Inverter intelligent control method based on adaptive algorithm

Abstract

The invention discloses an inverter intelligent control method based on a self-adaptive algorithm, and relates to the technical field of inverter intelligent control, and the method comprises the steps: multi-source temperature collection and environment recognition, machine learning-based electric-thermal coupling prediction, self-adaptive inverter parameter optimization, and cooperative thermal management and fault diagnosis self-repairing. According to the high-temperature or low-temperature failure risk of the motor and the inverter under the extreme climate, temperature distribution and load information are obtained in real time, the temperature rise trend is predicted in advance, and the current, voltage and modulation strategy of the inverter are actively adjusted, so that safe derating or torque compensation is achieved; multi-sensor cross validation and observer fusion are carried out when sensors drift or devices are aged, stable operation and efficient energy utilization of the system are kept, and the drivability and the whole vehicle reliability in an extreme environment are remarkably improved; in the process, fault diagnosis and self-repairing can be further improved through online model updating, and the durability and economical efficiency of the electric drive system are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of inverter intelligent control, and in particular to an inverter intelligent control method based on an adaptive algorithm. Background Art

[0002] When operating electric vehicles in extremely cold environments (such as high latitudes with temperatures exceeding tens of degrees Celsius below zero) or extremely hot climates (such as deserts with temperatures exceeding 40 degrees Celsius), the temperatures of key components such as the motor stator, permanent magnets, inverter power modules, and battery packs differ significantly from the ambient temperature. Heat accumulation and dissipation are further accelerated when the vehicle is traveling long distances at high speeds or under heavy loads. High temperatures can cause power electronics (such as IGBTs and MOSFETs) to experience a sharp increase in switching losses or overheating. In extremely cold conditions, poor lubrication and increased battery internal resistance can lead to insufficient output power. Furthermore, vehicles have high demands for power response and range. Slow startup in cold temperatures or power derating in high temperatures can result in a poor user experience and even pose safety risks. These demanding operating conditions not only test the ability of the vehicle's thermal management system to quickly adapt to changes in the external environment and load, but also require the inverter and motor to possess high reliability and a sophisticated temperature monitoring strategy to balance power performance and safety.

[0003] A Chinese invention patent with authorization publication number CN106240341B discloses a permanent magnet synchronous motor cooling system for electric vehicles and its control method. The cooling system includes a control unit, a motor winding ammeter, a cooling water pump, a cooling water jacket, a cooling water temperature sensor, a coolant flowmeter, a radiator, and a cooling fan. The control method detects the current passing through the motor stator winding to determine whether the permanent magnet synchronous motor has a tendency to heat up or cool down under the operating conditions, and adjusts the cooling capacity of the cooling system to achieve active adjustment of the cooling system. At the same time, auxiliary feedback adjustment is introduced. The motor operating point is queried through the motor efficiency MAP diagram to obtain the motor efficiency under the corresponding operating conditions, and the heat generated by the motor is calculated. The difference between the motor heat generated and the heat dissipated by the radiator is then fed back to the control unit to correct the adjustment of the cooling system. The present invention actively controls the cooling system before the motor temperature changes, saving energy while ensuring sufficient motor heat dissipation capacity.

[0004] However, combining the above practical application scenarios with the contents of the prior art:

[0005] Currently, under such extreme temperature conditions, the control methods of vehicle inverters and motors often adopt fixed safety thresholds or simple temperature switching strategies, and only limit power when significant overtemperature is detected. In extremely cold conditions, there is also a lack of forward-looking compensation and adaptive regulation of temperature dynamic characteristics. As a result, protection is triggered only when the temperature has seriously exceeded the design range, causing excessive stress accumulation in power devices or motor magnets, accelerating aging and even causing failures.

[0006] At the same time, sensors are prone to drift or failure in extreme temperatures, resulting in inaccurate identification of temperature fields and component performance status. Traditional solutions rely solely on single-point temperature measurement and shallow temperature thresholds, making it difficult to accurately grasp vehicle temperature distribution and changing trends. These issues can lead to significant vehicle deceleration, sudden torque drops, or unstable power output in actual operation, posing serious challenges to vehicle reliability and safety.

[0007] To this end, the present invention provides an inverter intelligent control method based on an adaptive algorithm. Summary of the Invention

[0008] (1) Technical problems solved

[0009] In response to the shortcomings of the existing technology, the present invention provides an intelligent control method for an inverter based on an adaptive algorithm. Through five steps of multi-source temperature acquisition and environmental identification, electro-thermal coupling prediction based on machine learning, adaptive inverter parameter optimization, and collaborative thermal management and fault diagnosis and self-repair, the method targets the high-temperature or low-temperature failure risks of motors and inverters under extreme climates, obtains temperature distribution and load information in real time, predicts temperature rise trends in advance, and actively adjusts inverter current, voltage, and modulation strategies to achieve safe derating or torque compensation. It also performs multi-sensor cross-validation and observer fusion when sensors drift or devices age, maintains stable system operation and efficient energy utilization, and significantly improves driving performance and vehicle reliability under extreme environments; thereby solving the technical problems recorded in the background technology.

[0010] (2) Technical solution

[0011] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0012] An intelligent inverter control method based on an adaptive algorithm includes, when it is detected that the ambient temperature significantly exceeds a preset range, integrating multi-source sensor readings, vehicle operating parameters and external meteorological data and using kernel function interpolation to generate a multi-source fusion temperature distribution field;

[0013] After collecting the multi-source fusion temperature distribution field and vehicle operating conditions, the electric-thermal coupling prediction model is called and error constraints are introduced to generate prediction quantities including at least the predicted temperature distribution field and the predicted heat load value;

[0014] If it is detected that the predicted temperature distribution field is about to exceed the temperature margin, the inverter control parameters are adaptively scheduled to actively limit the amplitude and reduce power loss before the critical temperature is reached, maintaining power output and device safety;

[0015] When the inverter output power and actual temperature distribution are transmitted back to the vehicle thermal management module in real time, the BMS-side temperature is compared with the heat dissipation information to predict the temperature distribution field and calibrate the electric-thermal coupling model. The fan or liquid cooling speed is then dynamically scheduled and the correction data is fed back to the previous step.

[0016] When the monitored sensor value deviates seriously from the predicted temperature distribution field, the fault type is determined and the metric function is calculated, and then the degraded operating value is initiated or the redundant signal is enabled to self-repair the power device or temperature measurement failure.

[0017] Preferably, a multi-point temperature sensing network is deployed in the vehicle power system, the readings of each sensor are recorded and marked, and the measurement results are mapped to the original temperature field according to the spatial position of each measuring point; based on the external environmental information obtained during vehicle operation, the external temperature curve is collected and the environmental impact coefficient is constructed.

[0018] Preferably, the environmental impact coefficient is coupled with the original temperature field to obtain a multi-source fusion temperature distribution field after environmental correction;

[0019] The multi-source fusion temperature distribution field after environmental correction and the external environment identification results are stored in the real-time database and interact with the historical operating condition database.

[0020] Preferably, the acquired multi-source fusion temperature distribution field, external environment recognition results, vehicle operating status, and relevant records in the historical operating condition database are synchronized to the feature construction module;

[0021] A time fusion method with an exponential decay kernel is used for feature extraction to generate comprehensive feature items for subsequent models. The sequence of comprehensive feature items between two historical moments is used in conjunction with the target value to construct the feature-label pairs required for supervised training.

[0022] Preferably, based on the thermophysical characteristics of electric vehicle inverters and motors under extreme temperatures, a hybrid electric-thermal coupling model with comprehensive feature terms as input is introduced into the machine learning framework, and the predicted temperature distribution field, heat load, and magnetic flux attenuation at future moments are output. A loss function integrating physical constraints is introduced to train the electric-thermal coupling prediction model.

[0023] The prediction effect is evaluated through offline verification and online update modes, and periodic fine-tuning training is performed. After the electro-thermal coupling prediction model is evaluated as qualified, its prediction output is passed to the inverter control parameter optimization module.

[0024] Preferably, the obtained predicted output is compared with the designed temperature safety upper limit and magnetic flux lower limit to form a temperature margin and a magnetic flux margin; the vehicle's driving intentions for the current and next time periods are collected, and a set of control parameters for the existing inverter control is simultaneously extracted. If the risk of overheating or insufficient magnetic flux is predicted, a corresponding power limiting strategy is selected based on the temperature margin and the magnetic flux margin;

[0025] The dynamic power limit value is combined with the original control parameter set to obtain a parameter mapping with upper and lower limit constraints; if temperature or flux risks are detected, the control parameters are dynamically limited or adjusted according to the dynamic power limit value.

[0026] Preferably, based on the determined parameter mapping, an adaptive algorithm is built into the inverter controller to realize online solution of each control parameter, and obtain the optimal control parameters finally applied to the execution end, wherein:

[0027] The predicted output is incorporated into the objective function, and the control parameters are optimized for future time periods to obtain the optimal control parameters that change over time. A dynamic control strategy is constructed and finally executed in combination with parameter mapping. If the actual temperature or flux demagnetization rate is still deviated from the expected value, the online prediction update mechanism is automatically called in the next control cycle for online fine-tuning or incremental training to correct the input of the electro-thermal coupling prediction model and the upper limit of the control parameters.

[0028] Preferably, the actual temperature distribution field and power flow information of the inverter after the dynamic control strategy is actually executed and the updated optimal control parameters are collected; the inverter-side data is compared with the predicted temperature distribution field, and the deviation between the prediction and the reality is evaluated through the control error term;

[0029] If the control error term is greater than expected, the machine learning model or sensor calibration is corrected;

[0030] The optimal control parameters, actual temperature distribution field, and control error terms are passed to the vehicle thermal management module. At the same time, the battery status and operating status of the cooling system are obtained from the BMS side and marked as battery management system status information. After alignment, a multi-terminal integrated data packet is formed.

[0031] Preferably, after obtaining the merged multi-terminal integrated data packet, the vehicle thermal management module and the BMS jointly perform a global scheduling; after calculating the optimal solution of the introduced global objective function, the vehicle thermal management module sends a scheduling instruction to the cooling system and guides the battery preheating or balancing operation on the battery management system BMS side.

[0032] Preferably, the actual action feedback generated by the final execution of the scheduling instruction is fed back in real time to the sensor fusion parameter correction and environmental recognition update, as well as the input correction and fine-tuning of the electro-thermal coupling prediction model; if it is found that the actual temperature response after the execution of the adaptive control strategy deviates significantly from the prediction, the scheduling strategy is dynamically adjusted, the global objective function is re-evaluated and local corrections are performed.

[0033] Preferably, after obtaining the actual temperature distribution field and other key sensor readings and the corresponding predicted values, a metric function based on observer and residual fusion is defined. When the metric function increases sharply in a short period of time and exceeds the threshold, the specific fault type of the detected suspicious signal is determined by multi-dimensional residual pattern matching.

[0034] Preferably, a degraded operating value indicating the extent of partial derating or backup parameter activation is defined based on the identified fault type and abnormality level;

[0035] When the fault is serious, the degraded operation value increases. If the impact of the fault is limited, only minor temporary replacement processing is performed. For sensor drift or local failure, redundant sensor data or correction values ​​based on predicted output are temporarily enabled as alternative inputs. When the degraded operation value exceeds the preset warning line, degraded operation is automatically executed.

[0036] (3) Beneficial effects

[0037] The present invention provides an inverter intelligent control method based on an adaptive algorithm, which has the following beneficial effects:

[0038] Through five closely linked steps, the system provides full process optimization for motors and inverters in extreme temperature environments, from real-time monitoring to fault diagnosis and self-repair. The overall beneficial effects are mainly reflected in the following aspects:

[0039] Through multi-source temperature measurement and environmental identification, not only has the quality of sensor data in extremely cold and hot areas been significantly improved, but it has also achieved dynamic adaptation to external conditions such as wind speed and humidity, thus providing high-precision input for subsequent machine learning models.

[0040] Based on the predicted temperature distribution field and power heat load The electric-thermal coupling prediction can identify the risk of overheating and flux attenuation in advance, and provide a safe margin ΔT safe , ΔΨ safe Quantitative evaluation is performed, thereby reducing the fatigue accumulation and failure probability of components under large temperature differences;

[0041] Adaptive inverter control uses prediction results to flexibly adjust control parameters before temperatures reach high-risk levels, breaking away from the traditional passive mode of reducing power only when temperatures exceed the specified threshold. This significantly enhances vehicle power and efficiency in extreme environments.

[0042] Using the actual temperature distribution field of the inverter The system uses real-time feedback information such as CMOS and CMOS to couple the inverter-side execution results with the BMS and vehicle cooling system, creating a unified optimization from battery preheating to fan liquid cooling adjustment. This collaborative thermal management can allocate more cooling resources to protect core power devices at high temperatures, while moderately preheating the battery and motor at low temperatures, balancing energy consumption and performance.

[0043] For sensor drift, power component aging or sudden failure, the multidimensional residual vector R s (x, t) and metric function Φ obs (t) and other multidimensional residual fusion methods to timely diagnose anomalies and use the degraded operating value Γ degrade (t) Triggering self-repair or graded derating. This allows the system to maintain safe operation even in extreme cold or heat, and the collected abnormal data feeds back into the machine learning model, forming self-learning and self-adaptive capabilities;

[0044] In summary, this solution can ensure the safety and stability of the power system while maintaining optimal efficiency and endurance in various extreme environments in terms of temperature monitoring, thermal prediction, inverter control, vehicle thermal management, and fault diagnosis and self-repair, thus achieving synergistic gains that take into account multiple objectives. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 The figure is a flow chart of the inverter intelligent control method based on the adaptive algorithm of the present invention. DETAILED DESCRIPTION

[0046] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0047] See also Figure 1 The present invention provides an inverter intelligent control method based on an adaptive algorithm, comprising:

[0048] Step 1: When the vehicle detects that the ambient temperature exceeds the preset range, it automatically integrates the output of multi-source temperature sensors with external meteorological station data, historical operating conditions information, and the vehicle's current operating parameters, and constructs a multi-source fusion temperature distribution field through a kernel function weighted interpolation strategy. At the same time, the environmental impact coefficient Θ is integrated ext (t) To capture the thermal coupling effects of key components;

[0049] The step 1 includes the following:

[0050] Step 101: Sensor placement and temperature data field construction

[0051] A series of sensors are placed in the vehicle power system (such as the motor stator, motor end cover, and inverter power module surface) to form a multi-point temperature sensing network, such as thermistor sensors, platinum resistance temperature sensors, thermocouples, and integrated temperature / humidity sensors. The readings of each sensor at time t are recorded and marked as (where i = 1, 2, ..., N represents the sensor number, and N is the total number of sensors); based on the spatial position x of each measuring point (i) (which can be expressed as three-dimensional coordinates), and its measurement results are mapped to the original temperature field T(x, t); to avoid local distortion caused by measurement differences between sensors, the following fusion formula with a weighted kernel function is introduced:

[0052]

[0053] Where: Ω represents the overall location domain of the sensor; Ψ(||xx (i) ||) is the kernel function, which is used to calculate the spatial distance ||xx (i) ||Weight the sensor measurements. Generally, a Gaussian kernel or other positive definite kernel can be used to achieve smooth interpolation. is the measurement value of the i-th sensor at time t; where the kernel function uses a Gaussian kernel,

[0054] K(x i ,x)=exp(-||x i -x|| 2 / (2σ 2 )), where σ = 0.12m is obtained by actual vehicle calibration;

[0055] When used, the kernel function Φ is used for weighting, which can significantly reduce the local deviation caused by single-point measurement, achieve more reliable temperature field reconstruction after multi-sensor fusion, and improve the accuracy of the temperature distribution field; by defining the original temperature field T(x, t) as a continuous distribution field function, it can be directly output to the machine learning model required in the second step for training or real-time prediction input, avoiding the secondary error of data format conversion.

[0056] Step 102: External environment identification and multi-source information fusion

[0057] Based on the real-time data from the meteorological station, cloud weather API or other external environmental information obtained during vehicle operation, the external temperature curve is marked as T env (t), and further define the environmental impact coefficient Θ based on the vehicle's current location and possible altitude, wind speed, humidity, etc. ext (t), please refer to the following form for details:

[0058]

[0059] Where: h(t) is the current altitude of the vehicle, w(t) is the wind speed, and RH(t) is the relative humidity; h0, w0, and RH0 are the reference values ​​of altitude, wind speed, and humidity, respectively; Take the natural logarithm of x(τ) / x0 (element-wise ratio) to form a column vector It is dimensionless; A is a 3×3 matrix and can be expanded or reduced according to the number of environmental factors; ω(τ) is a scalar function with no time-series weighting; or exponential decay, Gaussian kernel, etc. can be used; A = diag(1 / h0, 1 / v0, 1 / u0), where h0 = 1000m, v0 = 5m / s, and u0 = 60%, is established based on the average value of the national meteorological reference station;

[0060] If a new environmental factor is added (such as light intensity S), the column expansion method is adopted while maintaining the unit diagonal: A'=diag(A,1 / S0);

[0061] The environmental impact coefficient Θ ext (t) is coupled with the original temperature field T(x,t) to obtain the multi-source fusion temperature distribution field after environmental correction Its form can be expressed as:

[0062]

[0063] Where: Λ(·) is a mapping function coupled with the external environment, which is used to characterize the additional temperature compensation effect under air cooling, heat dissipation or extreme cold conditions. In other words, the dimensionless environmental impact coefficient Θ can be realized through two implementation paths: physical-parameterized linear combination or data-driven neural network. ext (t) is mapped to the temperature increment ΔT ext (x, t); For example, in extremely cold conditions, the sensor location may not be completely cooled but the external temperature is extremely low. In this case, Λ(Θ ext (t)) can give a negative offset based on simulation or data-driven to more accurately characterize the actual cooling effect; where Λ(·) is in the form of:

[0064]

[0065] is a dimensionless index of forced convection; is the dimensionless index of natural convection; is the dimensionless index of radiation heat; f c (x) is the spatial weight of forced convection; f r (x) is the natural convection space weight; fs (x) is the radiation space weight; γ c is the global weight of forced convection, ranging from 0.3 to 0.6; γ r is the global weight of natural convection, ranging from 0.1 to 0.4, γ s is the global radiation weight, ranging from 0.1 to 0.3; the constraint is γ c +γ r +γ s =1;T scale is the temperature scale quantity used to convert dimensionless terms into temperature increments;

[0066] The multi-source fusion temperature distribution field after environmental correction The external environment recognition results (such as classification labels such as extreme cold, extreme heat, and humid heat) are uniformly stored in the real-time database and interact with the historical working condition database, marked as

[0067] When used, the environmental impact coefficient Θ ext (t) Multi-source fusion temperature distribution field after environmental correction The integration of temperature and humidity can timely capture the heat flow differences caused by extreme temperatures, avoiding the hysteresis or deviation caused by internal temperature measurement alone. Once the external climate or altitude changes significantly, the environmental impact coefficient Θ ext (t) will change accordingly and be reflected in the mapping function Λ(·), which enhances the system's ability to adapt to the external environment; the environmental impact coefficient Θ is introduced ext (t) and mapping function Λ(·) as environmental correction units, compared with the traditional temperature limiting protection based only on the on-board sensor itself, it adds a quantitative characterization of the external dynamic environment. This enables thermal management decisions to no longer rely solely on internal measurement values, but to achieve a comprehensive assessment of the climate scenario in which the vehicle is located.

[0068] Step 2: Receive multi-source fusion temperature distribution field After analyzing the vehicle's operating status, the electric-thermal coupling model combining machine learning and physical priors is used to predict the inverter and temperature distribution field. and power heat load Make predictions and identify potential overheating or extreme cold risks through high-order error constraints and online updates, providing a quantitative reference for the adaptive control parameter π;

[0069] The second step includes the following:

[0070] Step 201: Multidimensional feature engineering and data reconstruction

[0071] The multi-source fusion temperature distribution field obtained External environment recognition result Θ ext(t), vehicle operating status u(τ) (vehicle speed, load, acceleration), and related records in the historical operating condition database are synchronized to the feature construction module;

[0072] Align all data on the time axis to have a unified sampling point t k , k=1,2,…,K, K is the total number of samples; the time fusion method with exponential decay kernel is used for feature extraction to emphasize the impact of recent temperature changes on the thermal response of the system. The following mapping is defined to generate the comprehensive feature term F(t k ):

[0073]

[0074] Where: Δt represents the fusion time length, which is used to comprehensively consider the temperature and external environment influences in a recent period of time;

[0075] α is the time decay factor, which ranges from 0.01 to 0.25 and is used to give greater weight to recent data in the integration process; G(·) is the feature combination function, which is used to convert the environmental impact coefficient Θ into ext (τ) is superimposed on the vehicle operating state u(τ) to form a temperature distribution field that can be fused with multiple sources The corresponding comprehensive characteristic term F(t k ); For example, two feasible implementation plans can be provided: a linear-physical hybrid method and a lightweight neural network mapping method, with detailed input features, normalization methods, weights / network structures, and safety clamping strategies to ensure that the output and temperature field dimensions are consistent and can be directly used in subsequent steps;

[0076] Comprehensive feature item F(t k The result dimension of ) can be simplified by discretizing or projecting the spatial position x. Specifically, the spatial position x can be divided into several key areas and the integrals are calculated separately. The results are then combined into a vector or matrix for subsequent use by machine learning algorithms.

[0077] In the real-time prediction scenario, using the historical moment t k-j to t k The comprehensive characteristic term F(t) sequence between the two, combined with the corresponding future temperature distribution field Or power device heat load H(t k+δ ) and other target values ​​to construct the feature-label pairs required for supervised training;

[0078] For key indicators such as motor flux attenuation or inverter overheating, corresponding label sequences Y(t k ), including the temperature rise rate or possible performance degradation, to facilitate the construction of multi-task learning models;

[0079] When used, it can more sensitively capture rapid changes in temperature and, combined with the external environment and vehicle operating conditions, improve the perception accuracy of sudden thermal shock or rapid temperature drop, highlighting the impact of recent conditions; the comprehensive feature item F(t k ) and simultaneously integrates the multi-source fusion temperature distribution field Environmental impact coefficient Θ ext (t) and vehicle operating status u(τ), which helps the machine learning model to more comprehensively characterize the electro-thermal coupling process, retaining historical information while highlighting the importance of recent changes. It can also reflect the combined effects of external climate and vehicle power demand at the model input stage.

[0080] Step 202: Electric-thermal coupling prediction model training and dynamic evaluation

[0081] According to the thermal physical characteristics of electric vehicle inverters and motors under extreme temperatures, a hybrid electric-thermal coupling model is introduced into the machine learning framework: on the one hand, a neural network (such as a multilayer perceptron, marked as Φ NN ) to capture the complex nonlinearity of the temperature distribution field; on the other hand, a small amount of physical priors (such as thermal conductivity in the heat transfer equation and stator copper loss characteristics) are introduced, integrated into the neural network structure and constrained in the loss function;

[0082] The input comprehensive feature term F(t k ), output the predicted temperature distribution field at the future time Inverter switching device thermal load Magnetic flux attenuation etc. The symbols here Indicates the model prediction value, which is compared with the actual measurement or verification data;

[0083] In order to reflect the mechanism of electrical-thermal coupling and reduce the inaccurate extrapolation caused by pure data drive, a loss function integrating physical constraints is introduced. Where W is the model weight vector, including the neural network weights and related physical hyperparameters, and the loss function can be defined as:

[0084]

[0085] Where: is the actual observed or marked temperature distribution field; Predict the temperature distribution field for the model;

[0086] Γ represents the spatiotemporal domain (covering the training period and the spatial range of key components), Γ f It is an adjustable factor with a value between 1 and 20, used to focus on key areas; It is a physical residual operator used to measure the degree of violation of the prediction results to the basic thermodynamic equations (such as heat conduction equations, heat dissipation models) or the inverter power loss mechanism. Its typical form is:

[0087]

[0088] ρ,c p is the material density and specific heat capacity; k is the thermal conductivity (the copper winding and aluminum shell partitions may be different);

[0089] q gen (x, t) is the local unit volume heat rate, which is calculated from the current density and device switching loss; h conv (x) is the local convective heat transfer coefficient; T ∞ is the ambient temperature, is the predicted temperature; β is the physical constraint weight coefficient, which is greater than 0 and is used to balance the prediction accuracy and physical consistency loss function;

[0090] After the electric-thermal coupling prediction model is trained, the prediction effect is evaluated through two modes: offline verification and online update. Offline verification refers to inputting complete test data (including extreme cold and hot conditions) into the model and comparing it with the measured temperature. Online update refers to continuously collecting the latest multi-source fusion temperature distribution field during actual vehicle driving. Periodically perform fine-tuning training to correct the model's deviations in different environments and aging conditions;

[0091] After the model is evaluated and qualified, its prediction output (i.e. the future predicted temperature distribution field) Thermal load prediction value of power devices Flux decay prediction The information (e.g., etc.) will be passed to the inverter control parameter optimization module in step three to guide the latter to actively adjust before overheating or performance degradation may occur.

[0092] When in use, the integration of physical mechanisms and machine learning can maintain high prediction accuracy in scenarios where data is missing or distributed off-site (such as extreme cold and high altitude or persistent heat). The dynamic evaluation and iterative training mechanism ensures that the model can be continuously corrected according to the changing vehicle operating conditions (such as component aging and sensor offset), thereby maintaining high reliability throughout the vehicle life cycle; in order to provide the necessary temperature and performance evolution reference, it helps the inverter controller to achieve active adaptive control in extreme environments, and can maintain the efficiency and accuracy of the model during the actual vehicle operation stage through online updates.

[0093] Step 3: When overheating or low temperature hazards are identified, the adaptive optimization algorithm is automatically loaded into the inverter controller to and temperature margin ΔT safe(t) Adjust the control parameters in real time, dynamically limit or schedule the power output before the temperature critical value, and adjust the magnetic flux margin ΔΨ safe (t) Prevent excessive demagnetization of magnetic steel;

[0094] The step three includes the following:

[0095] Step 301: Control requirements extraction and key parameter mapping

[0096] Read the prediction results in the future period [t k , t k+δ ] Future predicted temperature distribution of motor windings and power modules And the thermal load prediction value of the inverter and flux decay prediction wait;

[0097] The above predicted value is compared with the designed temperature safety upper limit T safe and the lower limit of magnetic flux Ψ safe Compare and form temperature margin ΔT safe (t) and flux margin ΔΨ safe (t) to determine whether there is a potential risk of overheating or severe demagnetization; collect the vehicle's driving intentions in the current and next time periods (such as acceleration, constant speed cruising, slope conditions, etc.), marked as D(t); and extract the control parameter set Π={ω switch ,i max , v dc …},in:

[0098] ω switch Indicates the inverter modulation switching frequency; i max Indicates the peak current allowed to be output by the inverter; v dc Indicates bus voltage, etc.

[0099] If the prediction is at time t k+δ There is a risk of overheating or insufficient magnetic flux near the temperature margin ΔT. safe (t) and flux margin ΔΨ safe (t) Pre-calculate the appropriate power limitation strategy, such as limiting the peak current amplitude or slightly reducing the upper limit of the bus voltage, to achieve early cooling or reduce losses. The specific mapping relationship can be expressed as:

[0100] Γ limit (t) = Θ ctrl [ΔT safe (t),ΔΨ safe (t),D(t)]

[0101] where Γ limit(t) represents the dynamic power limit imposed by the system at the future time t, Θ ctrl (·) A function that generates control limits for combining safety margins with vehicle demands, for example:

[0102]

[0103] Where: Θ ext (t) is the environmental impact coefficient; u(t) is the vehicle operating state; T dyn (t) is the real-time temperature data of the vehicle power system; Predict temperature distribution fields for the future; is the flux decay prediction quantity;

[0104] Set the dynamic power limit value Γ limit (t) Combined with the original control parameter set Π, the parameter mapping Π with upper and lower limit constraints is obtained bounded (t); When it is determined that no derating is required, the parameter mapping Π bounded (t) is the same as the default value of the control parameter set π;

[0105] If a temperature or flux risk is detected, the dynamic power limit value Γ limit (t) for switching frequency ω switch , maximum current i max Or DC bus voltage v dc Dynamically limit or adjust other control parameters;

[0106] When in use, by using the predicted value output in step 2, power scheduling preparation can be completed before the high temperature / low temperature extreme point arrives, rather than waiting until the system temperature alarm occurs before emergency de-rating; parameter mapping Π bounded (t) not only includes simple peak current limiting, but also allows for comprehensive adjustment of the inverter switching frequency or bus voltage to enhance protection against extreme temperatures; introduces a temperature margin ΔT safe (t) and flux margin ΔΨ safe The mapping mechanism of (t) can accurately indicate when and to what extent power limiting is performed, avoiding excessive or insufficient protection operations.

[0107] Step 302: Adaptive control algorithm implementation and closed-loop execution

[0108] Map π according to the determined parameters bounded (t), an adaptive algorithm is built into the inverter controller (nonlinear model predictive control or Lyapunov basis control method can be used) to solve the control parameters online and obtain the optimal control parameter Π that is finally applied to the execution end. * (t);

[0109] The heat load prediction value output from step 2 and flux decay prediction The objective function is also included to ensure that the driving needs are met to the maximum extent under the premise of considering the motor temperature and the thermal stress of the inverter power device. The following example gives a nonlinear cost function for the future period ΔT ctrl Optimize the control parameters:

[0110]

[0111] Where: Ω1, Ω2, Ω3 are weight coefficients, all ranging from 0 to 1, used to balance vehicle power demand and temperature safety; Θ drive (D(τ)) is the power demand that measures the vehicle's dynamic response; is the constraint penalty term for the inverter thermal load; margin (ΔΨ safe (τ) represents the allowable variation of the control parameters after considering the flux safety margin;

[0112] By the objective function The maximum (or minimum) value of is solved to obtain the optimal control parameter Π that changes with time * (t), and construct a dynamic control strategy and combine it with the parameter mapping Π bounded (t) to finally execute;

[0113] The inverter controller maps the bounded (t) Real-time update of the modulation strategy (such as Space, Vector, and PWM carrier frequency and duty cycle limit), current command upper limit, and bus voltage control signal, and compare them with the actual output feedback of the motor. If the actual temperature or flux demagnetization rate is still deviated from the expected value, the online prediction update mechanism in step 2 is automatically called in the next control cycle. Using the real-time feedback from actual operation, online fine-tuning or incremental training is performed to correct the input of the electro-thermal coupling prediction model and the upper limit of the control parameters.

[0114] The execution data of the closed-loop strategy (including actual temperature, real-time power output and optimal control parameter Π * The time series distribution field of (t) will be passed to step 4 to coordinate with the BMS and vehicle thermal management module to improve the thermal management and performance optimization of the entire vehicle;

[0115] When in use, the adaptive algorithm that incorporates heat load prediction is used to achieve dynamic and accurate control strategies under different temperature conditions. Through closed-loop interaction with the prediction model, when the actual operation is inconsistent with the prediction, it can be quickly corrected to maintain the long-term effectiveness of the solution. Considering the vehicle power demand Θ drive(D(τ)) and inverter thermal load prediction value and magnetic flux safety margin ΔΨ safe (τ) Achieve multi-objective nonlinear optimization, rather than simply pursuing temperature tolerance or a single maximum torque output;

[0116] By changing the space π margin (·) Incorporating the risk of magnetic flux attenuation into control constraints and combining it with a derating strategy based on thermal prediction can achieve synergy between magnetic steel protection and inverter safety.

[0117] Step 4: When the inverter completes the adaptive control and outputs the actual temperature distribution field After comparing the power data, the vehicle thermal management module combines the battery temperature and balancing information on the BMS side to predict the temperature distribution field. The actual temperature distribution field The difference is evaluated for residuals and fan or liquid cooling is activated to preheat the battery and motor at low temperatures. The correction data is then fed back to the machine learning model for continuous updating.

[0118] The step 4 includes the following contents:

[0119] Step 401: Real-time feedback and dynamic multi-terminal linkage

[0120] Collect the actual temperature distribution field of the inverter after the dynamic control strategy is actually implemented and power flow information (including output current, voltage, etc.), as well as the updated optimal control parameter Π * (t);

[0121] Combine the inverter-side data with the future predicted temperature distribution field For comparison, the error term Λ is controlled by err (t) is used to evaluate the deviation between prediction and reality. The formula is defined as follows:

[0122]

[0123] Where: is the actual temperature distribution field of the inverter and related components at position x; is the predicted temperature distribution field obtained based on the thermal model and machine learning prediction; W loc (x) is a local weight function used to emphasize the importance of specific parts (such as IGBT modules, DC bus terminals, etc.), usually W loc (x)>1 to amplify the deviation of the key area; for the secondary parts, W loc (x)<1;

[0124] α is the global adjustment factor, It is used to control the proportion of the cubic error term in the overall metric. It can be set according to the sensitivity to the risk of temperature peaks. Cubic operations are more sensitive to large deviations and can impose higher penalties in the event of extreme temperature changes.

[0125] It is the gradient of the difference field between the actual temperature and the predicted temperature in the spatial domain, reflecting the direction and rate of local temperature change;

[0126] ||·|| 2 is a vector The Euclidean norm square is used to measure the size of the gradient difference; β is the gradient difference weight, Ω inv This is the space where key components such as inverters and motors are located;

[0127] If the control error term Λ err The larger the (t) value is, the more serious the deviation between the prediction and the actual situation is, and the machine learning model or sensor calibration in the second step needs to be corrected in time; if the control error term Λ err (t) If the prediction model remains within a small preset range, it is considered that the current prediction model and the control measures are basically matched;

[0128] The optimal control parameter Π * (t), actual temperature distribution field Control error term Λ err (t) and other control feedback information are transmitted to the vehicle thermal management module, and the battery status (temperature, remaining capacity, internal resistance, etc.) and the operating status of the cooling system (fan, water pump or liquid cooling device) are obtained from the battery management system BMS side, and marked as battery management system status information BMS info (t k );

[0129] To ensure a unified time base, the information is aligned on the same time axis t to form a multi-terminal integrated data packet:

[0130]

[0131] The data package will serve as the basic input for global thermal management optimization decisions, providing complete status information;

[0132] When in use, the real-time data of the inverter control terminal, battery management system BMS and cooling system are uniformly packaged, which facilitates the vehicle thermal management module to make comprehensive decisions. By comparing the actual temperature distribution field of the inverter and predicted temperature distribution field It provides new correction data for the second-step model, gradually reduces the prediction error, and can impose differentiated monitoring efforts on different sensitive areas of the inverter (such as IGBT modules and DC bus contacts), greatly improving the precision.

[0133] Step 402: Global thermal management scheduling and comprehensive performance optimization

[0134] Get the combined multi-end integrated data packet M(t k ), the vehicle thermal management module and the battery management system (BMS) work together to perform a global scheduling to maximize the vehicle's power performance and energy utilization efficiency while ensuring safe temperature limits. The global objective function is introduced:

[0135]

[0136] Where: ΔT plan Planning time frame for thermal management; Ω A and Ω B They are weights for measuring vehicle driving performance and thermal safety, and their values ​​are between 0 and 1;

[0137] C drive (D(τ)) represents the contribution to satisfying driving requirements (such as torque output and acceleration requirements);

[0138] It is a safety penalty function for overheating risk or low temperature decay, and can also include factors such as heat dissipation power consumption and BMS preheating loss;

[0139] By the global objective function Π global Maximization can be performed to find the optimal balance between vehicle power and temperature safety under different operating conditions; the global objective function Π is calculated global After finding the optimal solution, the vehicle thermal management module sends scheduling instructions (such as fan speed and liquid cooling rate) to the cooling system and guides battery preheating or balancing operations on the battery management system (BMS).

[0140] For example, when the actual temperature distribution field of the inverter If the vehicle is nearing a safety threshold but the battery is relatively cold, the inverter's heat dissipation can be increased, while also preheating the battery appropriately to avoid a double temperature bottleneck caused by the upcoming high-load condition. If the inverter temperature is low but the motor demand is high during the early startup phase in cold areas, the thermal management module can reduce the heat dissipation of the heat sink accordingly, or even use a small amount of heat generated by the motor to help accelerate the system into the operating temperature range, thereby avoiding efficiency losses caused by low motor and inverter temperatures.

[0141] When used, by comparing with the optimal control parameter Π *(t) Coupling can reduce the inverter output power and battery discharge power in the same direction when necessary, or reduce the torque peak within the vehicle speed range, thereby achieving an integrated design of full-vehicle energy consumption optimization and hardware safety protection;

[0142] The actual action feedback generated by the final execution of the scheduling instruction (including fan speed curve, liquid cooling flow, BMS heating power, etc.) is fed back in real time to the sensor fusion parameter correction and environment recognition update of step one, as well as the input correction and fine-tuning of the electric-thermal coupling prediction model of step two, so that the subsequent machine learning model update can fully incorporate the temperature change effect caused by the heat dissipation and preheating action of the whole vehicle; if it is found that the actual temperature response of the adaptive control strategy is significantly different from the prediction (which can be determined by the control error term Λ err (t) monitoring results), the scheduling strategy is dynamically adjusted and the global objective function Π is evaluated again global And make local corrections to maintain continuous optimization of the system;

[0143] When in use, the thermal states of the inverter and battery ends are included in the same optimization target to avoid excessive heat dissipation or preheating of a single system leading to increased total energy consumption or insufficient power. The scheduling time range ΔT plan The objective function form can be adapted according to the vehicle operation scenario (long-distance highway, urban commuting) to enhance flexibility; by returning the actual execution and temperature results to the previous steps, the cycle system from data collection to full vehicle scheduling is further improved to enable the system to have the characteristics of self-iteration and self-correction; in the global objective function Π global The contribution of satisfying driving needs is also considered in C drive (D(t)) and security penalty function A multi-objective optimization framework has been built; through the two-way coordination of heat dissipation and preheating, proactive scheduling can be made in both high and low temperature extremes, rather than simply managing the heat dissipation system or the preheating system separately;

[0144] Taking the deviation between the actual operation of the inverter end and the previous prediction as the starting point, a real-time multi-terminal linkage feedback mechanism is constructed, which enables the system to quickly grasp the actual situation of temperature and power changes. This feedback is combined with the scheduling requirements of the BMS and cooling system, and the global objective function π is used to calculate the actual temperature and power changes. global Multi-objective optimization is performed to coordinate the vehicle's thermal management and power output in extreme temperatures. This closed-loop collaboration forms a complete technical chain that not only ensures the safety of the inverter and motor, but also provides drivers with optimal power and endurance performance in extreme environments. This demonstrates the systemic advantages of this solution at the vehicle level: efficient integration, active adaptation, and precise control.

[0145] Step 5: When the sensor or power device data is consistent with the predicted temperature distribution field When a serious deviation occurs, the fault diagnosis and self-repair process is triggered, and the multidimensional residual vector R of the redundant measurement points is fused by the observer. s (x, t) and metric function Φ obs (t) Quantify the degree of abnormality and locate the failure site. If it is determined to be high risk, the degraded operation value Γ degrade (t) Implementing graded derating or activating temporary replacement signals to feed back fault information to the inverter controller and thermal management module to collaboratively minimize the impact of the fault;

[0146] The step five includes the following:

[0147] Step 501: Fault identification and abnormality location

[0148] The actual temperature distribution field obtained from and other key sensor readings, extracting a redundant dataset S for cross-validation snes (t), such as temperature observations at multiple measurement points around the same component or cross-information from different signal channels;

[0149] At the same time, the predicted values ​​of the same physical quantities are obtained from the electro-thermal coupling prediction model, such as the predicted temperature distribution field

[0150] R s (x, t) represents the vector form of the difference between the actual value measured by the sensor and the predicted value, which can specifically include multi-dimensional components such as temperature, power, and magnetic flux decay rate to form a residual signal;

[0151] To identify the root cause of the fault, a metric function Φ based on the fusion of the observer (a state estimator based on the system model) and the residual is defined. obs (t):

[0152]

[0153] Where: Ω sens represents the sensor space domain; R s (x, t) is the multidimensional residual vector (e.g., temperature difference, power difference, magnetic flux difference) at position x and time t, which summarizes the multidimensional error between the actual measurement value and the model or predicted value; ||·|| F is the Frobenius norm of the matrix, used to measure R s (x),t)R s (x,t) T The overall amplitude of ; γ>1 is a high-order amplification factor, which is used to increase the sensitivity to significant deviations;

[0154] When the metric function Φ obs(t) If the value increases sharply within a short period of time and exceeds the threshold, it indicates that the corresponding sensor or power module data is suspected to be faulty or drifting; if the abnormality is significant only in a certain area, it may indicate aging of the local device or loose connection;

[0155] For the detected suspicious signals, the specific fault type is determined by comparing the cross redundancy between different sensors and the physical model prior (such as the stator temperature should be less than a certain threshold, the possible point of irreversible demagnetization of the magnetic steel, etc.), that is, through multi-dimensional residual pattern matching (rule base method) or machine learning classification, such as temperature sensor drift, IGBT gate fault, poor DC bus contact, etc. The preliminary judgment result is combined with the metric function Φ obs (t) Pass it to the next step to help the subsequent execution of self-repair or degradation mechanism;

[0156] When using, the multidimensional residual vector R s The vectorized structure of (x, t) can not only capture single temperature deviations, but also identify inconsistencies between multiple signals such as power and flux, thereby improving the comprehensiveness of fault detection. Based on high-order operations, sudden or local severe deviations are not masked by the average value, and are more sensitive to high temperature hot spots or rapid aging. The observer is fused with the residual matrix R s (x,t)R s (x,t) T Combined with the Frobenius norm and high-order power operations, in a metric function Φ obs In (t), the differences of multiple sensors and multiple physical quantities are evaluated simultaneously, which can maintain high recognition of complex fault modes.

[0157] Step 502: Self-repair and downgrade operation strategy

[0158] According to the identified fault type and abnormality level, the degraded operation value Γ is defined degrade (t), indicates the extent to which the system may require partial derating or backup parameter activation:

[0159] Γ degrade (t) = f deg (Φ obs (t), fault type, over-temperature / over-current margin)

[0160] where f deg It is a function that converts the observer output and fault properties into executable scheduling instructions;

[0161] When the fault is serious, the degraded operating value Γ degrade (t) increases. If the impact of the failure is limited, only a minor temporary replacement is performed;

[0162] In the event of sensor drift or partial failure, redundant sensor data or correction values ​​based on the second-step prediction model can be temporarily enabled as alternative inputs, allowing the third-step adaptive control logic to continue to operate smoothly.

[0163] When the degraded operating value Γ degrade (t) If the preset warning level is exceeded, the system will automatically perform downgraded operation;

[0164] The optimal control parameter Π in the third step * In step (t), a more conservative limit is added or the power limit is lowered to reduce device thermal stress. In conjunction with the global thermal management in step 4, if the failure causes a decrease in heat dissipation efficiency, the battery cooling or current diversion on the BMS side can be strengthened in advance. If a local high-temperature sensor fails, other measurement points and prediction models are used to maintain general temperature monitoring. More specifically, based on the degraded operating value Γ degrade α classifies the faults and obtains:

[0165] Mild fault: only the inverter peak current i max Reduce 5% to 10% to reduce IGBT switching frequency fluctuations;

[0166] Moderate fault: switch to the backup sensor channel or enable the prediction value substitution and set the bus voltage v dc The upper limit is lowered by a certain percentage to protect the main components;

[0167] Severe fault: If a critical component is detected to be endangering safety, an emergency safety mode can be triggered, significantly reducing power output and prompting the driver to limit speed or repair the vehicle as soon as possible;

[0168] At the same time, by recording the self-repair process (such as failure time, sensor replacement rate, aging characteristics, etc.), new abnormal data samples are incorporated into the machine learning model when updating, continuously improving the learning and early warning capabilities of similar failure scenarios;

[0169] When used, by degrading the operating value Γ degrade (t) Automatically adjust the operating mode to maintain the operation of the vehicle's core functions and avoid secondary damage even if some sensors or power devices fail. The degradation strategy is linked with inverter adaptive control, BMS battery management, and vehicle thermal management to ensure that single-point failures do not quickly evolve into systemic failures. Abnormal samples collected during the fault period will feed back into the machine learning prediction model and sensor calibration model, gradually accumulating fault characteristics and repair experience to build a more complete closed loop.

[0170] A dual mechanism of self-repair and degraded operation is proposed. In the case of minor faults, predictions or redundant data are prioritized for correction, avoiding unnecessary overall power reduction. In the case of severe faults, vehicle-wide thermal management and inverter power reduction are coordinated to ensure both safety and sustainable operation. Through fault identification and anomaly localization in sub-step 501 and the self-repair and degraded operation strategy in sub-step 502, timely responses can be made to sensor drift, power device aging, or sudden faults under extreme temperature conditions. Early warning and precise localization are achieved through observer and residual fusion methods. Degraded operation or backup data channels are used to maintain basic inverter and vehicle functionality during faults. Fault signature data is fed back to the electrical-thermal coupling prediction model in step 2 and vehicle-wide thermal management in step 4, enabling continuous improvement of fault handling experience in subsequent operations. This, combined with the previous four steps, enables vehicles to maintain a balanced balance of thermal safety, system performance, and fault tolerance in extreme environments, further enhancing the robustness and practical value of the entire solution.

[0171] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0172] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0173] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is only for some logical functions. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0174] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0175] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. An intelligent inverter control method based on an adaptive algorithm, characterized by: include, When the ambient temperature is detected to be significantly outside the preset range, the multi-source sensor readings, vehicle operating parameters and external meteorological data are integrated and kernel function interpolation is used to generate a multi-source fusion temperature distribution field; After collecting the multi-source fusion temperature distribution field and vehicle operating conditions, the electric-thermal coupling prediction model is called and error constraints are introduced to generate prediction quantities including at least the predicted temperature distribution field and the predicted heat load value; If it is detected that the predicted temperature distribution field is about to exceed the temperature margin, the inverter control parameters are adaptively scheduled to actively limit the amplitude and reduce power loss before the critical temperature is reached, maintaining power output and device safety; When the inverter output power and actual temperature distribution are transmitted back to the vehicle thermal management module in real time, the BMS-side temperature is compared with the heat dissipation information to predict the temperature distribution field and calibrate the electric-thermal coupling model. The fan or liquid cooling speed is then dynamically scheduled and the correction data is fed back to the previous step. When the monitored sensor value deviates seriously from the predicted temperature distribution field, the fault type is determined and the metric function is calculated, and then the degraded operating value is initiated or the redundant signal is enabled to self-repair the power device or temperature measurement failure.

2. The inverter intelligent control method based on the adaptive algorithm according to claim 1 is characterized in that: Deploy a multi-point temperature sensing network in the vehicle power system, record and mark the readings of each sensor; According to the spatial position of each measuring point, its measurement results are mapped to the original temperature field. Based on the external environment information obtained during vehicle operation, the external temperature curve is collected and the environmental impact coefficient is constructed.

3. The inverter intelligent control method based on the adaptive algorithm according to claim 2, characterized in that: The environmental impact coefficient is coupled with the original temperature field to obtain the multi-source fusion temperature distribution field after environmental correction. The multi-source fusion temperature distribution field after environmental correction and the external environment identification results are stored in the real-time database and interact with the historical operating condition database.

4. The inverter intelligent control method based on the adaptive algorithm according to claim 2, characterized in that: Synchronize the acquired multi-source fusion temperature distribution field, external environment recognition results, vehicle operating status, and related records in the historical operating condition database to the feature construction module; A time fusion method with an exponential decay kernel is used for feature extraction to generate comprehensive feature items for subsequent models. The sequence of comprehensive feature items between two historical moments is used in conjunction with the target value to construct the feature-label pairs required for supervised training.

5. The inverter intelligent control method based on the adaptive algorithm according to claim 4 is characterized in that: Based on the thermal physical characteristics of electric vehicle inverters and motors under extreme temperatures, a hybrid electro-thermal coupling model is introduced into the machine learning framework as input, with comprehensive feature terms. The model outputs the predicted temperature distribution field, heat load, and magnetic flux attenuation at future moments. A loss function that integrates physical constraints is introduced to train the electro-thermal coupling prediction model. The prediction effect is evaluated through offline verification and online update modes, and periodic fine-tuning training is performed. After the electro-thermal coupling prediction model is evaluated as qualified, its prediction output is passed to the inverter control parameter optimization module.

6. The inverter intelligent control method based on the adaptive algorithm according to claim 5, characterized in that: The predicted output is compared with the designed temperature safety upper limit and flux lower limit to form temperature margin and flux margin. The vehicle's driving intentions for the current and next period are collected, and the control parameter set of the existing inverter control is extracted. If the risk of overheating or insufficient flux is predicted, the corresponding power limiting strategy is selected based on the temperature margin and flux margin. The dynamic power limit value is combined with the original control parameter set to obtain a parameter mapping with upper and lower limit constraints. If temperature or flux risks are detected, the control parameters are dynamically limited or adjusted according to the dynamic power limit value.

7. The inverter intelligent control method based on the adaptive algorithm according to claim 6, characterized in that: Based on the determined parameter mapping, the inverter controller uses a built-in adaptive algorithm to solve the control parameters online and obtain the optimal control parameters that are ultimately applied to the execution end. The predicted output is incorporated into the objective function, and the control parameters are optimized for future time periods to obtain the optimal control parameters that change over time. A dynamic control strategy is constructed and finally executed in combination with parameter mapping. If the actual temperature or flux demagnetization rate is still deviated from the expected value, the online prediction update mechanism is automatically called in the next control cycle for online fine-tuning or incremental training to correct the input of the electro-thermal coupling prediction model and the upper limit of the control parameters.

8. The inverter intelligent control method based on the adaptive algorithm according to claim 7, characterized in that: Collect the actual temperature distribution field and power flow information of the inverter after the dynamic control strategy is actually implemented, as well as the updated optimal control parameters. Compare the inverter-side data with the predicted temperature distribution field, and evaluate the deviation between the prediction and reality through the control error term. If the control error term is greater than expected, the machine learning model or sensor calibration is corrected; The optimal control parameters, actual temperature distribution field, and control error terms are passed to the vehicle thermal management module. At the same time, the battery status and operating status of the cooling system are obtained from the BMS side and marked as battery management system status information. After alignment, a multi-terminal integrated data packet is formed.

9. The inverter intelligent control method based on the adaptive algorithm according to claim 8, characterized in that: After obtaining the merged multi-terminal integrated data package, the vehicle thermal management module and BMS work together to perform a global scheduling; After calculating the optimal solution of the introduced global objective function, the vehicle thermal management module sends scheduling instructions to the cooling system and guides battery preheating or balancing operations on the battery management system (BMS).

10. The inverter intelligent control method based on adaptive algorithm according to claim 9, characterized in that: Feedback of the actual action generated by the final execution of the scheduling instructions into real-time sensor fusion parameter correction and environment recognition update, as well as input correction and fine-tuning of the electro-thermal coupling prediction model; If it is found that the actual temperature response after the adaptive control strategy is implemented deviates significantly from the prediction, the scheduling strategy is dynamically adjusted, the global objective function is re-evaluated and local corrections are made.

11. The inverter intelligent control method based on adaptive algorithm according to claim 10, characterized in that: After obtaining the actual temperature distribution field and other key sensor readings and the corresponding predicted values, a metric function based on observer and residual fusion is defined. When the metric function increases sharply in a short period of time and exceeds the threshold, the specific fault type of the detected suspicious signal is determined by multidimensional residual pattern matching.

12. The inverter intelligent control method based on adaptive algorithm according to claim 10, characterized in that: Degradation values ​​indicating the extent of partial derating or backup parameter activation are defined based on the identified fault type and abnormality level; When the fault is serious, the degraded operation value increases. If the impact of the fault is limited, only minor temporary replacement processing is performed. For sensor drift or local failure, redundant sensor data or correction values ​​based on predicted output are temporarily enabled as alternative inputs. When the degraded operation value exceeds the preset warning line, degraded operation is automatically executed.

Images