation path after the excitation stage of the standard SENet channel attention framework. The physics-based attention mechanism learns the attention weights of each channel through data and combines them with the weights generated by physics rules to generate final attention weights that conform to statistical laws and follow physical common sense.

[0009] Preferably, the step of weighting and enhancing key temperature range features in local temporal features using an attention module specifically includes the following steps: For the key temperature range features in the local time series features that reflect the key state of thermal decay, an enhancement weight of 0.6~0.8 is generated through an MLP network; For the non-critical noise and redundant information parts in the key temperature range features of local time series features, a basic weight of 0.2~0.4 is generated through an MLP network; The generated enhanced weights and basic weights are then weighted element-wise with the key temperature range features in the local temporal features output by the 1D CNN to obtain the weighted feature sequence.

[0010] Preferably, the step of calculating the temperature change rate and real-time torque coefficient based on temperature, excitation current, and measured drum torque, and constructing a 3D feature vector containing temperature, temperature change rate, and real-time torque coefficient in multiple consecutive steps, specifically includes the following steps: Calculate the temperature change rate dT / dt based on the temperature T at multiple step sizes; The online identification of the real-time torque coefficient K_T_actual is based on the excitation current I_eddy, the measured drum speed v_act, and the measured drum torque T_actual of the eddy current excitation circuit. The calculation formula is as follows: K_T_actual=T_actual / (I_eddy²×v_act); A 3D feature vector [T, dT / dt, K_T_actual] with multiple consecutive steps is constructed using T, dT / dt, and K_T_actual.

[0011] Preferably, the normalized prediction deviation Δe is calculated as follows: Δe = |K_T_pred - K_T_actual| / K_T_ref; where K_T_pred is the predicted torque coefficient, K_T_actual is the real-time torque coefficient identified online, and K_T_ref is the room temperature reference torque coefficient.

[0012] Preferably, the PID parameters are adaptively adjusted based on Δe segmentation, and a dynamic thermal decay correction term is calculated based on the adjusted PID parameters; the dynamic thermal decay correction term is combined with the speed deviation to generate an excitation current command, specifically including the following steps: Let the PID baseline parameters be Kp0, Ki0, and Kd0. The PID parameters are adjusted based on the normalized prediction deviation Δe: Kp = Kp0 × [1 + β × (Δe)], where β is taken in segments according to the prediction quality: Ki = Ki0 × [1 - 0.5 × exp (-Δe)]; Kd = Kd0 × [1 + 0.2 × (dΔe / dt)]; The dynamic correction term ΔK for thermal decay is calculated based on the predicted torque coefficient K_T_pred, the online identified real-time torque coefficient K_T_actual, and the room temperature reference torque coefficient K_T_ref. ΔK=(K_T_actual-K_T_pred) / K_T_ref; Based on the adjusted PID parameters and the thermal decay dynamic correction term ΔK, the final excitation current command I_out is generated: I_out=Kp×e+Ki×∫edt+Kd×de / dt+ΔK×I_base; Where e is the speed deviation, e = v_ref - v_act; v_ref is the target speed of the drum, and I_base is the room temperature reference excitation current.

[0013] Preferably, the method also includes real-time acquisition of adjusted temperature, excitation current, measured drum speed and measured drum torque, feeding the measured torque coefficient back to the thermal decay prediction model, updating the model parameters online, and feeding the measured drum speed v_act back to the nonlinear adaptive PID controller to calculate the new speed deviation e, entering the next control cycle, generating a new I_out, and performing a new round of eddy current excitation.

[0014] This invention also proposes a dynamic compensation control system for thermal decay of an eddy current engine, comprising: The data acquisition module is used to collect in real time the temperature at the outer end of the excitation coil at the top of the eddy current machine, the excitation current of the excitation circuit of the eddy current machine, the measured speed of the drum, and the measured torque of the drum. Based on the temperature, excitation current, and measured torque of the drum, it calculates the temperature change rate and the real-time torque coefficient; and constructs a 3D feature vector containing temperature, temperature change rate, and real-time torque coefficient in multiple consecutive steps. The prediction module is used to input a 3D feature vector into a thermal decay prediction model. The thermal decay prediction model is based on a 1DCNN framework, with an attention module, a GRU layer, and a fully connected layer introduced sequentially after the 1D CNN. The 1D CNN extracts local temporal features representing thermal decay from the 3D feature vector, and the attention module weights and enhances the key temperature range features in the local temporal features. The weighted feature sequence is input into the GRU layer to obtain a high-dimensional hidden state vector containing all the temporal patterns of thermal decay. Based on the high-dimensional hidden state vector, the fully connected layer outputs the predicted torque coefficient value for the next control cycle. The compensation module is used to calculate the normalized prediction deviation Δe based on the predicted torque coefficient; to adaptively adjust the PID parameters in segments based on Δe, and to calculate the dynamic correction term for thermal decay based on the adjusted PID parameters; the dynamic correction term for thermal decay is combined with the speed deviation to generate an excitation current command; the excitation current command is sent to the excitation current adjustment unit of the eddy current machine to realize dynamic compensation for thermal decay of the eddy current machine; the speed deviation is the difference between the target speed of the drum and the measured speed of the drum.

[0015] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement any of the steps in the eddy current engine thermal decay dynamic compensation control method.

[0016] The present invention also provides a computer-readable storage medium storing a computer program, which, when loaded by a processor, is capable of executing any of the steps in the eddy current engine thermal decay dynamic compensation control method.

[0017] The dynamic compensation control method for thermal decay of eddy current engines provided by this invention has the following beneficial effects: This invention, through the deep integration of time-series prediction and nonlinear adaptive PID compensation, completely solves the problems of timing, nonlinearity, and lag in thermal decay, achieving a fundamental shift from reactive post-event adjustment to proactive pre-event prediction and precise in-event compensation. Relying on a 1D CNN-GRU hybrid prediction model, this invention accurately captures the long-term cumulative pattern of thermal decay and outputs the predicted torque coefficient value for the next control cycle in advance, achieving forward prediction of thermal decay trends and fundamentally overcoming the response lag problem of traditional control. Based on the normalized deviation Δe of the predicted value, the PID parameters are adjusted in a segmented adaptive manner. The constructed nonlinear adaptive mechanism can accurately characterize the complex changes in thermal decay. Combined with a dedicated thermal decay dynamic correction term, it can provide real-time and precise dynamic compensation for braking torque attenuation, effectively offsetting the time-series cumulative effect of thermal decay and perfectly adapting to its nonlinear characteristics. Simultaneously, the deep integration of torque coefficient prediction and speed deviation control generates an excitation current command that can precisely adjust the eddy current excitation current of the eddy current machine, allowing the compensation action to match the thermal decay process in real time. Ultimately, this significantly improves the real-time performance and accuracy of thermal fade compensation, substantially reduces system steady-state error, effectively avoids accuracy drift issues after long-term operation, ensures that the eddy current machine always outputs stable braking torque under continuous braking conditions, and greatly improves the testing accuracy and operational stability of the automotive environmental dynamometer in scenarios with high torque accuracy requirements such as exhaust emission testing and energy consumption testing. This ensures that the braking loading performance of the dynamometer always conforms to the testing standards, providing accurate and reliable equipment support for vehicle environmental performance testing. Attached Figure Description

[0018] To more clearly illustrate the embodiments and design schemes of the present invention, the accompanying drawings required for this embodiment will be briefly described below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of the dynamic compensation control method for thermal decay of an eddy current engine according to Embodiment 1 of the present invention.

[0020] Figure 2 This is a flowchart of the data preprocessing process of the present invention; Figure 3 This is a schematic diagram of the thermal decay prediction model structure of the present invention. Figure 4 This is a schematic diagram of the nonlinear adaptive PID parameter tuning and dynamic compensation of the present invention; Figure 5 This is a framework diagram of an attention mechanism guided by physics knowledge. Detailed Implementation

[0021] To enable those skilled in the art to better understand and implement the technical solutions of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be construed as limiting the scope of protection of the present invention.

[0022] Existing dynamometer control technologies are all designed under ideal conditions that ignore thermal fade, without constructing time-series modeling and dynamic compensation logic for thermal fade. They completely disregard the impact of thermal fade on braking torque, resulting in decreased accuracy due to thermal fade, which remains a core pain point in the industry. Currently, no solution in the industry combines time-series predictive models with adaptive control, making it impossible to anticipate and dynamically offset thermal fade; the control response consistently lags behind the progression of thermal fade.

[0023] To fill the technological gap in this field, this invention proposes a dynamic compensation control method that integrates GRU and nonlinear adaptive PID. Through the integrated design of time-series prediction and real-time correction, it achieves precise suppression of thermal decay and improves the testing accuracy and stability of the dynamometer.

[0024] This invention provides a dynamic compensation control method for thermal decay of an eddy current engine, specifically a dynamic compensation control method for thermal decay of an eddy current engine used in automotive environmental protection dynamometers based on the fusion of GRU and nonlinear adaptive PID. (Refer to...) Figure 1 The proposed eddy current engine thermal decay dynamic compensation control method relies on the automotive environmental protection dynamometer control system. This control system includes a temperature sensing module, a status acquisition module, a GRU prediction unit, a nonlinear adaptive PID controller, an excitation current regulation unit, and a fault protection module. The GRU prediction unit deploys a thermal decay prediction model, the temperature sensing module specifically uses a temperature sensor, and the status acquisition module includes a current sensor, a photoelectric encoder, and a torque sensor. All modules communicate via a CAN bus / SPI interface, with a control cycle of 10ms.

[0025] Before implementing dynamic compensation control for thermal decay of eddy current engines, it is necessary to first configure system modules and install and calibrate sensors, as well as train and deploy a thermal decay prediction model.

[0026] The system module configuration and sensor installation calibration are as follows: 1. Module Configuration: The functions and key parameters of each core module are shown in Table 1. Table 1 Functions and Key Parameters of Each Core Module 2. Precise temperature sensing design: (1) Sensing part: Temperature sensing is performed at the end of the excitation coil of the eddy current machine. This part is the main area where electromagnetic loss is generated. Temperature change directly determines the torque coefficient attenuation range and can most accurately reflect the true state of thermal decay.

[0027] (2) Sensor selection and parameters are shown in Table 2: Table 2 Temperature Sensing Module Selection and Parameters (3) Sensor installation: Installation location: The bracket with reserved holes corresponding to the end of the eddy current machine excitation coil.

[0028] Fixing method: Use high-temperature resistant adhesive to firmly bond the sensor pressure ring to the end of the excitation coil, ensuring that the sensor is in close contact with the measured surface (without gaps).

[0029] Cable protection: High-temperature armored cables are secured with metal clips, keeping them away from rotating parts and high-temperature areas to prevent vibration from causing poor contact.

[0030] 3. Sensor calibration: After completing the wiring, calibrate the temperature sensor to ensure that the measurement error is ≤ ±2℃ for 25~200℃.

[0031] Based on the module configuration, the thermal decay prediction model is deployed and trained.

[0032] Reference Figure 3 The thermal decay prediction model proposed in this invention (i.e. Figure 3 The GRU prediction network in the model is a hybrid neural network architecture based on 1D CNN-GRU. This architecture is a cascaded multi-scale learning architecture that combines CNN, attention modules, and GRU layers, enabling multi-level joint learning that achieves rapid local changes due to thermal decay → feature enhancement and selection → long-term cumulative modeling. This architecture avoids the limitations of a single network structure and forms a complete feature learning path from microscopic transient response to macroscopic trend evolution.

[0033] This improved architecture is specifically designed for the multi-timescale characteristics of thermal decay in eddy current engines. A CNN layer captures the second-level thermal shock response, an attention module intelligently selects key features, and a GRU layer models the minute-level heat accumulation and dissipation process. The thermal decay prediction model is based on a 1D CNN framework, followed by a physics-guided attention module, a GRU layer, and a fully connected layer. Specifically, the 1D convolutional layer uses three 3x3 convolutional kernels and automatically extracts local temporal features of thermal decay through the ReLU activation function, forming a feature enhancement layer. This layer specifically captures rapid dynamic changes during thermal decay, such as the instantaneous correlation between temperature surges and current responses, significantly reducing the learning complexity of subsequent networks.

[0034] A physics-knowledge-guided attention mechanism is introduced after CNN feature extraction. Based on the prior knowledge that the temperature is in the critical thermal fade interval (50~150℃), a small neural network dynamically generates attention weights to perform adaptive weighting on the deep features extracted by the CNN. As Figure 5 shown in the figure, the physics-knowledge-guided attention mechanism introduces a parallel and interpretable physical rule weight generation path after the Excitation stage of the standard SENet channel attention framework; the physics-knowledge-guided attention mechanism learns the attention weights of each channel through data, and combines the weights generated by physical rules to generate final attention weights that conform to statistical laws and follow physical common sense. This mechanism enhances the focus on features of the critical thermal fade stage, while suppressing the interference of non-critical information.

[0035] The rule function is: Low temperature interval (T<25℃): The brake is in a cold state, and there is almost no thermal fade. The main influencing factor is the cold-state friction characteristic of the material itself. The physical weights of temperature, temperature change rate, and real-time torque coefficient can be set to (0.4, 0.3, 0.3).

[0036] Normal operation / low-temperature fade interval (25℃<T<50℃): The brake starts to work, the temperature rises slowly, the friction coefficient is relatively stable, and the thermal fade phenomenon is slight. All parameters change gently, and there is no need to particularly emphasize a certain channel. The physical weights of temperature, temperature change rate, and real-time torque coefficient can be set to (0.4, 0.3, 0.3).

[0037] Temperature rise / moderate fade interval (50℃<T<100℃): This is the critical interval where thermal fade starts to occur and develop. The organic components in the friction material may begin to decompose or generate a gas film, and the temperature change rate (temperature rise speed) is a leading indicator for predicting the fade trend. The faster the temperature rises, the greater the braking power or the poorer the heat dissipation, and the fade may accelerate. The physical weights of temperature, temperature change rate, and real-time torque coefficient can be set to (0.6, 0.8, 0.2).

[0038] High temperature / severe fade interval (100℃<T<150℃): The brake is already in a high-temperature plateau, the material may undergo phase transition or severe thermal decomposition, and the friction coefficient will stabilize at a relatively low level. At this time, the absolute temperature value is the main basis for judging whether the material has entered the high-temperature fade plateau. The importance of the temperature change rate decreases, because the system may be close to thermal equilibrium. The real-time torque coefficient directly reflects the current fade degree, and its importance increases. The physical weights of temperature, temperature change rate, and real-time torque coefficient can be set to (0.7, 0.2, 0.7).

[0039] Ultra-high temperature range (T>150℃): Irreversible damage such as material failure and ablation may occur. It is a dangerous working condition and requires special protection strategies. It can be set to (1.0, 0, 0).

[0040] The weights are merged as follows: s_final = µ × s_data + (1-µ) × s_physics; where s_final is the final feature vector after thermal decay fusion, µ is the fusion weight of data-driven features, s_data is the weight vector generated by the pure data-driven branch in the attention mechanism, and s_physics is the thermal decay feature vector driven by physical knowledge.

[0041] In the low-temperature range, the primary task is to establish a stable, reliable, and unbiased braking performance benchmark. The physical model, with a weight of 1 - µ = 0.65, dominates, ensuring output stability and avoiding misjudgments caused by small data fluctuations. The data (s_data) is assigned a weight of 0.35 (this refers to the weight vector generated by the purely data-driven branch in the attention mechanism), enabling it to slowly and finitely learn and compensate for the fixed deviation between the current specific brake and the ideal calibration model, providing the system with fine-tuning capabilities. The µ value sets the initial confidence state of the system. As the temperature transitions from the low-temperature range to the warming range, the µ value can be dynamically adjusted based on this, according to operating conditions and data consistency. For example, if the data is consistently stable and highly consistent with the model, µ can be slightly increased; if the data fluctuates significantly, µ can be maintained or slightly decreased.

[0042] Furthermore, a 16-neuron GRU layer is used to replace the traditional LSTM to capture the cumulative dependence of thermal decay. While maintaining the ability to learn long-term dependencies, the GRU layer reduces the number of model parameters by about 25% through a simplified gating mechanism (update gate, reset gate), significantly improving real-time inference efficiency. The GRU layer is specifically designed to learn the long-term cumulative effect and slow dynamic process of thermal decay.

[0043] Training a thermal decay prediction model involves the following steps: S1. Data Acquisition: Collect 50-80 sets of time-series data on the automotive environmental protection dynamometer, covering temperatures of 25-200℃ and loads of 10%-100% of rated torque. Each set includes core component temperature T, temperature change rate dT / dt, excitation current I_eddy, and measured torque coefficient K_T_actual.

[0044] S2. Preprocessing: The data is smoothed by using a moving average filter (window 5) to effectively suppress on-site noise interference; the data is mapped to [0,1] by feature normalization to eliminate the influence of different physical dimensions, and the preprocessed data is divided into a training set (80%) and a validation set (20%).

[0045] S3. Model Training: The deployed thermal decay prediction model is trained using the training set, and the training results are validated using the validation set.

[0046] Perform INT8 precision quantization on the trained model to ensure inference time ≤1.5ms and memory usage ≤30KB (adapted to embedded controller); burn the quantized model to the embedded controller.

[0047] Example 1 Based on the above hardware and software deployments, and with a control cycle of 10ms, this invention proposes a dynamic compensation control method for thermal decay of an eddy current engine used in automotive environmental protection dynamometers, based on the fusion of GRU and nonlinear adaptive PID. The specific implementation steps are as follows: Step 1: Perform data collection and feature construction.

[0048] Specifically, the temperature T (°C) at the outer end of the excitation coil at the top of the eddy current generator used in the automotive environmental protection dynamometer, the excitation current I_eddy (A) of the eddy current generator excitation circuit, the measured speed v_act (rad / s) of the drum, and the measured torque T_actual of the drum are collected in real time using detection devices such as temperature sensors, current sensors, photoelectric encoders, and torque sensors. Among them, the speed is collected by photoelectric encoders and the torque is collected by torque sensors.

[0049] Next, preprocessing is performed on the collected data, such as... Figure 2 As shown, it includes the following steps: (1) Calculate the temperature change rate dT / dt: First-order difference combined with moving average filtering is used to suppress measurement noise and smooth the temperature signal.

[0050] The specific calculation method for the temperature change rate is as follows: Based on a 10ms control cycle, the temperature at the end of the eddy current turbine excitation coil is collected in real time by a temperature sensor to obtain a continuous time-series temperature data sequence. T represents the current temperature, T₋1 represents the temperature of the previous control cycle, and the time interval Δt = 10 ms = 0.01 s.

[0051] Take the temperature difference between the current moment and the previous moment, divide it by the time interval, and obtain the original instantaneous rate of change: dT / dt original = (TT₋1) / Δt.

[0052] A moving average filter with a window size of 5 is used to calculate the average of the raw instantaneous rate of change over five consecutive control cycles, resulting in the final smoothed rate of temperature change.

[0053] (2) Calculation of real-time torque coefficient K_T_actual: Based on the eddy current braking principle, T_actual = K_T_actual × I_eddy² × v_actual. Combining the measured or calculated braking torque T_actual, the real-time torque coefficient K_T_actual can be identified online using the formula K_T_actual = T_actual / (I_eddy² × v_actual). The braking torque is the torque collected by the torque sensor.

[0054] (3) A 3-dimensional feature vector [T, dT / dt, K_T_actual] is constructed using T, dT / dt and K_T_actual, which is used for subsequent time series prediction.

[0055] Step 2: Perform 1D CNN-GRU temporal prediction, which includes the following steps: Step 21, Temporal Feature Extraction: Input the 3D feature vectors of 60 consecutive time steps (corresponding to a 300ms history window) into a 1D convolutional neural network (CNN). Use three convolutional kernels of size 3 to extract local temporal patterns and output a deep feature sequence with dimensions [58,3]. This sequence contains key feature information of local rapid dynamic changes during the thermal decay of the eddy current engine, such as the instantaneous correlation between temperature rise and excitation current and torque coefficient.

[0056] Step 22: Dynamically weighted features of the attention module containing a small neural network, specifically calculating attention weights based on a combination of physical mechanisms and data-driven approaches, including: (1) For the part of the deep feature sequence that reflects the key state of thermal decay (such as the rapid decay of torque coefficient and the sudden increase of temperature change rate in the range of 50~150℃), a small neural network is used to generate an enhancement weight of 0.6~0.8 to amplify the signal proportion of such key features.

[0057] For parts of deep feature sequences that belong to non-critical noise / redundant information (such as small fluctuations in the temperature stable range or feature changes that have no practical significance), a basic weight of 0.2 to 0.4 is generated through a small neural network to reduce the interference of this type of information.

[0058] The small neural network is the core component of the attention module in the 1D CNN-GRU hybrid architecture. It has no complex convolutional / recurrent layers, consisting only of a few hidden layers (1-2 layers) and neurons, with a much lower parameter count than 1D CNN and GRU, thus avoiding increased model inference time. In this embodiment, the small neural network can be a customized shallow fully connected network (MLP). Instead of generating fixed weights solely based on temperature ranges, the small neural network comprehensively integrates the deep feature sequences of the 1D CNN, the current temperature value, and historical temperature trends to calculate attention weights. Through the attention weights generated by the small neural network, the 1D CNN-GRU layer can accurately focus on key features of thermal decay, effectively suppress noise, improve the prediction accuracy of the torque coefficient, and ultimately provide reliable look-ahead data for the subsequent dynamic compensation of the nonlinear adaptive PID.

[0059] (2) The generated attention weights are weighted element-wise with the deep feature sequence output by the 1D CNN to obtain the weighted feature sequence. This sequence retains and enhances only the core features related to thermal decay and removes redundant noise.

[0060] Step 23, GRU time series prediction: The weighted feature sequence is input into a lightweight GRU layer (16 neurons). The GRU layer captures the long-term cumulative dependence and slow dynamic process of torque coefficient change during the thermal decay of the eddy current engine, and obtains a high-dimensional hidden state vector containing all the time series laws of thermal decay. The state vectors are integrated with features and then input into a fully connected layer. Finally, the torque coefficient prediction value K_T_pred for the next control cycle is output, realizing the look-ahead prediction of thermal decay.

[0061] Meanwhile, the prediction quality is evaluated by calculating the normalized prediction deviation Δe. The normalized prediction deviation Δe = |K_T_pred - K_T_actual| / K_T_ref, where K_T_ref is the room temperature (25℃) reference torque coefficient, and Δe will be used for subsequent adaptive adjustment of the controller.

[0062] This invention automatically learns the local temporal patterns of thermal decay using 1D CNN, strengthens key features by combining a physics-guided attention mechanism, and captures long-term dependencies using a GRU layer, forming a multi-level prediction architecture from local feature extraction to global temporal modeling, which significantly improves the prediction accuracy of thermal decay dynamics.

[0063] Step 3: Dynamically adjust the PID parameters and calculate the final excitation current command to achieve dynamic compensation for thermal decay.

[0064] Reference Figure 4Step 3, based on the GRU prediction results and prediction deviation data, dynamically adjusts the PID parameters through an innovative nonlinear adaptive mechanism and generates the final excitation current command to achieve precise compensation for thermal decay. The specific adjustment method is as follows: (1) Segmented adaptive adjustment of self-assessment of predicted quality Let the PID reference parameters be Kp0, Ki0, and Kd0 (obtained through system identification under cold conditions), and perform intelligent adjustment based on the normalized prediction deviation Δe: Proportional gain segmented adaptive: Kp = Kp0 × [1 + β × (Δe)], where β is segmented according to the prediction quality: when Δe < 5%, β = 0.3 (high model confidence, strong adjustment); when Δe ≥ 5%, β = 0.1 (low model confidence, weak adjustment). This invention introduces a segmented adaptive mechanism based on prediction quality for the first time, enabling the controller to intelligently adjust the response intensity according to its own prediction accuracy.

[0065] Exponential decay integral anti-saturation: Ki = Ki0 × [1 - 0.5 × exp(-Δe)]. When the prediction bias is large (high model uncertainty), a strong integral action is maintained to eliminate the bias; when the prediction bias is small (accurate model), the integral action is automatically weakened to prevent integral saturation in the small error range. The exponential function ensures a smooth transition without abrupt changes.

[0066] Proactive differential damping adjustment: Kd = Kd0 × [1 + 0.2 × (dΔe / dt)]. This invention adjusts the system damping based on the rate of change of prediction deviation dΔe / dt instead of the traditional rate of change of error de / dt. When the prediction deviation increases rapidly, the system damping is enhanced in advance to achieve preventative oscillation suppression.

[0067] (2) Direct feedforward compensation for prediction error Calculate the dynamic correction term for thermal decay: ΔK=(K_T_actual-K_T_pred) / K_T_ref. The GRU prediction error is directly converted into a feedforward compensation amount, forming a dual-channel correction architecture that combines real-time compensation for prediction bias with gradual elimination of feedback error, significantly improving response speed.

[0068] (3) Generation of composite control commands The final excitation current command I_out is generated by using a standard PID algorithm combined with a feedforward correction term. I_out=Kp×e+Ki×∫edt+Kd×de / dt+ΔK×I_base; Where e is the speed deviation, e = v_ref - v_act; v_ref is the target drum speed, and I_base is the room temperature reference excitation current. This invention achieves optimized synergy between feedforward compensation and feedback control. The ΔK term provides fast initial compensation, ensuring that the PID term can be finely adjusted.

[0069] The calculated excitation current command I_out is sent to the excitation current adjustment unit to adjust the excitation current of the eddy current machine, thereby realizing real-time correction of the measured torque of the eddy current machine drum and offsetting the effect of thermal decay.

[0070] Online model updates and continuous optimization compensation: Repeat the above steps, collect parameters such as the adjusted excitation current, measured drum speed, core temperature, and measured torque coefficient in real time, feed the measured torque coefficient K_T_actual back to the thermal decay prediction model, update the model parameters online, and feed the measured drum speed v_act back to the nonlinear adaptive PID controller to calculate the new speed deviation e, enter the next 10ms control cycle, generate a new I_out, and start a new round of eddy current excitation.

[0071] The 10ms high-frequency closed-loop feedback in this invention is the core support for achieving accurate dynamic compensation for thermal decay of the eddy current machine. It forms a comprehensive performance improvement advantage from three levels: model, control, and system. At the model level, the measured torque coefficient is fed back to the GRU prediction unit to realize online parameter updates, allowing the model to adapt to changes in the actual operating conditions of the dynamometer, offsetting the cumulative prediction error, and reducing the sample cost of offline training. At the control level, real-time parameters such as the measured speed of the drum and the excitation current are fed back to the nonlinear adaptive PID controller, which not only realizes the real-time self-cancellation of speed deviation and ensures the accuracy of the test speed, but also allows the PID gain adjustment and feedforward compensation terms to be calibrated with measured data, forming a dual-channel correction of fast feedforward compensation and fine feedback adjustment, improving the response speed and accuracy of compensation. At the system level, it provides real-time status information for fault protection and engineering redundancy control, accurately distinguishing between instantaneous interference and real faults, upgrading equipment protection from passive alarm to active adjustment of control strategy, while ensuring the stability of control accuracy during long-term continuous operation of the dynamometer and avoiding accuracy drift caused by thermal decay.

[0072] Overall, closed-loop feedback becomes the key link between GRU look-ahead prediction and PID real-time control, enabling the entire control system to form a virtuous cycle of prediction, compensation, measurement, correction and re-prediction. This effectively overcomes the core challenges of the eddy current engine's thermal decay, including its timing, nonlinearity, and hysteresis, and truly achieves a qualitative leap from reactive adjustment to proactive compensation and real-time correction, ensuring high precision of the dynamometer's torque output and high stability of the system operation.

[0073] Step 4: Parallel fault monitoring and protection.

[0074] The fault protection module monitors sensor signals and prediction deviations in real time. When continuous anomalies are detected, a fault switching mechanism is automatically triggered, switching to standby control mode. This ensures the safe and reliable operation of the system when the prediction model is inaccurate or the sensor fails.

[0075] Step 4: Engineering redundancy control, mainly including: Fault diagnosis is performed based on an intelligent fault diagnosis mechanism using a predictive model. Specifically, by comparing the deviation Δe between the GRU's predicted torque coefficient and the actual value, the predictive model itself is used as a fault detector. When Δe continuously exceeds a threshold, sensor faults or model mismatches are automatically identified, thus upgrading from threshold alarms to model diagnosis.

[0076] Mode switching is achieved through a hierarchical multimodal adaptive control architecture. Specifically, this invention designs a multi-level control strategy including an intelligent prediction mode, a segmented compensation mode, and a cooling protection mode. The modes automatically switch according to the fault level, ensuring optimal performance while guaranteeing system safety. The segmented compensation mode is triggered when the sensor signal is abnormal (temperature remains constant or exceeds the range) or the prediction deviation is too large. It divides the temperature range into segments, using a fixed linear compensation coefficient (a fixed proportional value obtained based on the current temperature) for each temperature range, while the PID parameters use those from the cold state. The aim is to provide basic thermal decay compensation functionality even when the prediction model fails.

[0077] The prediction output constraint is implemented based on physical mechanisms. Specifically, a meaningful limit ([0.7×K_T_ref,1.0×K_T_ref]) based on thermodynamic principles is applied to the GRU prediction output to ensure that the predicted value is always within a physically reasonable range, avoiding non-physical predictions that may be generated by a purely data-driven model.

[0078] The system makes judgments on timing-accumulated faults. Specifically, it uses the accumulation of anomalies over multiple consecutive control cycles as the switching condition, effectively distinguishing between transient disturbances and real faults, thus improving system reliability.

[0079] The system integrates temperature protection and control. Specifically, when an overheating risk (T>200℃) is detected, it not only issues an alarm but also proactively adjusts the control strategy, such as reducing the excitation current by 30%, upgrading traditional passive protection to proactive preventive control.

[0080] The above actions are executed cyclically to achieve 10ms closed-loop control and continuously optimize the compensation effect. In other words, the above steps are executed sequentially within the above time intervals in each control cycle to achieve dynamic compensation for thermal decay of the eddy current engine used in automotive environmental dynamometers.

[0081] The eddy current engine thermal decay dynamic compensation control method provided by this invention has the following advantages: This invention establishes a dynamic compensation mechanism for thermal decay of eddy current engines in the field of automotive environmental dynamometers for the first time, solving the problem of accuracy drift caused by neglecting thermal decay in traditional control systems and filling a technological gap. Specifically, addressing the problem of insufficient adjustment capability and severe accuracy drift of traditional control schemes under thermal decay conditions of automotive environmental dynamometer eddy current engines, this invention innovatively integrates GRU timing prediction with nonlinear adaptive PID control to construct an integrated architecture of precise temperature sensing, GRU timing prediction, and nonlinear adaptive PID dynamic correction. It obtains core thermal decay data through precise temperature measurement and uses the GRU layer to capture the timing accumulation dependency and nonlinear characteristics of thermal decay, providing a reliable data source for thermal decay modeling. In the feedforward channel, an attention-enhanced lightweight GRU layer is constructed to capture the timing accumulation and nonlinear characteristics of thermal decay and predict the torque coefficient of future control cycles. In the feedback channel, a nonlinear adaptive PID dynamic correction module is designed to dynamically adjust the PID gain based on the GRU prediction results and introduce a prediction deviation correction term to dynamically adjust the PID parameters, thus offsetting the impact of thermal decay in real time. By employing GRU timing prediction and PID real-time correction, the steady-state error of thermal decay compensation is ≤±1.0%, and the overshoot is <2%, achieving more than 4 times the accuracy compared to traditional solutions, with high dynamic compensation accuracy. A 10ms control cycle meets dynamic response requirements, and the lightweight GRU layer is compatible with embedded controllers, allowing convergence with only a small sample training (50 sets of data), resulting in low deployment costs. The prediction bias correction term, coupled with engineering redundancy, ensures stable system operation in the event of sensor failure or temperature anomalies, adapting to the long-term continuous operation requirements of dynamometers.

[0082] Simultaneously, redundant engineering design is added to ensure control stability and equipment safety. This compensation strategy fully leverages the advantages of GRU in timing prediction while retaining the reliability of PID control in engineering applications. By combining the timing prediction capability of GRU with the dynamic correction capability of nonlinear adaptive PID, precise dynamic compensation for thermal decay of eddy current engines is achieved, significantly improving the testing accuracy and stability of automotive environmental dynamometers. This represents a fundamental shift from reactive post-event adjustment to proactive pre-event compensation, effectively overcoming the torque decay problem caused by thermal inertia and time lag during continuous braking of eddy current engines. It significantly improves the testing accuracy and operational stability of automotive environmental dynamometers, making it suitable for scenarios with high torque output accuracy requirements, such as exhaust emission testing and energy consumption testing.

[0083] Based on the same inventive concept, this invention also proposes a dynamic compensation control system for thermal decay of an eddy current engine, comprising: The data acquisition module is used to collect in real time the temperature at the outer end of the excitation coil at the top of the eddy current machine, the excitation current of the excitation circuit of the eddy current machine, the measured speed of the drum, and the measured torque of the drum. Based on the temperature, excitation current, and measured torque of the drum, the module calculates the temperature change rate and the real-time torque coefficient; and constructs a 3D feature vector containing temperature, temperature change rate, and real-time torque coefficient in multiple consecutive steps.

[0084] The prediction module is used to input a 3D feature vector into a thermal decay prediction model. The thermal decay prediction model is based on a 1DCNN framework, with an attention module, a GRU layer, and a fully connected layer introduced sequentially after the 1D CNN. The 1D CNN extracts local temporal features representing thermal decay from the 3D feature vector, and the attention module weights and enhances the key temperature range features in the local temporal features. The weighted feature sequence is input into the GRU layer to obtain a high-dimensional hidden state vector containing all the temporal patterns of thermal decay. Based on the high-dimensional hidden state vector, the fully connected layer outputs the predicted torque coefficient value for the next control cycle.

[0085] The compensation module is used to calculate the normalized prediction deviation Δe based on the predicted torque coefficient; to adaptively adjust the PID parameters in segments based on Δe, and to calculate the dynamic correction term for thermal decay based on the adjusted PID parameters; the dynamic correction term for thermal decay is combined with the speed deviation to generate an excitation current command; the excitation current command is sent to the excitation current adjustment unit of the eddy current machine to realize dynamic compensation for thermal decay of the eddy current machine; the speed deviation is the difference between the target speed of the drum and the measured speed of the drum.

[0086] Each module in the aforementioned dynamic compensation control system for thermal decay of the eddy current engine can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0087] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement the steps in the embodiment of the dynamic compensation control method for thermal decay of eddy current machines. Specific implementation methods can be found in the method embodiments, and will not be repeated here.

[0088] Furthermore, the present invention also provides a non-transitory computer-readable storage medium containing instructions, on which a computer program is stored. For example, a memory containing instructions that can be executed by a processor of a computer device to perform the above-described method. For example, the non-transitory computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc. When the computer program is executed by the processor, it can implement the steps in the embodiments of the eddy current engine thermal decay dynamic compensation control method. Specific implementation methods can be found in the method embodiments, which will not be repeated here.

[0089] Those skilled in the art will understand that embodiments of the present invention can provide methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0090] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0091] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0092] These computer program instructions can also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1The steps of the function specified in one or more boxes.

[0093] It should be noted that the specific embodiments described above enable those skilled in the art to more fully understand the present invention, but do not limit the present invention in any way. Therefore, although the present invention has been described in detail in this specification and embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention; and all technical solutions and improvements that do not depart from the spirit and scope of the present invention are covered within the protection scope of the present invention patent. No reference numerals in the claims should be construed as limiting the scope of the claims. Any simple variations or equivalent substitutions of technical solutions that can be readily obtained by those skilled in the art within the scope of the technology disclosed in the present invention are within the protection scope of the present invention.

Claims

1. A dynamic compensation control method for thermal decay of an eddy current engine, characterized in that, Includes the following steps: The temperature at the outer end of the top excitation coil of the eddy current generator, the excitation current of the excitation circuit of the eddy current generator, the measured speed of the drum, and the measured torque of the drum are collected in real time. The temperature change rate and the real-time torque coefficient are calculated based on the temperature, excitation current, and measured torque of the drum. A 3D feature vector containing temperature, temperature change rate, and real-time torque coefficient with multiple consecutive steps is constructed. A 3D feature vector is input into a thermal decay prediction model, which is based on a 1D CNN framework. An attention module, a GRU layer, and a fully connected layer are sequentially introduced after the 1D CNN. The 1D CNN extracts local temporal features representing thermal decay from the 3D feature vector. The attention module then weights and enhances the key temperature range features in the local temporal features. The weighted feature sequence is input into the GRU layer to obtain a high-dimensional hidden state vector containing all temporal patterns of thermal decay. Based on the high-dimensional hidden state vector, the fully connected layer outputs the predicted torque coefficient value for the next control cycle. The normalized prediction deviation Δe is calculated based on the predicted torque coefficient; the PID parameters are adaptively adjusted piecewise based on Δe, and a dynamic thermal decay correction term is calculated based on the adjusted PID parameters; the dynamic thermal decay correction term is combined with the speed deviation to generate an excitation current command; the excitation current command is sent to the excitation current adjustment unit of the eddy current machine to realize dynamic compensation for thermal decay of the eddy current machine; the speed deviation is the difference between the target speed of the drum and the measured speed of the drum.

2. The dynamic compensation control method for thermal decay of an eddy current engine according to claim 1, characterized in that, The attention module introduces a physics-guided attention mechanism. This mechanism introduces a parallel and interpretable physics-rule weight generation path after the excitation stage of the standard SENet channel attention framework. The physics-guided attention mechanism learns the attention weights of each channel through data and combines them with the weights generated by physics rules to generate final attention weights that conform to statistical laws and follow physical common sense.

3. The dynamic compensation control method for thermal decay of an eddy current engine according to claim 2, characterized in that, The step of weighting and enhancing key temperature range features in local temporal features using an attention module specifically includes the following steps: For the key temperature range features in the local time series features that reflect the key state of thermal decay, an enhancement weight of 0.6 to 0.8 is generated through an MLP network; For the non-critical noise and redundant information in the key temperature range features of the local time series features, a basic weight of 0.2 to 0.4 is generated through the MLP network; The generated enhanced weights and basic weights are then weighted element-wise with the key temperature range features in the local temporal features output by the 1D CNN to obtain the weighted feature sequence.

4. The dynamic compensation control method for thermal decay of an eddy current engine according to claim 1, characterized in that, The calculation of temperature change rate and real-time torque coefficient based on temperature, excitation current, and measured drum torque; and the construction of a 3D feature vector containing temperature, temperature change rate, and real-time torque coefficient at multiple consecutive steps, specifically includes the following steps: Calculate the temperature change rate dT / dt based on the temperature T at multiple step sizes; The online identification of the real-time torque coefficient K_T_actual is based on the excitation current I_eddy, the measured drum speed v_act, and the measured drum torque T_actual of the eddy current excitation circuit. The calculation formula is as follows: K_T_actual=T_actual / (I_eddy²×v_act); A 3D feature vector [T, dT / dt, K_T_actual] with multiple consecutive steps is constructed using T, dT / dt, and K_T_actual.

5. The dynamic compensation control method for thermal decay of an eddy current engine according to claim 4, characterized in that, The formula for calculating the normalized prediction deviation Δe is: Δe=|K_T_pred-K_T_actual| / K_T_ref; where K_T_pred is the predicted torque coefficient, K_T_actual is the real-time torque coefficient identified online, and K_T_ref is the room temperature reference torque coefficient.

6. The dynamic compensation control method for thermal decay of an eddy current engine according to claim 5, characterized in that, Based on the piecewise adaptive adjustment of PID parameters using Δe, a dynamic thermal decay correction term is calculated according to the adjusted PID parameters. This dynamic thermal decay correction term, combined with the speed deviation, generates an excitation current command, specifically including the following steps: Let the PID baseline parameters be Kp0, Ki0, and Kd0. The PID parameters are adjusted based on the normalized prediction deviation Δe: Kp = Kp0 × [1 + β × (Δe)], where β is taken in segments according to the prediction quality: Ki = Ki0 × [1 - 0.5 × exp (-Δe)]; Kd = Kd0 × [1 + 0.2 × (dΔe / dt)]; The dynamic correction term ΔK for thermal decay is calculated based on the predicted torque coefficient K_T_pred, the online identified real-time torque coefficient K_T_actual, and the room temperature reference torque coefficient K_T_ref. ΔK=(K_T_actual-K_T_pred) / K_T_ref; Based on the adjusted PID parameters and the thermal decay dynamic correction term ΔK, the final excitation current command I_out is generated: I_out=Kp×e+Ki×∫edt+Kd×de / dt+ΔK×I_base; Where e is the speed deviation, e = v_ref - v_act; v_ref is the target speed of the drum, and I_base is the room temperature reference excitation current.

7. The dynamic compensation control method for thermal decay of an eddy current engine according to claim 1, characterized in that, It also includes real-time acquisition of adjusted temperature, excitation current, actual drum speed and actual drum torque, feeding the actual torque coefficient back to the thermal decay prediction model, updating the model parameters online, and feeding the actual drum speed v_act back to the nonlinear adaptive PID controller to calculate the new speed deviation e, entering the next control cycle, generating a new I_out, and starting a new round of eddy current excitation.

8. A dynamic compensation control system for thermal decay of an eddy current engine, characterized in that, include: The data acquisition module is used to collect in real time the temperature at the outer end of the excitation coil at the top of the eddy current machine, the excitation current of the excitation circuit of the eddy current machine, the measured speed of the drum, and the measured torque of the drum. Based on the temperature, excitation current, and measured torque of the drum, it calculates the temperature change rate and the real-time torque coefficient; and constructs a 3D feature vector containing temperature, temperature change rate, and real-time torque coefficient in multiple consecutive steps. The prediction module is used to input a 3D feature vector into a thermal decay prediction model. The thermal decay prediction model is based on a 1D CNN framework, with an attention module, a GRU layer, and a fully connected layer introduced sequentially after the 1D CNN. The 1D CNN extracts local temporal features representing thermal decay from the 3D feature vector, and the attention module weights and enhances the key temperature range features in the local temporal features. The weighted feature sequence is input into the GRU layer to obtain a high-dimensional hidden state vector containing all temporal patterns of thermal decay. Based on the high-dimensional hidden state vector, the fully connected layer outputs the predicted torque coefficient value for the next control cycle. The compensation module is used to calculate the normalized prediction deviation Δe based on the predicted torque coefficient; to adaptively adjust the PID parameters in segments based on Δe, and to calculate the dynamic correction term for thermal decay based on the adjusted PID parameters; the dynamic correction term for thermal decay is combined with the speed deviation to generate an excitation current command; the excitation current command is sent to the excitation current adjustment unit of the eddy current machine to realize dynamic compensation for thermal decay of the eddy current machine; the speed deviation is the difference between the target speed of the drum and the measured speed of the drum.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is loaded by the processor, it is able to perform the steps of the method according to any one of claims 1 to 7.

Images