Vehicle heat damage risk point prediction method, device and equipment and storage medium

Abstract

The invention discloses a whole vehicle heat damage risk point prediction method, device and equipment and a storage medium, and the method comprises the steps: extracting the multi-dimensional features of all parts based on collected multi-vehicle-type heat damage test data, and generating a feature set; constructing a heterogeneous graph neural network model, wherein the construction process comprises the following steps of: decoupling a complex whole vehicle part network into three independent sub-graphs according to three heat transfer modes of conduction, convection and radiation; in each sub-graph, each part carries out message transmission with the adjacent part only through the corresponding heat transfer mode, so that local heat features, obtained from the three sub-graphs, of each part are fused into a comprehensive heat feature, and indirect heat influences among the parts are captured through iterative message transmission; and training the heterogeneous graph neural network model by using the feature set so as to output a prediction result of the heat damage risk of each component through the trained heterogeneous graph neural network model. According to the method, the heat damage risk point can be quickly and accurately predicted, and the test cost is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vehicles, in particular to a whole vehicle heat hazard risk point prediction method, device, equipment and storage medium. BACKGROUND

[0002] During the operation of the automobile, the internal combustion engine and the electric drive system will generate a large amount of heat energy, which is transmitted to the surrounding components through conduction, convection and radiation, etc. When the heat energy accumulates in the components and causes the temperature to exceed the allowable temperature of the components, it will trigger reactions such as reduced aging performance of rubber parts, failure of electronic components, carbonization of lubricating oil, etc. In severe cases, it may cause mechanical jamming and even fire hazards.

[0003] In order to find out these heat hazard risk points during vehicle development, the current industry generally uses an experience-oriented temperature monitoring method: during the prototype development stage, engineers arrange hundreds of thermocouple sensors in suspected risk areas according to historical data, and collect temperature data through some extreme working condition tests to see if it is over temperature. This scheme has three significant defects: (1) the selection of monitoring points lacks theoretical basis and relies on the experience of engineers. Engineers usually try to arrange as many points as possible to avoid missing risk points, resulting in a great waste of test resources; (2) the installation of multiple point sensors will change the local aerodynamic characteristics, and the wire harness interference will affect the test accuracy; (3) the single test period is as long as 60-80 hours, and the improvement scheme needs to be verified repeatedly.

[0004] Therefore, how to quickly and accurately predict the heat hazard risk points under different vehicle models and different working conditions is a technical problem that needs to be solved at present. SUMMARY

[0005] The main purpose of the present application is to provide a whole vehicle heat hazard risk point prediction method, device, equipment and storage medium, which can quickly and accurately predict the heat hazard risk points under different vehicle models and different working conditions, and greatly shorten the development cycle and reduce the test cost.

[0006] In a first aspect, the present application provides a whole vehicle heat hazard risk point prediction method, wherein the method comprises the following steps: Based on the collected multi-vehicle heat hazard test data, multi-dimensional features of each component are extracted, and a feature set is generated; An heterogeneous graph neural network model is constructed, wherein the construction process comprises: decoupling the complex whole vehicle component network into three independent subgraphs according to the three heat transfer modes of conduction, convection and radiation; in each subgraph, each component only performs message passing with adjacent components through the corresponding heat transfer mode, so as to fuse the local heat features obtained by each component from the three subgraphs into comprehensive heat features, and capture the indirect heat influence between components through iterative message passing; The heterogeneous graph neural network model is trained by using the feature set, so as to output a prediction result of the thermal hazard risk of each component through the trained heterogeneous graph neural network model.

[0007] In combination with the first aspect, as an optional implementation manner, the acquired physical features of different heat transfer modes are input into the weight prediction neural network, and the physical features include distance, contact area, material thermal conductivity, fluid velocity and surface emissivity. The weight prediction neural network outputs the weights of different heat transfer modes according to the input physical features by learning a mapping function of edge features and weights.

[0008] In combination with the first aspect, as an optional implementation manner, the conduction message of the component is calculated according to a formula: msg1 = w1 x f (u, v, conduction feature). The convection message of the component is calculated according to a formula: msg2 = w2 x f (u, v, convection feature). The radiation message of the component is calculated according to a formula: msg3 = w3 x f (u, v, radiation feature). The total message is calculated according to a formula: msg_total = msg1 + msg2 + msg3, wherein w1 is the weight of the conduction mode, w2 is the weight of the convection mode, w3 is the weight of the radiation mode, f (u, v, conduction feature) is a message function, u and v are components, and heat is transferred from u to v.

[0009] In combination with the first aspect, as an optional implementation manner, in the process of training the heterogeneous graph neural network model by using the feature set, the temperature of the component output by the heterogeneous graph neural network model is compared with the real temperature, and a predicted temperature error is calculated. The predicted temperature error is taken as a loss function, and the weight predictor and the heterogeneous graph neural network model are synchronously and jointly optimized through a back propagation algorithm.

[0010] In combination with the first aspect, as an optional implementation manner, multi-vehicle thermal hazard test data is collected, and the thermal hazard test data includes temperature sensor data, vehicle parameters, environmental conditions, operating states and vehicle general layout information. Multi-dimensional features of each component are extracted from the thermal hazard test data, and the multi-dimensional features include basic features, time series features, spatial features and label result features. The extracted multi-dimensional features of each component are converged to generate a feature set of each component.

[0011] ​In combination with the first aspect, as an optional implementation manner, the collected multi-vehicle thermal disaster test data is preprocessed, and the preprocessing includes: missing value interpolation, abnormal value correction and time sequence data alignment on the thermal disaster test data.

[0012] In combination with the first aspect, as an optional implementation manner, the feature set is divided into a training set and a validation set, the heterogeneous graph neural network model is trained and validated until the output result of the heterogeneous graph neural network model reaches a preset standard, and the training of the heterogeneous graph neural network model is completed. The environmental parameters, vehicle power and heat dissipation system configuration, vehicle operation parameters and component parameters are taken as inputs of the trained heterogeneous graph neural network model to output the prediction result of the component thermal disaster risk and temperature.

[0013] In a second aspect, the present application provides a whole vehicle thermal disaster risk point prediction device, which comprises: An extraction module is configured to extract multi-dimensional features of each component based on collected multi-vehicle thermal disaster test data, and generate a feature set. A construction module is configured to construct a heterogeneous graph neural network model, wherein the construction process includes: decoupling a complex whole vehicle component network into three independent subgraphs according to three heat transfer modes of conduction, convection and radiation; in each subgraph, each component only performs message passing with adjacent components through the corresponding heat transfer mode to fuse the local thermal features obtained by each component from the three subgraphs into comprehensive thermal features, and capture indirect thermal effects between components through iterative message passing. A prediction module is configured to train the heterogeneous graph neural network model using the feature set to output a prediction result of the thermal disaster risk of each component through the trained heterogeneous graph neural network model.

[0014] In a third aspect, the present application further provides an electronic device, which comprises: a processor; a memory having computer readable instructions stored thereon, wherein the computer readable instructions are executed by the processor to implement the method of any one of the first aspect.

[0015] In a fourth aspect, the present application further provides a computer readable storage medium storing computer program instructions, wherein the computer program instructions are executed by a computer to make the computer execute the method of any one of the first aspect.

[0016] The application provides a vehicle heat hazard risk point prediction method, device, equipment and storage medium. The method comprises the following steps: based on collected heat hazard test data of multiple vehicle models, multi-dimensional features of each component are extracted, and a feature set is generated; a heterogeneous graph neural network model is constructed, and the construction process comprises the following steps: a complex vehicle component network is decoupled into three independent subgraphs according to three heat transfer modes, namely conduction, convection and radiation; in each subgraph, each component only performs message passing with adjacent components through the corresponding heat transfer mode, so as to fuse the local heat features of each component obtained from the three subgraphs into comprehensive heat features, and the indirect heat influence between the components is captured through iterative message passing; the feature set is used to train the heterogeneous graph neural network model, so that the trained heterogeneous graph neural network model outputs a prediction result of the heat hazard risk of each component. The application can quickly and accurately predict the heat hazard risk points of different vehicle models and different working conditions, and greatly shortens the development cycle and reduces the test cost.

[0017] It should be understood that the above general description and the following detailed description are only exemplary and do not limit the application. BRIEF DESCRIPTION OF DRAWINGS

[0018] The accompanying drawings, which are incorporated into and form part of the specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the application.

[0019] Figure 1 A flow chart of a vehicle heat hazard risk point prediction method provided in an embodiment of the application; Figure 2 A schematic diagram of a vehicle heat hazard risk point prediction device provided in an embodiment of the application; Figure 3 A schematic diagram of an electronic device provided in an embodiment of the application; Figure 4 A schematic diagram of a computer-readable program medium provided in an embodiment of the application. DETAILED DESCRIPTION

[0020] The exemplary embodiments will be described in detail herein below with reference to the drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of apparatuses and methods consistent with some aspects of the present application as detailed in the appended claims.

[0021] In addition, the accompanying drawings are merely schematic illustrations of the present disclosure and are not necessarily drawn to scale. Some of the block diagrams shown in the accompanying drawings are functional entities, which do not necessarily correspond to physically or logically independent entities.

[0022] The embodiments of the present application are further described in detail below with reference to the accompanying drawings.

[0023] Referring to Figure 1 , Figure 1 Fig. 1 shows a flow chart of a whole vehicle heat hazard risk point prediction method provided by the present application, as shown in Fig. 2, the method comprises the steps of: Figure 1 Step S101: Based on the collected multi-vehicle heat hazard test data, multi-dimensional features of each component are extracted, and a feature set is generated.

[0024] Collecting multi-vehicle heat hazard test data, the heat hazard test data includes: temperature sensor data, vehicle parameters, environmental conditions, running state and vehicle overall layout information; Pretreating the collected multi-vehicle heat hazard test data, the pretreatment includes: missing value interpolation, abnormal value correction and time series data alignment of the heat hazard test data. From the heat hazard test data, multi-dimensional features of each component are extracted, the multi-dimensional features include: basic features (extracted from vehicle parameters, environmental conditions and running state data), time series features (extracted from temperature sensor data), spatial features (extracted from vehicle overall layout information) and label result features; the extracted multi-dimensional features of each component are converged to generate a feature set of each component.

[0025] For the convenience of understanding and illustration, collecting multi-vehicle heat hazard test data includes temperature sensor data (time series), vehicle parameters (model, component material, etc.), environmental conditions (air temperature, humidity, wind speed), running state (vehicle speed, load, duration), etc., vehicle overall layout, etc.

[0026] Interpolating the missing values of the data (for example, filling with the previous value, filling with the average of the previous and next values, or using linear interpolation, spline interpolation for estimation), correcting the abnormal values (which can be directly deleted, or corrected by interpolation method), aligning the time stamps of the multi-source time series data (for example, taking a master clock as a reference, correcting the time stamps of all other data sources to this reference), ensuring that the sensor data is synchronized.

[0027] From the collected data, basic features are extracted: vehicle model, heat dissipation system configuration, component attributes (three-dimensional coordinates, material thermal conductivity coefficient, etc.), environmental parameters (temperature, humidity, light intensity, etc.), vehicle running parameters (road resistance coefficient, vehicle speed, engine power, battery discharge power, air conditioner power, etc.). Time series features: set a sliding window (the time length covers a typical heat accumulation period) to extract statistical features such as temperature mean value, rising slope, fluctuation rate, maximum value in the window. Spatial features: relative position between components, heat conduction path, etc. Result label: according to the test over-temperature record, mark the risk point (binary classification label: risk / non-risk).​

[0028] It can be understood that the basic features can be understood as the causes and conditions leading to heat damage. It includes the attributes of the vehicle itself (who), the external environment (what environment), and the operating conditions (what to do).

[0029] The time series feature can be understood as heat damage is a cumulative and evolving dynamic process. It is not enough to look at the temperature value at a certain moment. By calculating the mean value (average heat level), the rising slope (heating speed), and the volatility (stability) through a sliding window, this process can be quantified to predict its future trend.

[0030] The spatial feature can be understood as heat can be conducted between components. The heat generated by one component can bake its adjacent components. The spatial feature (such as relative position, heat conduction path) is introduced to let the model understand this physical interaction, so as to more accurately assess the risk.

[0031] Step S102: Constructing a heterogeneous graph neural network model, wherein the construction process comprises: decoupling the complex whole vehicle component network into three independent subgraphs according to conduction, convection, and radiation three heat transfer modes; in each subgraph, each component only performs message passing with adjacent components through the corresponding heat transfer mode to fuse the local heat features obtained by each component from the three subgraphs into comprehensive heat features, and to capture the indirect heat influence between components through iterative message passing.

[0032] Specifically, the spatial heat interaction is modeled based on the graph neural network GNN, wherein the nodes in the graph structure are each sensor-distributed component, and the edges are the connection relationship between different nodes, i.e., the heat interaction mode between components. The heterogeneous graph is divided into three different edge types: radiation heat transfer, convection heat transfer, and conduction heat transfer. If all heat transfers are considered as the same kind of relationship (such as isomorphic graph), the model will ignore the differences of these physical mechanisms, resulting in a decrease in prediction accuracy. The heterogeneous graph explicitly distinguishes different types of heat interaction, making the model more consistent with the actual physical law. Independent weight matrices are designed for different edge types, and the information of different types of neighbors of the same target node is aggregated and combined.

[0033] The specific modeling process is: first, the whole graph (automobile component network) is divided into 3 subgraphs according to the heat transfer mode, corresponding to 3 message passing layers: conduction layer: only the conduction edge (directly connected components, such as engine and support) is retained, and the message passing of conduction heat transfer is specially processed; the convection layer: only the convection edge (components connected by fluid, such as engine and water tank) is retained, and the message passing of convection heat transfer is specially processed; the radiation layer: only the radiation edge (components that can transfer heat by radiation, such as high-temperature exhaust pipe and nearby plastic parts) is retained, and the message passing of radiation heat transfer is specially processed. In each (layer), the node (component) only transmits heat influence messages to adjacent components through the edges (corresponding to the heat transfer mode) of the layer, and the weight coefficient of the mode is used in the calculation. Each layer only calculates the influence of its own mode, and does not interfere with each other. When the 3 layers complete the message passing, each node (component) will obtain 3 local features, and then the 3 features are combined to obtain the comprehensive heat feature of the node. One layer transfer may not be enough (such as heat indirectly transmitted through multiple components). For example, engine-support (conduction), support-wire harness (conduction), and the heat of the wire harness is indirectly affected by the engine. At this time, the above steps can be repeated, and after each round of transmission, the comprehensive heat feature of the node will be more accurate (including the influence of indirect heat transfer).

[0034] For the convenience of understanding and illustration, the whole vehicle is taken as a huge relationship network (large graph). According to the heat transfer mode, the large graph is divided into three sub-networks: 1. Conduction graph: only concerned about directly contacting components. For example, the engine and its support. In this graph, only conduction is studied.

[0035] 2. Convection graph: only concerned about components that exchange heat through fluid (air, coolant). For example, the engine and the water tank. In this graph, only convection is studied.

[0036] 3. Radiation graph: only concerned about components that do not contact but transfer heat through infrared radiation. For example, the high-temperature exhaust pipe and the nearby plastic wire harness. In this graph, only radiation is studied. It can be understood that in the conduction graph, the node (component) only receives the heat influence from the directly contacting neighbors. In the convection graph, the node only receives the heat influence from the fluid-related neighbors. In the radiation graph, the node only receives the heat influence from the radiation-related neighbors. For the same component, there are three kinds of heat transfer information, namely conduction, convection and radiation, which are fused to generate the comprehensive feature of the component. The comprehensive feature comprehensively reflects the thermal state of the component under the joint action of the three heat transfer modes.

[0037] Where iterative deepening - simulating indirect effects - is the first round of passes: the model only sees the direct thermal effects. For example, the engine (node A) directly heats the bracket (node B). Second round of passes: when the model goes through the above process of disassembly - computation - fusion again, node B (the bracket) already contains the thermal information from A (the engine). Now, when node B does the conduction computation with the harness (node C), node B passes the thermal information from A (albeit indirectly) to C. Repeat this for many rounds, and the heat can be passed down through multiple components (A -> B -> C) like a relay race. The model can thus capture indirect, remote thermal hazard risks.

[0038] For example: the engine (A) is hot, it heats the bracket (B) through conduction, and the bracket (B) contacts the fuel harness (C) through conduction. The end goal is to judge whether the harness (C) is at risk.

[0039] First round: conduction graph: A -> B (B knows that A is hot).

[0040] Conduction graph: B -> C (C knows that B is a little hot, but does not know why B is hot.

[0041] Second round: conduction graph: A -> B (B knows more accurately the heat of A).

[0042] Conduction graph: B -> C (Now B's integrated thermal signature contains a strong signal from A, so C can more accurately receive the indirect information "I am near a very hot engine" and thus be more likely to be correctly predicted as a risk point).

[0043] The modeling process further comprises inputting the obtained physical characteristics of different heat transfer modes into a weight prediction neural network, the physical characteristics including distance, contact area, material thermal conductivity, fluid velocity, and surface emissivity; the weight prediction neural network outputs the weights of different heat transfer modes according to the input physical characteristics by learning a mapping function of edge features-weight.

[0044] In the message passing process of the GNN, the message of each edge (i.e., the heat transfer influence between two nodes) needs to be weighted with the corresponding weight. For example: the conduction message passed from node u to node v: msg1= w1× f (u, v, conduction features) The convection message: msg2= w2× f (u, v, convection features) The radiation message: msg3= w3× f (u, v, radiation features) Total message: msg_total = msg1+ msg2+ msg3 Where f ( ) is a message function (such as the temperature of node u multiplied by the conduction coefficient), and w1, w2, w3 are adaptive weights output by the weight predictor. In this way, the weight directly determines the proportion of each mode in the total message: the mode with a high weight has a greater impact on the target node (for example, a high conduction weight means that the temperature of the neighbor has a greater impact on the current node through conduction).

[0045] In an embodiment, the obtained physical characteristics of different heat transfer modes are input into a weight prediction neural network, the physical characteristics including distance, contact area, material thermal conductivity, fluid velocity, and surface emissivity; the weight prediction neural network outputs the weights of different heat transfer modes according to the input physical characteristics by learning a mapping function of edge features-weight.

[0046] It can be understood that by optimizing the model (heterogeneous graph neural network model) through the loss function, the model needs to learn reasonable weights through training, and the core is to adjust the parameters of the weight predictor in the reverse direction with the prediction error. The temperature of the node is predicted by the GNN, and the error L is obtained by comparing with the true temperature. The error L is propagated in the reverse direction along the network, which not only optimizes the parameters of the message passing in the GNN, but also optimizes the parameters of the weight predictor. For example: if the true heat transfer of a certain edge is mainly by conduction, but the model predicts a low w1, resulting in insufficient msg1 and a low predicted temperature → the error L increases → when back propagation, the parameters of the weight predictor are adjusted, so that w1 is increased next time the edge with similar characteristics is encountered. Conversely, if the convection weight is too high, the error will force the weight predictor to reduce w2 in the corresponding scenario. Repeat the process of input data → predict weight → pass message → calculate error → adjust parameter until the prediction error of the model is small enough. At this time, the "feature → weight" mapping learned by the weight predictor is consistent with the true heat transfer law.

[0047] Adaptive weight learning is adopted, that is, a learnable neural network is used to predict the weight according to the characteristics of the edge, the weight is integrated into the message passing of the GNN, and the prediction error is optimized in the reverse direction, so that the model automatically learns "which mode is more important in which scenario". The whole process does not need to manually set the weight rule, and is completely data-driven, so it can adapt to complex and variable actual scenarios.

[0048] Among them, the loss function adopts a multi-task loss function, and the model needs to complete the prediction of whether the component will overheat (classification task: yes / no) and the highest temperature of the component (regression task: specific value, such as 150°C). First, calculate a "loss" for each task (for example, task 1 is wrong, deduct 2 points, task 2 temperature deviation is large, deduct 1 point); then assign a weight to the loss of each task (for example, task 1 is the most important, weight 0.6; task 2 is the second, weight 0.4); finally, calculate the total loss: (2x0.6) + (1x0.4) = 1.6 points. In automobile safety, "whether it is overheated" is more critical than the specific temperature (overheating may cause a fire), so the loss of "whether it is overheated" is given a slightly higher weight, so that the model can prioritize the accuracy of this task. The two tasks are actually related (for example, the higher the "maximum temperature", the higher the "overheating probability"). When learning to "predict the maximum temperature", it will indirectly help it understand the judgment standard of "whether it is overheated"; conversely, learning "whether it is overheated" can also help it accurately grasp the threshold of the temperature. This "knowledge sharing" can make the model learn faster and more stable.

[0049] For easy understanding, a complete example is given: suppose predicting whether plastic harness (P) will overheat. Involved components: E: engine (high temperature heat source), B: metal bracket (directly connected with engine), P: plastic harness (in contact with bracket, and near engine).

[0050] Heat interaction: E B: conduction (direct contact), B P: conduction (direct contact), E → P: radiation (radiation in space).

[0051] 1. Data collection: time series: install temperature sensors on E, B, P, record temperature changes over time.

[0052] Vehicle parameters: record the materials of E, B, P components (such as engine cast iron, bracket steel, harness PVC).

[0053] Environmental conditions: record air temperature, wind speed during the test.

[0054] Operating state: record vehicle speed, engine speed.

[0055] Overall arrangement: obtain the three-dimensional coordinates of E, B, P, and calculate the distance between them.

[0056] 2. Preprocessing: found that the data of P sensor had missing values at a certain second, and interpolated and filled in the data before and after the time point.

[0057] Found that B sensor had an instantaneous abnormal value (such as 1000°C), judged as noise, and corrected to a reasonable value.

[0058] Align the data timestamps of all sensors, environment, and vehicle state, to ensure that at T=10 seconds, all data are synchronized.

[0059] 3. Feature extraction Suppose focus on T=10 seconds, and set a time window (such as the previous 5 seconds to the next 5 seconds) centered on this time. Basic features: P.material = "PVC", B.material = "Steel", E.material = "CastIron"; ambient_temp = 35°C, vehicle_speed = 80 km / h.

[0060] Time series features (calculated on temperature data in the window): P.temp_mean = 85°C (average temperature in the window), P.temp_slope = 0.8 (temperature is rising rapidly).

[0061] Spatial features: dist(E, P) = 0.3m (linear distance between E and P); contact_area(B, P) = 0.001㎡ (contact area between B and P).

[0062] Result label: According to the test record, the temperature of P has not yet exceeded the standard, but is in a rapid rising trend, and the expert marks it as a risk.

[0063] 4. Constructing the heterogeneous graph Conduction subgraph: contains edge E-B and B-P. Radiation subgraph: contains edge E-P. Convection subgraph: none in this case (assuming air convection has little effect).

[0064] For edge E-P: Input features x_E-P = [distance=0.3m, surface emissivity=0.9,...] to the weight predictor.

[0065] Predictor output weights: [w_conduction=0.05, w_convection=0.05, w_radiation=0.9]. The model judges that this edge is dominated by radiation heat transfer For edge B-P: Input features x_B-P = [contact area=0.001㎡, material thermal conductivity=0.2,...] to the weight predictor.

[0066] Predictor output weights: [w_conduction=0.95, w_convection=0.03, w_radiation=0.02]. The model judges that this edge is dominated by conduction heat transfer.

[0067] Node P receives messages from its neighbors (E and B). The message from the engine (E): since there is only a radiation edge between E and P, only the radiation message is calculated. Total message_E = w_radiation * f_radiation(E's temperature, E's surface properties...).

[0068] The message from the bracket (B): since there is only a conduction edge between B and P, only the conduction message is calculated. Total message_B = w_conduction * f_conduction(B's temperature, B's material...) P aggregates and updates: P aggregates (e.g., sums) the total message_E and the total message_B. P combines its own original features (temp=85°C, material="PVC") and updates to generate a new, more comprehensive feature. This new feature contains the information that "I am being radiated by the engine, and also being heated by the conduction from the bracket".

[0069] 5. Multiple rounds of passing (reflecting indirect effects) Round 1: B tells P, "I'm hot" (but P doesn't know where B's heat comes from).

[0070] Round 2: E tells B: I am very hot! (Through the conduction subgraph). B updates its own characteristics and knows that the root cause of its heat is E.

[0071] In the third round, B tells P again: I'm updating you, I'm not only hot now, but my heat source is that super hot engine E!

[0072] After multiple rounds, the comprehensive characteristics of P clearly include direct radiative heat from E and indirect conductive heat transferred through B.

[0073] 6. Output and Optimization The final integrated features of P are input into a prediction head (small neural network).

[0074] Output: Regression task: Predicted temperature = 102°C; Classification task: Overheating risk probability = 92% (classified as risk).

[0075] Optimization (through the loss function): Assuming the actual situation: the actual temperature of P is 108°C, and it is already overheated (risk = 1).

[0076] Loss Calculation: Regression Loss: |102°C - 108°C| = 6 Classification Loss: The model predicts a 92% risk probability, but the actual risk is 100%, hence the discrepancy. Total Loss = 0.4 * Regression Loss + 0.6 * Classification Loss (because classification is more important) Backpropagation: This total loss is backpropagated, adjusting the parameters of the weight predictor. For example, it might find that for edge EP, a w_radiation of 0.9 is underestimated, and should be 0.95 next time. It also adjusts the message function f and the parameters of the final prediction head. Ultimately: Through thousands of such adjustments, the model learns how to accurately assign weights based on edge features, how to accurately pass messages, and how to make accurate predictions. (The above is for ease of understanding only.) Step S103: Train the heterogeneous graph neural network model using the feature set, so as to output the prediction results of the thermal damage risk of each component through the trained heterogeneous graph neural network model.

[0077] Specifically, the feature set is divided into a training set and a validation set to train and validate the heterogeneous graph neural network model until the output of the heterogeneous graph neural network model reaches the preset standard, thus completing the training of the heterogeneous graph neural network model. Environmental parameters, vehicle power and cooling system configuration, vehicle operating parameters and component parameters are used as inputs to the trained heterogeneous graph neural network model to output the predicted results of component thermal damage risk and temperature.

[0078] Understandably, the dataset is divided into a training set (80%) and a validation set (20%). The model is trained using the training set data, with the goal of achieving a risk point identification accuracy of over 98% on the validation set. The inputs are environmental parameters (temperature, humidity, solar radiation intensity), vehicle power and cooling system configuration, vehicle operating parameters (drag, speed, engine power, etc.), and component parameters (3D coordinates, material thermal conductivity, allowable temperature, etc.). The output is whether a component has an overheating risk and its predicted temperature value. Furthermore, the model is dynamically updated based on new experimental data.

[0079] In summary, this application can provide prediction results within seconds, which can be used for rapid iterative design (parameter optimization, multi-solution screening), significantly shorten the test and development cycle, learn complex correlations in historical data (multi-physics coupling, etc.), is applicable to multi-condition migration, and has low prediction cost for new scenarios.

[0080] Reference Figure 2 , Figure 2 The diagram shown is a schematic of a vehicle thermal damage risk prediction device provided by the present invention. Figure 2 As shown, the device includes: Extraction module 201: It is used to extract multi-dimensional features of each component based on the collected thermal damage test data of multiple vehicle models, and generate a feature set.

[0081] Module 202: It is used to build a heterogeneous graph neural network model. The construction process includes: decoupling the complex vehicle component network into three independent subgraphs according to the three heat transfer modes of conduction, convection and radiation; within each subgraph, each component communicates with its neighboring components only through the corresponding heat transfer mode, so as to fuse the local thermal features obtained by each component from the three subgraphs into comprehensive thermal features, and to capture the indirect thermal effects between components through iterative message passing.

[0082] Prediction module 203: It is used to train the heterogeneous graph neural network model using the feature set, so as to output the prediction results of the thermal damage risk of each component through the trained heterogeneous graph neural network model. Furthermore, in one possible implementation, the construction module is also used to input the physical features of the different heat transfer modes acquired into a weighted prediction neural network, the physical features including: distance, contact area, material thermal conductivity, fluid velocity and surface emissivity; The weight prediction neural network learns the feature-weight mapping function of the edges and outputs the weights of different heat transfer modes based on the input physical features.

[0083] Furthermore, in one possible implementation, the building module is also used to calculate the conductive message of the component according to the formula: msg1 = w1 × f (u, v, conductive characteristics); calculating a convection message of the component according to a formula: msg2 = w2xf(u, v, convection feature); calculating a radiation message of the component according to a formula: msg3 = w3xf(u, v, radiation feature); calculating a comprehensive thermal message according to a formula: msg_total = msg1+ msg2+ msg3, wherein w1 is a weight of the conduction mode, w2 is a weight of the convection mode, w3 is a weight of the radiation mode, f ( ) is a message function, and u and v are components, indicating that u transfers heat to v.

[0084] Further, in a possible implementation, the prediction module is further configured to, in the process of training the heterogeneous graph neural network model by using the feature set, compare a component temperature output by the heterogeneous graph neural network model with a real temperature, and calculate a predicted temperature error; use the predicted temperature error as a loss function, and synchronously and jointly optimize the weight predictor and the heterogeneous graph neural network model by using a back propagation algorithm.

[0085] Further, in a possible implementation, the extraction module is further configured to collect multi-vehicle thermal hazard test data, the thermal hazard test data including temperature sensor data, vehicle parameters, environmental conditions, operating states, and vehicle overall arrangement information; extract multi-dimensional features of each component from the thermal hazard test data, the multi-dimensional features including basic features, time series features, spatial features, and label result features; converge the extracted multi-dimensional features of each component to generate a feature set of each component.

[0086] Further, in a possible implementation, the extraction module is further configured to perform preprocessing on the collected multi-vehicle thermal hazard test data, the preprocessing including missing value interpolation, abnormal value correction, and time series data alignment on the thermal hazard test data.

[0087] Further, in a possible implementation, the prediction module is further configured to divide the feature set into a training set and a verification set, train and verify the heterogeneous graph neural network model, and complete training of the heterogeneous graph neural network model after an output result of the heterogeneous graph neural network model reaches a preset standard. use environmental parameters, vehicle power and heat dissipation system configurations, vehicle operating parameters, and component parameters as inputs of the trained heterogeneous graph neural network model, to output a predicted result of a component thermal hazard risk and temperature.

[0088] The electronic device 300 according to this implementation of the present application will be described below with reference to Figure 3 .Figure 3 The electronic device 300 illustrated is merely one example and should not be taken as limiting the scope of functionality or use of embodiments of the application.

[0089] As shown, the electronic device 300 is in the form of a general-purpose computing device. The components of the electronic device 300 can include, but are not limited to, the at least one processing unit 310, the at least one storage unit 320, and a bus 330 that connects the various system components, including the storage unit 320 and the processing unit 310. Figure 3

[0090] The storage unit stores program code that can be executed by the processing unit 310 such that the processing unit 310 performs the steps described in the above "Embodiment Methods" section of this specification in accordance with the various exemplary embodiments of this application.

[0091] The storage unit 320 can include a readable medium in the form of volatile storage such as random access memory (RAM) 321 and / or cache memory 322, and can further include non-volatile storage such as read only memory (ROM) 323.

[0092] The storage unit 320 can further include program / utility 324 having a set of at least one program modules 325, including operating system, one or more application programs, other program modules, and program data, each or some combination thereof, which may

[0093] The bus 330 can represent one or more of several types of bus structures, including a storage bus or bus controller, a peripheral bus, a graphics acceleration bus, a processor or local bus using any of a variety of bus architectures.

[0094] ​The electronic device 300 can also communicate with one or more external devices such as a keyboard or a pointing device, a Bluetooth device, etc., and can also communicate with one or more devices that enable a user to interact with the electronic device 300 and / or one or more devices that enable the electronic device 300 to communicate with one or more other computing devices. Such communication can occur via Input / Output (I / O) interface 350. Also, the electronic device 300 can communicate with one or more networks such as a local area network (LAN), a wide area network (WAN), and / or the Internet through network adapter 360. As depicted, network adapter 360 communicates with the other components of the electronic device 300 via bus 330. It should be understood that although not shown, other hardware and / or software components could be used in conjunction with the electronic device 300. These include, but are not limited to, microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data archival storage systems, etc.

[0095] Those skilled in the art will readily understand that the example embodiments described herein can be implemented by software and / or by software in combination with the requisite hardware. Thus, the technical solutions according to the embodiments of the present disclosure can be embodied in the form of a software product, which can be stored in an non-volatile storage medium (which can be a CD-ROM, a USB flash disk, a mobile hard disk, etc.) or a network, and includes a number of instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to perform the methods according to the embodiments of the present disclosure.

[0096] According to the solutions of the present disclosure, a computer readable storage medium is also provided, which stores the program product capable of implementing the above-mentioned methods. In some possible embodiments, various aspects of the present disclosure can also be implemented in the form of a program product, which includes program codes for causing the terminal device to perform the steps according to various example embodiments of the present disclosure described in the above-mentioned “example method” section when the program product is run on the terminal device.

[0097] Reference Figure 4 As shown, the program product 400 for implementing the above-mentioned methods according to the embodiments of the present disclosure is described, which can adopt a portable compact disc read-only memory (CD-ROM) and includes program codes, and can be run on a terminal device, such as a personal computer. However, the program product of the present disclosure is not limited to this, and in this document, the readable storage medium can be any tangible medium containing or storing a program, which can be used by or in combination with an instruction execution system, device or apparatus.

[0098] The program product can employ any combination of one or more computer-readable media. The computer-readable media can be a computer-readable storage medium or a computer-readable signal medium. The computer-readable storage medium can be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer-readable storage medium include the following: an electrical connection having one or more wires, a portable disc, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0099] The computer-readable signal medium can include a computer-readable storage medium that is propagated as a carrier wave. The computer-readable signal medium can further be any computer-readable medium that is not a storage medium. The computer-readable signal medium can be a computer-readable storage medium that is a propagated signal on a computer-readable storage medium.

[0100] The program code embodied on the computer-readable media can be transmitted using any appropriate medium, including but not limited to wireless, wired, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0101] The program code can be executed by one or more programmable processors, which can be implemented in one or more computer devices including any combination of a microprocessor, a microcontroller, a central processing unit (CPU), a graphics processing unit (GPU), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic device. The program code can be written in any of a number of suitable programming languages and can be compiled into an executable program.

[0102] In addition, the above-described flowcharts are merely illustrative of the processes included in the method according to the exemplary embodiments of the present application, and are not intended to limit the present application. It is readily understood that the processes shown in the above-described flowcharts do not indicate or limit the time sequence of the processes. In addition, it is readily understood that the processes can be executed synchronously or asynchronously, for example, in a plurality of modules.

[0103] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

[0104] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

Claims

1. A method for predicting a risk point of heat disaster of a whole vehicle, characterized in that, The method comprises the steps of: Based on the collected multi-vehicle heat damage test data, multi-dimensional features of each component are extracted, and a feature set is generated; A heterogeneous graph neural network model is constructed, wherein the construction process comprises: decoupling the complex vehicle component network into three independent subgraphs according to conduction, convection and radiation three heat transfer modes; In each subgraph, each component only transmits messages with adjacent components through the corresponding heat transfer mode, so as to fuse the local heat features obtained by each component from the three subgraphs into comprehensive heat features, and capture the indirect heat influence between components through iterative message transmission; The feature set is used to train the heterogeneous graph neural network model, so as to output the prediction result of the heat damage risk of each component through the trained heterogeneous graph neural network model.

2. The method of claim 1, wherein, The construction process further comprises: The physical features of different heat transfer modes obtained are input into a weight prediction neural network, and the physical features include distance, contact area, material thermal conductivity, fluid velocity and surface emissivity; The weight prediction neural network outputs the weight of different heat transfer modes according to the input physical features by learning the mapping function of edge feature-weight.

3. The method of claim 1, wherein, Each component only transmits messages with adjacent components through the corresponding heat transfer mode, so as to fuse the local heat features obtained by each component from the three subgraphs into comprehensive heat features, which comprises: The conduction message of the component is calculated according to the formula: msg1 = w1×f (u, v, conduction feature); The convection message of the component is calculated according to the formula: msg2 = w2×f(u, v, convection feature); The radiation message of the component is calculated according to the formula: msg3 = w3×f(u, v, radiation feature); The total heat message is calculated according to the formula: msg_total = msg1 + msg2 + msg3, where w1 is the weight of the conduction mode, w2 is the weight of the convection mode, w3 is the weight of the radiation mode, f ) is the message function, and u and v are components, indicating heat transfer from u to v.

4. The method of claim 1, wherein, Further comprising: During the process of training the heterogeneous graph neural network model using the feature set, the temperature error of the component output by the heterogeneous graph neural network model is compared with the real temperature, and the predicted temperature error is calculated; The predicted temperature error is used as a loss function, and the weight predictor and the heterogeneous graph neural network model are simultaneously and jointly optimized through a back propagation algorithm.

5. The method of claim 1, wherein, The method comprises the steps of: Collecting multi-vehicle heat damage test data, wherein the heat damage test data comprises temperature sensor data, vehicle parameters, environmental conditions, running states and vehicle overall arrangement information; Multi-dimensional features of each component are extracted from the heat damage test data, wherein the multi-dimensional features comprise basic features, time series features, spatial features and label result features; The extracted multi-dimensional features of each component are aggregated to generate a feature set of each component.

6. The method of claim 1, wherein, Before the step of extracting multi-dimensional features of each component and generating a feature set, the method comprises the steps of: The collected multi-vehicle heat damage test data is preprocessed, and the preprocessing comprises missing value interpolation, abnormal value correction and time series data alignment of the heat damage test data.

7. The method of claim 1, wherein, The method comprises the steps of: The feature set is divided into a training set and a validation set, the heterogeneous graph neural network model is trained and validated until the output result of the heterogeneous graph neural network model reaches a preset standard, and the training of the heterogeneous graph neural network model is completed; The environmental parameters, vehicle power and heat dissipation system configuration, vehicle operation parameters and component parameters are taken as inputs of the trained heterogeneous graph neural network model to output the prediction results of the component heat hazard risk and temperature.

8. A vehicle heat hazard risk point prediction device characterized by comprising: Comprise: An extraction module for extracting multi-dimensional features of each component based on collected multi-vehicle heat hazard test data and generating a feature set; A construction module for constructing a heterogeneous graph neural network model, wherein the construction process comprises: decoupling the complex vehicle component network into three independent subgraphs according to conduction, convection and radiation three heat transfer modes; In each subgraph, each component only transmits messages with adjacent components through the corresponding heat transfer mode to fuse the local thermal features obtained by each component from the three subgraphs into comprehensive thermal features, and capture the indirect thermal effects between components through iterative message passing; A prediction module for training the heterogeneous graph neural network model using the feature set to output the prediction results of the heat hazard risk of each component through the trained heterogeneous graph neural network model.

9. An electronic device, comprising: The electronic device comprises: A processor; A memory, the memory has computer readable instructions stored thereon, and the computer readable instructions are executed by the processor to implement the method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores computer program instructions, which, when executed by a computer, cause the computer to execute the method according to any one of claims 1 to 7.

Images