Test method and test system of electric vehicle thermal management system

Abstract

The application belongs to the technical field of test of electric vehicle thermal management system, and discloses a test method and a test system of the electric vehicle thermal management system.The method comprises the following steps: obtaining time sequence data of thermal sensitive modules such as battery packs, inverters and motors of the electric vehicle under the condition of a set thermal load change; then, a graph structure is constructed with the thermal sensitive modules as nodes and thermal influence relations as edges; a thermal coupling relation is generated by using a graph neural network based on message passing; a dynamic Bayesian network is constructed based on the time sequence data and the thermal coupling relation; a temperature response time dependence relation among the modules is obtained; then, a thermal hysteresis distribution graph is generated to represent the thermal response delay relation and its intensity among the modules; finally, the response coordination and the adaptability of the thermal management control strategy under the test of the extreme thermal load condition are evaluated based on the thermal hysteresis distribution graph.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of test of electric vehicle thermal management system, in particular to a test method and test system of electric vehicle thermal management system, electronic equipment, medium and computer program product. BACKGROUND

[0002] In the thermal management system of an electric vehicle, multiple subsystems (or thermal sensitive modules, such as batteries, motors, inverters, on-board chargers, etc.) share heat dissipation resources, such as cooling liquid circulation, heat dissipation air ducts, etc. These modules have different materials (different heat capacity and thermal conductivity), different power dynamics (for example, the temperature of the motor rises sharply while the temperature of the battery rises smoothly), different layouts (space position affects thermal coupling), and different control strategies (such as battery priority cooling strategy), etc., resulting in different temperature change speeds, peak response times, and falling times of each module under the same thermal disturbance condition (such as power impact during acceleration), which is called thermal lag heterogeneity.

[0003] Traditional test methods generally only use temperature sensors to record single-point data; or perform linear fitting or empirical analysis based on time-temperature curves; or cannot depict the dynamic dependence relationship between modules, especially in complex structures with thermal coupling (thermal conduction / thermal convection influence), which is prone to misjudgment of the root cause of response delay.

[0004] Therefore, there is an urgent need for a test method to improve the response coordination and adaptability of the thermal management system of an electric vehicle. SUMMARY

[0005] To this end, the present application provides a test method and test system of an electric vehicle thermal management system, electronic equipment, medium and computer program product to at least partially solve the above technical problems.

[0006] The present application provides a test method of an electric vehicle thermal management system, comprising the following method steps:

[0007] Obtaining time series data of multiple thermal sensitive modules of an electric vehicle under a set thermal load change condition, the thermal sensitive modules at least including a battery pack, an inverter and a motor;

[0008] Constructing a graph structure with each thermal sensitive module as a node and the thermal influence relationship between the thermal sensitive modules as an edge, and generating a thermal coupling relationship based on the graph structure by using a message passing-based graph neural network;

[0009] Based on the time series data and the thermal coupling relationship, constructing a dynamic Bayesian network to obtain a temperature response time dependence relationship between the thermal sensitive modules;

[0010] generate a thermal hysteresis distribution map based on the time series data, the thermal coupling relationship and the dependency relationship, for representing thermal response delay relationships and their strengths among the plurality of modules;

[0011] evaluate response coordination and fitness of a thermal management control strategy under extreme thermal load conditions based on the thermal hysteresis distribution map.

[0012] In another aspect, the present application also provides a test system for an electric vehicle thermal management system, comprising:

[0013] a first obtaining module configured to obtain time series data of a plurality of thermal sensitive modules of an electric vehicle under a set thermal load change condition, the thermal sensitive modules at least including a battery pack, an inverter and a motor;

[0014] a first generating module configured to construct a graph structure with each of the thermal sensitive modules as a node and a thermal influence relationship among the thermal sensitive modules as an edge, and generate a thermal coupling relationship based on the graph structure by using a message passing based graph neural network;

[0015] a second obtaining module configured to construct a dynamic Bayesian network based on the time series data and the thermal coupling relationship to obtain a temperature response time dependency relationship among the thermal sensitive modules;

[0016] a second generating module configured to generate a thermal hysteresis distribution map based on the time series data, the thermal coupling relationship and the dependency relationship, for representing thermal response delay relationships and their strengths among the plurality of modules;

[0017] an evaluating module configured to evaluate response coordination and fitness of a thermal management control strategy under extreme thermal load conditions based on the thermal hysteresis distribution map.

[0018] In another aspect, the present application also provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the test method for an electric vehicle thermal management system as described above.

[0019] In another aspect, the present application provides a computer readable storage medium having stored thereon computer program instructions executable by a processor to implement the test method for an electric vehicle thermal management system as described above.

[0020] In another aspect, the present application provides a computer program product comprising a computer program executable by a processor to implement the test method for an electric vehicle thermal management system as described above.

[0021] Through the scheme provided by the embodiment of the application, the time sequence data of multiple thermal sensitive modules is collected, a graph structure is constructed, a GNN model is generated, a thermal coupling relationship is generated, a temperature response time dependence is obtained by combining DBN, a thermal hysteresis distribution graph is generated, the thermal response delay relationship and intensity between modules can be accurately reflected, based on the thermal hysteresis distribution graph, the response coordination and adaptability of the thermal management control strategy under extreme thermal load conditions can be effectively evaluated; the structure gating mechanism is introduced, the message passing round number and time window are dynamically adjusted, and the like, so that the model can be dynamically adjusted according to the actual working condition, the change of the thermal influence path under different vehicle arrangements and working conditions is adapted, such as activating the efficient heat transfer path under high-speed working conditions, weakening the non-key edges under low load, improving the generalization ability of the topological model, and thus improving the accuracy of the test; through sparse structure learning, multi-index fusion and screening of prior dependence structure, the invalid edges are pruned by combining L1 penalty, the model structure is simplified, the key thermal dependence is focused, the thermal transfer path analysis is more accurate, the invalid edge interference is reduced, and the test efficiency and accuracy are improved. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some of the embodiments of the application, and therefore should not be regarded as a limitation to the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0023] Other features, objects and advantages of the application will become more apparent from the following detailed description of the non-limiting embodiments, made with reference to the accompanying drawings:

[0024] Figure 1 A schematic diagram of a test method of an electric vehicle thermal management system provided by the embodiment of the application.

[0025] Figure 2 A topological schematic diagram of DBN provided by the embodiment of the application.

[0026] Figure 3 A structure schematic diagram of a thermal hysteresis distribution graph provided by the embodiment of the application.

[0027] Figure 4 A structure schematic diagram of a test system of an electric vehicle thermal management system provided by the embodiment of the application.

[0028] Figure 5 A structure schematic diagram of an electronic device provided by the embodiment of the application. DETAILED DESCRIPTION

[0029] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application.

[0030] In a typical configuration of the present application, the terminal and the device of the service network each include one or more processors (CPU), input / output interfaces, network interfaces and memories.

[0031] The memory can include a non-permanent memory in a computer readable medium, random access memory (RAM) and / or non-volatile memory such as read only memory (ROM) or flash memory (flash RAM). The memory is an example of the computer readable medium.

[0032] The computer readable medium includes permanent and non-permanent, removable and non-removable media, and can be implemented by any method or technology to store information. The information can be computer program instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read only memory (ROM), electrically erasable programmable read only memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic tape disk storage or other magnetic storage device, or any other non-transmission medium that can be used to store information accessible to a computing device.

[0033] As shown in Figure 1 The present application discloses a test method for an electric vehicle thermal management system, which comprises the following steps:

[0034] Step S100, obtaining time sequence data of a plurality of thermal sensitive modules of an electric vehicle under a set thermal load change condition, wherein the thermal sensitive modules at least include a battery pack, an inverter and a motor.

[0035] In some embodiments, the multiple core thermal sensitive modules of the electric vehicle have different thermal capacities, cooling paths and physical layouts, and there are differences in the thermal response characteristics of the power load. For example: the motor rapidly heats up under high torque output; the inverter instantaneously generates a lot of heat under frequent commutation or PWM modulation; the battery heats up more slowly and lags behind the power change. In order to detect the thermal response lag phenomenon of these modules in real operation, representative thermal disturbances (i.e. thermal load change conditions) are applied in controlled tests, and the responses are recorded.

[0036] Specifically, the thermal load change condition refers to causing the temperature of the thermal sensitive module to rise by setting the load input curve (such as current, voltage, torque) of the power system, and inducing the thermal management system to respond. Common types include: step load (Step Input), such as sudden acceleration, start-up impact, motor current or system power suddenly increases in a short time; PWM modulation load, such as frequent acceleration and deceleration under urban working conditions, high frequency, periodic power fluctuation; continuous high load, such as high-speed cruising, long slope climbing, continuous stable high output (high power / current); sweep load, the load frequency gradually changes, simulating the full spectrum load disturbance of electromagnetic response; positive and negative load switching, rapid positive torque / negative torque conversion, thermal impact of the process of sudden acceleration→braking, etc.

[0037] Exemplarily, taking the sudden acceleration step load as an example, simulating the thermal load mutation in high-speed start-up or lane change acceleration, suddenly applying 100% motor torque output in the vehicle static state; the current jumps from 0A to 250A and lasts for 45 seconds; the temperature, current, voltage, etc. of the battery, inverter and motor are recorded, and the typical data trend is shown in the following table:

[0038] Time (s) Current (A) Motor temperature (°C) Inverter temperature (°C) Battery temperature (°C) 0 0 25.0 25.0 25.1 5 250 27.8 30.0 25.3 20 250 36.5 42.0 26.5 45 250 43.0 49.0 28.2 90 0 41.0 46.5 29.0

[0039] As can be seen from the data in the above table, the motor and the inverter respond quickly and have small thermal inertia; while the battery heats up significantly lags behind.

[0040] Step S200, constructing a graph structure with each thermal sensitive module as a node and the thermal influence relationship between the thermal sensitive modules as an edge, and based on the graph structure, using a graph neural network to model the graph structure to generate a thermal coupling relationship.

[0041] In some embodiments, in the thermal management system of the electric vehicle, there are thermal conduction paths (such as through structural members, heat sinks) between different modules (such as motors, inverters, battery packs, DC / DC, etc.); thermal convection paths (such as cooling liquid flow sequence); indirect thermal interference (such as sharing heat exchangers or heat sinks) and other thermal influence relationships. This embodiment is based on the thermal influence relationship to construct a graph structure with each thermal sensitive module as a node and the thermal influence relationship between the thermal sensitive modules as an edge; exemplarily, the specific composition of the graph structure is as follows:

[0042] Node: Each thermal module (e.g., battery, motor);

[0043] Edge: The thermal coupling relationship between modules (e.g., coolant flows through sequentially, sharing structural thermal paths);

[0044] Node characteristics: Temperature, power, current, voltage, coolant temperature difference, etc. of each module;

[0045] Edge characteristics: coupling strength, cooling flow rate, physical connection method, heat conduction efficiency, etc.

[0046] In some embodiments, thermal coupling in electric vehicle thermal management is nonlinear and dynamic. For example, changes in motor temperature can affect the inverter through the liquid cooling system; battery temperature rise can delay the thermal load of the entire cooling circuit; and changes in the cooling strategy at a certain flow point can indirectly affect the temperature control performance of other modules. Based on these complex relationships of local dependence and global propagation, this embodiment uses a message-passing-based graph neural network model (MP-GNN) to model the coupling effect. Each node passes "thermal state messages" to its neighbors in each iteration and integrates the influence of its neighbors on itself.

[0047] As an example, taking a high-power long hill climb test lasting 600 seconds as an example, the thermal module may include a battery pack (BATT), a motor (MOT), an inverter (INV), a motor controller (MCU), and a cooler (CHX); the coolant flow sequence between modules is: BATT→INV→MCU→MOT→CHX; in addition, the inverter and motor are connected by an aluminum alloy bracket, which creates a structural heat conduction path.

[0048] Based on the aforementioned graph structure, the example graph structure is constructed as follows:

[0049] Nodes (5): BATT, INV, MCU, MOT, CHX

[0050] Edges (including directionality):

[0051] Coolant path: BATT→INV→MCU→MOT→CHX

[0052] Structural thermal coupling edge: (Two-way)

[0053] The input features for each node include: current temperature; current power input / output; coolant inlet / outlet temperature difference; ambient temperature; and the temperature rise rate at the previous time step. The features for each edge include: coolant flow rate (used for heat transfer estimation); path thermal resistance (e.g., ...). Structural connection thermal resistance; thermal delay estimate (from historical data playback).

[0054] Specifically, the MP-GNN structure is briefly described as follows:

[0055] The input layer is a graph structure used to receive the income, including nodes and edges; the temporal characteristics of each node / edge;

[0056] The message passing layer (K ​​layer) is used to perform K rounds of message passing and updating in the graph, simulating hot coupling;

[0057] The aggregation and update mechanism is used by each node to receive thermal state information from its neighbors and perform aggregation, nonlinear transformation, and state update.

[0058] The output layer is used to predict the thermal state (temperature or rate of temperature rise) of each module node.

[0059] For example, the input layer receives an input graph structure, including a graph G = (V, E), wherein, for example, the node set in the graph may include V = {BATT, INV, MCU, MOT, CHX, ...}; the edge set E includes: cooling path edges: BATT→INV→MCU→MOT→CHX (directed edge); structural thermal coupling edges: (Bidirectional edges), etc., the nodes and edges in the graph structure can be set according to the actual situation, and this invention does not impose any restrictions.

[0060] The feature vector Xt of each node (v) Taking an inverter (INV) as an example, for instance, a dimension d is used. n The vector representation of this is denoted as: Specific feature types may include the current thermal state, represented by the current temperature (°C) and the rate of temperature change; electrical parameters, represented by the current power, current, voltage, load rate, power flow intensity, etc.; thermal sensing information, represented by the temperature difference between the inlet and outlet of the coolant; time correlation, represented by the temperature rise rate (ΔT / Δt) at the previous moment; environmental influence, represented by the external ambient temperature; and cooling information, represented by the cooling flow rate. It is understood that the feature type of each node can be set according to the actual situation, and this invention does not impose any restrictions.

[0061] The feature vector Xe of each edge (e) Taking INV→MCU as an example, for instance, a dimension d is used. e The vector representation of this is denoted as: Specific feature types may include cooling path information, represented by coolant flow rate, flow direction, and connecting pipe length; thermal resistance estimation, represented by thermal coupling conduction efficiency; topology information, represented by whether it is a direct structural connection or a serial / parallel cooling flow path; distance coefficient, represented by the physical distance between coolers in modules; coupling confidence, etc. It can be understood that the feature type of each edge can be set according to the actual situation, and this invention does not impose any restrictions.

[0062] Based on the received data, the message passing layer of the MP-GNN iterates K times, each time simulating one hot coupling propagation cycle. For example, the process is as follows:

[0063] 1. Message Construction

[0064] For each edge (u→v), constructing the message passed from node u to node v can be simplified to the following formula:

[0065]

[0066] This means that in a graph structure, for each directed edge u→v, using the information of the current time t, a message is calculated from node u to node v. This represents the feature vector of the sending node u, such as the current temperature, load rate, and temperature change rate of the MOT. This represents the feature vector of the receiving node v, for example, INV's current temperature, flow rate, and thermal conductivity. The feature vectors representing the edges from u to v include, for example, the coolant flow rate, connecting pipe length, thermal resistance, and edge coupling confidence between u and v. MLP1 is a multilayer perceptron (multilayer neural network) that nonlinearly combines these features to generate an information-rich thermal impact message.

[0067] Taking the edge MOT→INV (drive motor→inverter) as an example; node u = MOT (sender), feature vector Includes: Current temperature: 61.3℃, temperature change rate: +1.2℃ / s, motor load rate: 95%, local cooling flow rate: 0.6L / min; node v = INV (receiver);

[0068] Feature vector Features include: current temperature: 42.1℃, temperature change rate: +0.5℃ / s, cooling flow rate: 0.5L / min, power flow intensity: high;

[0069] Edge features The parameters include: coolant flow rate through this side: 0.55 L / min; connecting pipe length: 1.2 m; estimated thermal resistance: 0.03 K / W; and initial coupling confidence level: 0.85.

[0070] In MP-GNN, the message computation module for this edge (as described in the formula above) concatenates these features together and computes a vector using the neural network. This vector can include heat conduction trend (positive / negative), coupling direction confidence, thermal impact priority, etc.; it is used to represent the intensity of the thermal coupling effect of MOT on INV under the current thermal state.

[0071] In thermal management systems, motors heat up rapidly and preferentially transfer heat to the inverter via liquid cooling pipes or metal structures. The strength of this behavior is not linear and is influenced by factors such as load, cooling status, and path parameters. This message construction mechanism learns the strength of these complex, multi-factor nonlinear thermal influence paths and iteratively propagates them within a graph structure to ultimately form a thermally coupled topology graph.

[0072] 2. Message Aggregation

[0073] For each node v, messages are received from all its neighboring nodes N(v) and aggregated. The aggregated value represents the combined thermal effect of all neighbors on a given node at time t.

[0074] The target node is v = INV (inverter). Assume that in the current graph structure, INV has three neighbor nodes: MOT (drive motor), BATT (battery pack), and CHX (cooler). Each neighbor node u generates a message vector to send to INV through the message construction step in the previous stage. As mentioned earlier, the numerical values ​​of the vectors represent their respective impact on the current thermal state of the INV (which may include dimensions such as heat flux, thermal conductivity, thermal disturbance caused by the load, and risk level). For example, all messages of the INV node and their message vector values ​​are shown in the table below:

[0075] Neighbor u Message vector (simplified to a single-dimensional value, e.g. heat impact strength) MOT→INV 0.92 BATT→INV 0.76 CHX→INV 0.31

[0076] The message vector value can be calculated according to the above embodiments. The aggregation method can include summation (SUM), mean (MEAN), maximum value (MAX), or weighted summation, etc., and the present invention does not limit it.

[0077] Taking the mean value as an example, the heat-affected zone intensity value of polymerization,

[0078] 3. Node Status Update

[0079] Updating the node's own thermal state representation can be simplified to the following formula:

[0080]

[0081] This represents the thermal state vector of node v at time step t+1. Based on its current state Aggregate messages with neighbors Together, we determine that the update function is a GRU (Gated Recurrent Unit), but it can also be an LSTM or an MLP.

[0082] Taking the INV module as an example, where, This represents the thermal state vector of node v at time t, such as INV current temperature, cooling rate, current load status, temperature change trend, etc. This represents aggregated messages from neighboring nodes, such as the heat-affected aggregation results from MOT, BATT, and CHX; GRU() represents the state update unit, which is used to combine past states with external inputs to generate new states.

[0083] As an example, GRU is a gated recurrent neural network that uses historical states (the current thermal state of the INV) and "aggregated messages" about heat conduction from neighbors to determine the thermal risk trend of the INV in the next moment, such as whether it will heat up. For example, the GRU process consists of three gating mechanisms:

[0084] 1. Update the door

[0085] The update gate determines how much of the current state to retain (whether to continue the existing temperature rise trend). For example, if the current temperature rise is not significant (+0.5℃ / s) and the neighbor's thermal influence is not severe (0.663), the update gate tends to be conservative, meaning the state change is minimal. If the temperature rise is faster (e.g., +1.5℃ / s) or the neighbor's message is stronger (e.g., >0.9), the update gate weight increases, indicating that INV needs to prepare for drastic temperature changes.

[0086] 2. Reset the door

[0087] The system controls whether to "forget" old information and prioritizes external input. For example, if the current cooling is already strong (coolant flow rate Q = 0.5 L / min), but the temperature continues to rise, the reset gate will increase its value, prompting the model to abandon the current judgment and consider readjustment. If the external thermal influence is weak, the reset gate will decrease, indicating that historical judgments can still be referenced.

[0088] 3. New candidate status

[0089] Current thermal state Impact on neighbors The predicted thermal state change trend at the next moment is calculated by fusion. This generates the thermal coupling relationship of each edge, which can be exemplarily represented as the intensity of thermal influence between modules learned through a graph neural network.

[0090] Based on the above process, after several rounds of iteration (e.g., 3 rounds), the representation vector of each node stabilizes, containing the thermal response features of itself and its neighbors. By automatically learning the complex thermal coupling between modules through a graph neural network, without relying on manually creating the topology graph structure, it is understood that the number of iteration rounds can be set according to the actual situation, and this invention does not impose any limitations.

[0091] For example, based on the above process, the thermal coupling strength output by the GNN model is shown in the table below:

[0092]

[0093]

[0094] For example, based on the above process, the GNN model can also output a thermal impact path ranking to explain the main thermal response transfer path, ranked as follows: MOT→INV→BATT→CHX (main thermal propagation path, response chain length), MOT→CHX (secondary path), ACU→BATT (auxiliary path). Specifically, the main thermal propagation path contains multiple modules, and heat transfer requires multiple couplings, while also reflecting the lag step length. For example, after a sudden increase in motor power, its heat is first transferred to the inverter through structural heat conduction or coolant, the inverter then affects the battery through the cooling loop, and finally the heat is dissipated through the cooler, forming a relatively long thermal response chain. For the secondary path, the motor heat does not pass through the inverter and battery, but is directly transferred to the cooler through coolant circulation, the path is shorter, and the lag time is shorter; for the auxiliary path, the coupling strength of ACU→BATT is about 0.15, indicating that the thermal impact of the air conditioning controller on the battery is weak, belonging to a secondary path.

[0095] It is understood that the GNN model can output both coupling strength and thermal impact path ranking simultaneously, or it can output only one of the two; this invention does not impose any restrictions.

[0096] Step S300: Based on the time series data and the coupling relationship, construct a dynamic Bayesian network to model the temperature response time dependency between each of the thermal modules.

[0097] In some embodiments, the input data sources of Dynamic Bayesian Network (DBN) and their roles in DBN are shown in the following table:

[0098]

[0099] Taking the modules MOT (motor), INV (inverter), and BATT (battery pack) as an example, the construction process of the dynamic Bayesian network is as follows:

[0100] 1. Define node variables

[0101] Based on time-series data, the following variable sequence is constructed to describe the temperature state of different modules in continuous time slices:

[0102]

[0103] ...

[0104]

[0105] Expanding over time, each node can be set as a discrete variable (e.g., low temperature, medium temperature, high temperature) or a continuous variable (e.g., real value in °C).

[0106] 2. Add structural edges (causal edges between modules)

[0107] Based on the coupling path and coupling strength provided by GNN, determine whether to connect the following dependencies:

[0108]

[0109] At this point, the DBN topology is as follows: Figure 2 As shown, the time slices are divided into three time slices, representing the module temperature status at times t, t+1, and t+2, respectively.

[0110] The motor temperature at time t is taken as the starting point of heat transfer.

[0111] The inverter temperature at time t+1 is delayed by one time step due to the influence of motor heat.

[0112] The battery temperature at time t+2 is delayed by two time steps due to the thermal effect of the inverter.

[0113] Solid arrows represent heat transfer paths, reflecting the thermal response chain sequence of "motor → inverter → battery" and the time dependence of temperature response between modules.

[0114] It should be noted that DBN outputs differ from GNN outputs in that their edge directions must explicitly reflect temporal dependencies.

[0115] 3. Conditional Probability (CPT) of Quantized Edges

[0116] For each node, for example Define its dependency variables, for example The following conditional probability:

[0117] Assume the temperature state is discretized as follows:

[0118] L: Low temperature (<40℃)

[0119] M: Medium temperature (40–60℃)

[0120] H: High temperature (>60℃)

[0121] Example CPT (Conditional Probability Table):

[0122] For edge MOT→INV+1:

[0123]

[0124] The table above shows that if the motor temperature is high (H), the probability of the inverter experiencing high temperature in the next moment increases significantly.

[0125] As an example, consider the following time-series data and thermally coupled path sorting.

[0126] Time series data:

[0127] Time t (s) MOT (°C) INV (°C) BATT (°C) 0 36.2 36.1 35.8 5 49.6 38.4 36.2 10 57.9 44.1 37.8 15 61.5 50.8 40.5

[0128] Thermal coupling path sorting:

[0129] MOT→INV→BATT

[0130] Lag relationship:

[0131] MOT to INV: 1 time step

[0132] INV vs. BATT: 2 time steps

[0133] The constructed DBN network topology reflecting the temperature response time dependency:

[0134]

[0135] Step S400: Generate a thermal hysteresis distribution map based on the time series data, the thermal coupling relationship, and the dependency relationship to represent the thermal response delay relationship and its intensity among multiple modules.

[0136] In some embodiments, the thermal hysteresis distribution diagram can be understood as a graphical structure showing the time delay (hysteresis) relationship and intensity between various thermally sensitive modules in the thermal management system of an electric vehicle. Specific display content includes:

[0137] Nodes: Thermal modules (such as BATT, MOT, INV, etc.);

[0138] Directed edge: heat-affected path;

[0139] Edge weight:

[0140] Lag time Δt (unit: time step);

[0141] Influence strength G (derived from MP-GNN inference);

[0142] The lag probability P (derived from DBN conditional probability).

[0143] For example, the generation process includes:

[0144] 1. Obtaining thermal coupling relationships (from MP-GNN)

[0145] Based on the MP-GNN model, the following results were obtained:

[0146] Module pair G (coupling strength) Δt (lag time) MOT→INV 0.83 1 INV→BATT 0.77 2 MOT→BATT 0.15 2 (weak coupling)

[0147] For example, only paths with coupling strength exceeding a threshold (e.g., 0.3) are retained.

[0148] 2. Obtaining conditional probability dependencies (from DBN)

[0149] For example, based on DBN inference, the following conditional dependencies are obtained:

[0150]

[0151] (exclude)

[0152] 3. Construct thermal hysteresis boundary sets

[0153] By combining the two sources mentioned above, we obtain the edge set that is formally used for graph drawing, as shown in the table below:

[0154] Start point End point Lag step Δt Coupling strength G Lag probability P MOT INV 1 0.83 0.88 INV BATT 2 0.77 0.81

[0155] The structure of the thermal hysteresis distribution map is exemplarily as follows: Figure 3 As shown,

[0156] Node representation: The rectangles represent the module temperature status at times t, t+1, and t+2, respectively.

[0157] The meaning of the edge:

[0158] The solid arrows represent the main thermal influence path, including parameters such as lag time (Δt), thermal coupling strength (G), and lag probability (P). Specifically, Δt = 1 (lag by 1 time step), G = 0.83 (strong coupling), P = 0.88 (high response probability); Δt = 1 (lags by 1 time step, cumulatively 2 time steps), G = 0.77 (strong coupling), P = 0.81 (high response probability).

[0159] Dashed arrow + X: Indicates an ignored path. Δt = 2 (lagging by 2 time steps), G = 0.15 (weak coupling), P = 0.22 (low response probability) were excluded because the parameters were below the threshold.

[0160] Step S500: Evaluate the response coordination and adaptability of the thermal management control strategy under extreme thermal load conditions based on the thermal hysteresis distribution map.

[0161] In some embodiments, the thermal hysteresis distribution map can quantify and analyze the control strategy's ability to regulate the heat transfer path under extreme conditions by visualizing the thermal response delay relationship between modules (hysteresis time Δt, coupling strength G, hysteresis probability P).

[0162] Specifically, the following can be evaluated:

[0163] Changes in main hot path lag time: Whether the total lag time of the main path (e.g., MOT→INV→BATT) is shortened reflects the optimization effect of the strategy on the response speed of key modules.

[0164] Coupling strength balance: Whether the coupling strength G of each path tends to be balanced, so as to avoid a certain module becoming a thermal bottleneck.

[0165] Hysteresis probability under extreme conditions: Whether the probability P of inter-module response remains high (e.g., P>0.8) under high load to ensure the stability of heat transfer.

[0166] As an example, taking extreme thermal load conditions: continuous ramping + high temperature environment (40℃), with the thermal management control strategy being a battery-priority cooling strategy, the test process based on the thermal hysteresis distribution map is as follows:

[0167] 1. Key parameters of thermal hysteresis distribution map:

[0168] Main path: MOT→INV→BATT

[0169] Δt_total = 1(MOT→INV) + 1(INV→BATT) = 2 time steps (approximately 25 seconds)

[0170] G = 0.83 × 0.77 = 0.64 (product of coupling strength)

[0171] P = 0.88 × 0.81 = 0.72 (product of response probabilities)

[0172] Ignored path: MOT→BATT (Δt=2, G=0.15, P=0.22, still ignored).

[0173] 2. Assessment of Response Coordination Issues:

[0174] Because the battery is cooled preferentially, the coolant flow rate is high, but the motor and inverter do not receive sufficient heat dissipation, resulting in:

[0175] The Δt of INV→BATT is extended to 2 time steps (total lag of 3 time steps, approximately 35 seconds), G decreases to 0.65, and P decreases to 0.70.

[0176] As the motor temperature continued to rise, although G for MOT→INV remained at 0.83, P dropped to 0.80 due to insufficient heat dissipation, resulting in decreased thermal transfer stability.

[0177] 3. Compatibility issues:

[0178] During continuous uphill climbing, the motor and inverter are at risk of overheating (temperature > 80℃) due to lagging heat dissipation, while the battery temperature is controlled below 40℃, resulting in an uneven distribution of heat dissipation resources.

[0179] By quantifying parameters such as Δt, G, and P, the effect of the control strategy on the heat transfer path can be intuitively reflected, helping to identify response bottlenecks under extreme conditions (such as the motor overheating problem during continuous ramping in the original strategy).

[0180] Optionally, the step of generating thermal coupling relationships using a message-passing-based graph neural network includes: obtaining the temperature change rate of the thermal module and cooling control state parameters; introducing a structural gating mechanism to design a dynamic weight gating function for each heat transfer edge between the thermal modules in the graph structure; dynamically adjusting the edge connection strength or retaining / removing the heat transfer edge based on the output of the gating function; if the thermal response is strong, retaining and strengthening the corresponding edge connection; if the cooling path is not shared, weakening or removing the edge connection.

[0181] In some embodiments, within the vehicle thermal management system testing scenario, the thermal impact paths between modules (e.g., MOT, INV, BATT, etc.) will change under different vehicle layouts (e.g., motors are located in the front compartment in some vehicles, while their positions differ due to space design) or operating conditions (e.g., high-speed driving, stop-and-go traffic). For example, at high speeds, motor heat may be directly and quickly transferred to the cooler, while in congested traffic, more heat transfer occurs between the motor, inverter, and battery. The preset static diagram structure (which fixes the connections between modules) cannot flexibly adapt to these changes.

[0182] To address the aforementioned issues, this embodiment introduces a structural attention gate mechanism. Simply put, it dynamically determines the connection strength of edges (heat transfer connections) between modules, and even whether to retain those edges, based on the vehicle's actual operating environment (temperature, road conditions, etc.) or collected temperature sequences. The specific workflow includes:

[0183] Input: Module temperature change rate ΔT i (t) (e.g., how many degrees the motor temperature rises per second), cooling control status (e.g., how fast the coolant pump is, how much the motor load is).

[0184] Dynamic weighted gating function: For each possible heat transfer edge between modules (e.g., from motor to inverter, denoted as i→j), design a function to calculate a weight α using the temperature change rate difference, pump speed, etc. ij (t).

[0185] If the temperature change rate difference between modules is |ΔT i (t)-ΔT j If (t) is large, it indicates a strong thermal response, so this edge is retained, allowing the heat transfer relationship to play a role in the graph neural network; if the cooling path is not shared (for example, the motor runs through one coolant pipe and the battery runs through another, with no intersection), the connection of the edge is weakened, because actual heat transfer is difficult to occur through this "path".

[0186] For example, under high-speed operating conditions, the rate of change of motor temperature ΔT MOT The rate of temperature change ΔT in the cooler (CHX) is very large. CHX When the two values ​​are relatively stable and the difference is large, the weight of the edge related to the motor to the cooler (such as the edge from the inverter to the cooler) is high, and the heat transfer relationship is strengthened.

[0187] The presence or absence of edges between modules and their initial weights are determined by the actual operating data, rather than being fixed at the beginning. This allows the graph neural network to better adapt to different situations (improving the generalization ability of the topology model). No matter how the vehicle layout changes or the operating conditions change, it can find a reasonable heat transfer path at the current time.

[0188] As an example, under high-speed operating conditions, the thermal impact of the motor (MOT) is rapidly transferred to the cooler (CHX). According to this mechanism, because the motor's temperature changes rapidly and the temperature difference between it and other modules like the cooler is large, the edge from the inverter (INV) to the cooler (CHX) in the graph neural network is activated. Originally, the main path is MOT→INV→BATT→CHX, but at high speeds, through structural gating, the weight of the INV→CHX edge increases, allowing heat transfer to take this "shortcut" more efficiently, which aligns with the actual requirement of prioritizing rapid heat dissipation from the motor to the cooler under high-speed conditions.

[0189] In this way, the adaptive adjustment mechanism of the thermal coupling structure allows the heat transfer path in the graph neural network to change flexibly according to different vehicle layouts and operating conditions, solving the problem of poor adaptability of static graph structures. This makes the testing of the thermal management system (such as thermal hysteresis and thermal response path) more closely match the actual vehicle operation and improves the test performance under extreme conditions.

[0190] Optionally, the step of generating thermal coupling relationships using a message-passing-based graph neural network further includes: calculating the state change amplitude of module nodes after each round of thermal signal message passing in the graph neural network, wherein the state change includes temperature change and thermal influence weight change; introducing a propagation termination gate and setting a state change threshold; if the state change amplitude of module nodes is less than the threshold after two consecutive rounds of thermal signal message passing, it is determined that the thermal signal propagation tends to be stable, and subsequent message passing is terminated in advance; and dynamically pruning the number of GNN message passing rounds based on the determination result of the propagation termination gate.

[0191] In some embodiments, the depth of thermal signal propagation is not fixed in the testing of a vehicle thermal management system based on a graph neural network (GNN). For example, when the battery overheats locally, the thermal impact may initially only propagate within the battery and adjacent modules (such as the inverter) before spreading further; while when the motor heats up under high load, the heat transfer path and range change over time. However, in traditional GNNs, the number of message passing rounds K is fixed (e.g., always passing through 3 rounds), which cannot flexibly adapt to such dynamic changes in the depth of thermal signal propagation. This can lead to problems such as insufficient transmission (failure to capture the full thermal impact) or excessive transmission (mistakenly transmitting the thermal impact to irrelevant modules).

[0192] To address the aforementioned issues, this embodiment introduces a propagation stop gate, enabling the GNN to determine when to stop message passing and dynamically adjust the K value. The specific workflow includes:

[0193] After each round of message passing (e.g., one round of thermal information exchange between modules in a GNN), the magnitude of the state change of the evaluation node (thermal-sensitive module, such as battery, motor, etc.) is assessed. This state change can be a change in module temperature, a change in the thermal impact weight, etc.

[0194] At the same time, a threshold is set (for example, the temperature change is less than 0.5℃ for two consecutive rounds). If the node state change is less than this threshold after two consecutive rounds of message transmission, it is considered that the heat signal has "stabilized" at the current depth and there is no need to continue transmission. The transmission is terminated in advance, and the K value is dynamically trimmed (for example, if it was originally set to transmit for 5 rounds, it is now set to stop after 3 rounds).

[0195] As an example, in high-speed operating conditions where the motor (MOT) continuously generates heat, the heat signal propagates gradually under sustained high load:

[0196] In the first two rounds of message transmission, the motor's own state changes significantly (temperature rises), then the message is transmitted to the inverter (INV), where the state also changes significantly, so the transmission continues. In the third and fourth rounds, the heat signal travels from the inverter to the battery (BATT), where the battery's state changes noticeably but hasn't reached a stable threshold, so the transmission continues. Propagation ends in the fifth round when the state changes of the motor, inverter, and battery are less than 0.2℃ for two consecutive rounds (e.g., all stabilizing within a temperature change of 0.2℃), and the K value is dynamically adjusted (originally, a maximum of 6 rounds might have been set, but now it stops after 5 rounds). This ensures that the complete propagation path of the heat signal from the motor → inverter → battery is captured, but avoids excessive transmission to unrelated modules (such as the air conditioning controller ACU, which doesn't blindly transmit unless there is strong thermal coupling), adapting to the characteristic of heat signal propagation depth changing over time.

[0197] Optionally, the construction of a dynamic Bayesian network to model the temperature response time dependency among the thermal modules further includes obtaining the maximum correlation delay time based on the heat transfer path between the modules and the temperature sequence correlation between the modules, automatically configuring the time window length for each heat transfer path based on the maximum correlation delay time, and constructing the conditional probability distribution of the heat transfer path based on the time window length.

[0198] In some embodiments, during vehicle thermal management system testing, the thermal response lag times differ significantly between different thermally sensitive modules (BATT, CHX, MOT). For example, changes in battery temperature affecting the cooler may take several seconds to manifest; while cooler adjustments affecting motor temperature may occur very quickly. If a fixed time window is used (e.g., a 5-second time window is used for analysis regardless of the type of thermally sensitive module), it becomes impossible to consider both local (short lag, like from the cooler to the motor) and global (long lag, like from the battery to the cooler) thermal response, leading to inaccurate analysis.

[0199] To address the aforementioned issues, this embodiment assigns a time window that best reflects the heat transfer lag between different modules (i→j, e.g., BATT→CHX) based on the correlation of their thermal responses. This time window is then used as a dedicated time window to build a lag model.

[0200] Specifically, the module thermal response correlation function is set:

[0201]

[0202] Find the maximum relevant delay time τ ij That is, observe the temperature T of module i (e.g., battery) at time t. i (t), and the temperature T of module j (e.g., the cooler) at t+Δt. j (t+Δt), at which Δt is the correlation between the two highest? This Δt is τ. ij , is used to define the time window for the i→j edge.

[0203] Each heat transfer path (e.g., BATT→CHX, CHX→MOT) uses its own τ ij During the time window, the conditional probability distribution (CPD) in the dynamic Bayesian network (DBN) is constructed to make the model more closely reflect the actual heat transfer lag.

[0204] Optionally, the generation of thermal coupling relationships using a message-passing-based graph neural network further includes dynamically selecting edges using sparse structure learning combined with heuristic indicators. The heuristic indicators include at least one of the following: quantifying the physical coupling degree of the heat transfer path based on the physical connection characteristics between the thermal modules; or, statistically analyzing the correlation between the temperature and thermal response of the thermal modules in historical data; or, scoring the importance of thermally dependent edges through L1 penalty, and automatically pruning edges below a threshold.

[0205] In some embodiments, during the construction of a graph neural network, it is necessary to accurately identify the real hot dependencies between modules. This embodiment uses sparse structure learning (such as Sparsity-Inducing Priors, which makes the model tend to select key edges and prune non-key edges) and combines three types of heuristic indicators to dynamically determine the retention or removal of edges.

[0206] Specifically, heuristic metrics include physical path coupling metrics: these assess the closeness of physical connections between modules, such as the length of liquid cooling pipes (the shorter the pipe, the more direct the heat transfer, and the more likely the edge should be preserved). Heat transfer between heat-sensitive modules is affected by the physical connection method. Physical path coupling metrics are used to measure the tightness of the physical basis for heat transfer between modules. The core logic is: the more direct and closer the physical connection, the easier it is for heat transfer to occur, and the more likely the corresponding thermal dependencies (edges in the model) should be preserved. Shorter / more direct channels result in less heat loss and faster speed during heat transfer, leading to a significant thermal impact between modules; longer / more indirect channels result in more dispersed and delayed heat transfer, weakening the thermal impact between modules. For example, if a motor and inverter share a short liquid cooling pipe, the physical coupling is strong, and the corresponding edge should be preserved first.

[0207] Historical mutual information: This statistically analyzes the correlation between module temperature and thermal response in historical data (higher mutual information indicates more significant thermal dependence). Historical mutual information is a key indicator for uncovering thermal dependencies between modules. The core logic is: by statistically analyzing historical operating data, we can detect whether the temperature and thermal response changes of two modules are synchronously correlated—the stronger the correlation (higher mutual information), the more significant the thermal dependence, and the more important the corresponding heat transfer edge in the model should be.

[0208] Mutual information measures how much information two random variables (such as the temperature of module A and the temperature of module B) share. If the temperature of module A (e.g., a motor) changes, and the temperature and thermal response (e.g., the start-up of the cooling system, the rate of temperature rise) of module B (e.g., an inverter) also change accordingly, it indicates that there is an informational relationship between the two—the mutual information value is high. If module B does not react to changes in the temperature of module A, it indicates a weak relationship—the mutual information value is low.

[0209] For example, the analysis process of mutual information includes:

[0210] Extract 1000 sets of data on motor load changes, motor temperature, and inverter temperature from the thermal management system test data. For example, under conditions such as rapid acceleration, hill climbing, and high-speed cruising, the temperature is recorded every second.

[0211] Using mutual information algorithms (such as joint probability distribution and entropy calculation), analyze the correlation between "motor temperature change" and "inverter temperature change":

[0212] When the motor load suddenly increases (e.g., rapid acceleration), the motor temperature rises at time t → the inverter temperature also rises at time t+1, and the two changes synchronously.

[0213] Statistics show that in over 80% of operating conditions, the inverter temperature changes within 1-2 seconds after the motor temperature changes – resulting in a very high calculated mutual information value (e.g., reaching 0.8 out of 1).

[0214] Because the historical mutual information is high, it indicates that the thermal behavior of the motor and the inverter is significantly related. The edge MOT→INV in the model (representing the heat transfer relationship) will be judged as "strong thermal dependence" and will be retained first, corresponding to the main heat propagation path (MOT→INV→BATT→CHX).

[0215] Structural importance pruning: Use methods such as L1 penalty to score edges, and prune edges that are below a threshold to simplify the structure.

[0216] Structural importance pruning calculates an importance score for each hot-dependent edge (e.g., the edge from module A to module B) from the perspective of model complexity and contribution. It retains only edges with high scores (those that contribute greatly to the model) and prunes edges with low scores (those that contribute little or even interfere with the model), making the model structure simpler and more accurate.

[0217] L1 penalty: During model training, a penalty term proportional to the absolute value of the weight is added to the edge weight (representing the heat dependency strength). In this way, unimportant edges (with small weights) will be forced to reduce their weights due to the penalty, even approaching 0; Threshold screening: Set a score threshold (e.g., 0.2). If the importance score of an edge is lower than the threshold, it is determined that it has little contribution to the model and is directly cut off.

[0218] For example, taking the thermally dependent edge of motor (MOT) → inverter (INV) → battery (BATT) as an example, the process of structural importance pruning is as follows:

[0219] Model Training and Scoring: In a Graph Neural Network (GNN), initial weights are assigned to edges such as MOT→INV, INV→BATT, and MOT→BATT. During training, an L1 penalty is introduced. For the MOT→INV edge, which has strong physical coupling (short liquid cooling pipe) and high historical mutual information (temperature linkage), the weight increases continuously during training (e.g., final weight 0.8), resulting in a high importance score. For the INV→BATT edge, which is crucial in the heat transfer path (main path link), the weight is also large (e.g., 0.7), resulting in a high score. For the MOT→BATT edge, which has a long physical path (liquid cooling pipes are not directly connected) and low historical mutual information (weak temperature linkage), the weight is suppressed by the L1 penalty during training, approaching 0 (e.g., final weight 0.1), resulting in a low score.

[0220] Pruning operation: Set a threshold (e.g., 0.3). Edges with an importance score (0.1) below the threshold are pruned. In this way, the model only retains key edges such as MOT→INV and INV→BATT, which are consistent with the main heat propagation path (MOT→INV→BATT→CHX).

[0221] Simply put, it automatically determines which thermally dependent edges should be kept and which should be deleted based on three dimensions: physical connectivity, historical data association, and structural importance. This makes the graph structure of the thermal management model more realistic and improves the accuracy of thermal management system testing.

[0222] Figure 4 A test system 400 for an electric vehicle thermal management system is shown. Embodiments of this system are... Figure 1 Corresponding to the method embodiments shown, the system can be specifically applied to various electronic devices.

[0223] like Figure 4 As shown, the test system 400 for the electric vehicle thermal management system provided in this application embodiment includes:

[0224] The first acquisition module 401 is used to acquire time-series data of multiple thermal modules of an electric vehicle under a set thermal load change condition. The thermal modules include at least a battery pack, an inverter, and a motor.

[0225] The first generation module 402 is used to construct a graph structure with each of the thermal modules as nodes and the thermal influence relationship between the thermal modules as edges, and to generate thermal coupling relationships based on the graph structure using a message-passing-based graph neural network.

[0226] The second acquisition module 403 is used to construct a dynamic Bayesian network based on the time series data and the thermal coupling relationship to obtain the temperature response time dependency relationship between each of the thermal modules.

[0227] The second generation module 404 is used to generate a thermal hysteresis distribution map based on the time series data, the thermal coupling relationship and the dependency relationship, which is used to represent the thermal response delay relationship and its intensity among multiple modules.

[0228] Evaluation module 405 is used to evaluate the response coordination and adaptability of the thermal management control strategy under extreme thermal load conditions based on the thermal hysteresis distribution map.

[0229] Based on the same inventive concept, this application also provides an electronic device. The method corresponding to the electronic device can be the method in the foregoing embodiments, and its problem-solving principle is similar to that method. The electronic device provided in this application includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the methods and / or technical solutions of the foregoing embodiments of this application.

[0230] The electronic device can be a user device, or a device formed by integrating user devices and network devices through a network, or it can be an application running on the aforementioned devices. The user device includes, but is not limited to, various terminal devices such as computers, mobile phones, tablets, smartwatches, and wristbands. The network device includes, but is not limited to, network hosts, single network servers, multiple network server sets, or cloud computing-based computer sets, and can be used to implement some processing functions when setting an alarm clock. Here, the cloud consists of a large number of hosts or network servers based on cloud computing. Cloud computing is a type of distributed computing, consisting of a virtual computer composed of a group of loosely coupled computer sets.

[0231] Figure 5The diagram illustrates the structure of an apparatus suitable for implementing the methods and / or technical solutions in the embodiments of this application. The apparatus 500 includes a Central Processing Unit (CPU) 501, which can perform various appropriate actions and processes based on a program stored in a Read Only Memory (ROM) 502 or a program loaded from a storage portion 508 into a Random Access Memory (RAM) 503. The RAM 503 also stores various programs and data required for system operation. The CPU 501, ROM 502, and RAM 503 are interconnected via a bus 504. An Input / Output (I / O) interface 505 is also connected to the bus 504.

[0232] The following components are connected to I / O interface 505: an input section 506 including a keyboard, mouse, touchscreen, microphone, infrared sensor, etc.; an output section 507 including a cathode ray tube (CRT), liquid crystal display (LCD), LED display, OLED display, etc., and speakers, etc.; a storage section 508 including one or more computer-readable media such as hard disk, optical disk, magnetic disk, semiconductor memory, etc.; and a communication section 509 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 509 performs communication processing via a network such as the Internet.

[0233] In particular, the methods and / or embodiments in this application can be implemented as computer software programs. For example, the embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowchart. When the computer program is executed by the central processing unit (CPU) 501, it performs the functions defined in the methods of this application.

[0234] Another embodiment of this application provides a computer-readable storage medium having computer program instructions stored thereon, which can be executed by a processor to implement the methods and / or technical solutions of any one or more embodiments of this application described above.

[0235] Specifically, this embodiment may employ any combination of one or more computer-readable media. A computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device.

[0236] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

[0237] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in a device claim may also be implemented by a single unit or device through software or hardware. The terms "first," "second," etc., are used to indicate names and do not indicate any specific order.

Claims

1. A test method for an electric vehicle thermal management system, characterized in that, include: The timing data of multiple thermal modules of an electric vehicle under a set thermal load change condition are obtained, wherein the thermal modules include at least a battery pack, an inverter, and a motor. A graph structure is constructed with each of the thermally sensitive modules as nodes and the thermal influence relationship between the thermally sensitive modules as edges. Based on the graph structure, a thermal coupling relationship is generated using a graph neural network based on message passing. Based on the time-series data and the thermal coupling relationship, a dynamic Bayesian network is constructed to obtain the temperature response time dependency between each of the thermal modules. A thermal hysteresis distribution map is generated based on the time series data, the thermal coupling relationship, and the dependency relationship to represent the thermal response delay relationship and its intensity among multiple modules. The thermal hysteresis distribution map is used to evaluate the response coordination and adaptability of the thermal management control strategy under extreme thermal load conditions.

2. The test method for an electric vehicle thermal management system according to claim 1, characterized in that: The method of generating thermal coupling relationships using a message-passing-based graph neural network also includes obtaining the temperature change rate of the thermal module and cooling control state parameters. A structural gating mechanism is introduced, and a dynamic weighted gating function is designed for each heat transfer edge between the thermal modules in the graph structure. Based on the output of the gate function, the strength of the edge connection is dynamically adjusted or the heat transfer edge is retained / removed. If the thermal response is strong, the corresponding edge connection is retained and strengthened. If the cooling path is not shared, the edge connection is weakened or removed.

3. The test method for an electric vehicle thermal management system according to claim 1, characterized in that: The method of generating thermal coupling relationships using a message-passing-based graph neural network further includes calculating the state change amplitude of the module nodes after each round of thermal signal message passing in the graph neural network, wherein the state change includes temperature change and thermal influence weight change. A propagation termination gate is introduced, and a state change threshold is set. If the state change amplitude of the module node is less than the threshold after two consecutive rounds of hot signal message transmission, it is determined that the hot signal propagation tends to be stable, and the subsequent message transmission is terminated in advance. Based on the determination result of the propagation termination gate, the number of GNN message passing rounds is dynamically reduced.

4. The test method for an electric vehicle thermal management system according to claim 3, characterized in that, The construction of a dynamic Bayesian network to model the temperature response time dependency among the various thermal modules also includes, Based on the heat transfer paths between the modules and the temperature sequence correlation between the modules, the maximum correlation delay time is obtained. The maximum correlation delay time is used as the time window length for each heat transfer path, and the conditional probability distribution of the heat transfer path is constructed based on the time window length.

5. The test method for an electric vehicle thermal management system according to claim 4, characterized in that, Also includes: The method of generating hot-coupled relationships using a message-passing-based graph neural network further includes dynamically selecting edges using sparse structure learning combined with heuristic metrics, wherein the heuristic metrics include at least one of the following metrics: Based on the physical connection characteristics between the thermal modules, the physical coupling degree of the heat transfer path is quantified; Alternatively, the correlation between the temperature and thermal response of the thermistor module in the statistical historical data; Alternatively, the importance of hot-dependent edges can be scored using L1 penalties, and edges below a threshold can be automatically pruned.

6. A testing system for an electric vehicle thermal management system, characterized in that, include: The first acquisition module is used to acquire time-series data of multiple thermal modules of an electric vehicle under a set thermal load change condition. The thermal modules include at least a battery pack, an inverter, and a motor. The first generation module is used to construct a graph structure with each of the thermally sensitive modules as nodes and the thermal influence relationship between the thermally sensitive modules as edges, and to generate thermal coupling relationships based on the graph structure using a message-passing-based graph neural network. The second acquisition module is used to construct a dynamic Bayesian network based on the time series data and the thermal coupling relationship to obtain the temperature response time dependency relationship between each of the thermal modules; The second generation module is used to generate a thermal hysteresis distribution map based on the time series data, the thermal coupling relationship, and the dependency relationship, which is used to represent the thermal response delay relationship and its intensity among multiple modules. The evaluation module is used to evaluate the response coordination and adaptability of the thermal management control strategy under extreme thermal load conditions based on the thermal hysteresis distribution map.

7. The test system for an electric vehicle thermal management system according to claim 6, characterized in that: The method of generating thermal coupling relationships using a message-passing-based graph neural network also includes obtaining the temperature change rate of the thermal module and cooling control state parameters. A structural gating mechanism is introduced, and a dynamic weighted gating function is designed for each heat transfer edge between the thermal modules in the graph structure. Based on the output of the gate function, the strength of the edge connection is dynamically adjusted or the heat transfer edge is retained / removed. If the thermal response is strong, the corresponding edge connection is retained and strengthened. If the cooling path is not shared, the edge connection is weakened or removed.

8. An electronic device, the electronic device comprising: At least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, characterized in that: the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1-5.

9. A computer-readable medium storing computer program instructions thereon, characterized in that: The computer program instructions can be executed by a processor to implement the method as described in any one of claims 1-5.

10. A computer program product, comprising a computer program, characterized in that: When the computer program is executed by a processor, it implements the method as described in any one of claims 1-5.

Images