Vehicle motor temperature control method and device and vehicle motor temperature prediction method

Abstract

This application relates to a vehicle motor temperature control method, apparatus, and vehicle motor temperature prediction method. The method includes: based on target runtime timing data of the vehicle motor at a target time and multiple coolant flow rate acquisition data, iteratively optimizing the output data of a motor temperature prediction model by minimizing the output data to obtain target flow rate control data for controlling the coolant flow rate of the vehicle motor; the motor temperature prediction model is used to divide the input runtime timing data and coolant flow rate timing data into multiple runtime subsequences and multiple coolant flow rate subsequences, and to perform graph convolution feature extraction and temperature prediction processing using partial data from each subsequence; in each iterative optimization process, the coolant flow rate data included in the partial coolant flow rate subsequence of the coolant flow rate timing data input to the motor temperature prediction model is the coolant flow rate candidate data for the current iterative optimization process. This method can improve the efficiency of vehicle motor temperature control.

Description

Technical Field

[0001] This application relates to the field of vehicle motor control technology, and in particular to a vehicle motor temperature control method, device, and vehicle motor temperature prediction method. Background Technology

[0002] Permanent magnet synchronous motors (PMSMs) have gradually become the mainstream choice for drive motors in new energy vehicles due to their advantages such as high power density, high efficiency, fast response speed, and wide speed range. However, energy loss is inevitable in the process of converting electrical energy into mechanical energy. For motor systems with rated power ranging from tens to hundreds of kilowatts, even a small energy loss (e.g., less than 5%) is enough to cause a high temperature rise in the motor. This heat can cause permanent damage such as aging to the motor's insulation system and can also cause demagnetization of the permanent magnets in the PMSM, affecting the control accuracy of the motor controller and the transmission efficiency of the motor.

[0003] The vehicle motor temperature control method in related technologies constructs a digital twin model of the motor temperature field, and then predicts the motor temperature data at future times by using the digital twin model of the motor temperature field and real-time operating data collected under the current operating conditions, and solves the optimal coolant flow rate under the current operating conditions so that the maximum temperature of the motor is maintained at a low level for a period of time in the future.

[0004] However, the above-mentioned vehicle motor temperature control methods are inefficient. Summary of the Invention

[0005] Therefore, it is necessary to provide a vehicle motor temperature control method, device, and vehicle motor temperature prediction method that can improve efficiency in response to the above-mentioned technical problems.

[0006] In a first aspect, this application provides a method for controlling the temperature of a vehicle motor, including:

[0007] Acquire the target runtime sequence data of the vehicle motor at the target time and multiple coolant flow acquisition data, wherein the target runtime sequence data includes multiple runtime acquisition data;

[0008] Based on the target runtime sequence data and multiple coolant flow rate acquisition data, the target flow control data is obtained by iteratively optimizing the output data of the motor temperature prediction model by minimizing the output data. The motor temperature prediction model is used to divide the input runtime sequence data and coolant flow rate time series data into multiple runtime subsequences and multiple coolant flow rate subsequences according to a preset time window. Partial data in each runtime subsequence and each coolant flow rate subsequence are used for graph convolution feature extraction and temperature prediction processing. The output is the temperature prediction data corresponding to the input runtime sequence data and coolant flow rate time series data. In each iteration optimization process, the coolant flow rate data included in the partial coolant flow rate subsequence of the coolant flow rate time series data input to the motor temperature prediction model is the coolant flow rate candidate data corresponding to the current iteration optimization process.

[0009] The coolant flow rate used to cool the vehicle's motor is controlled based on the target flow control data.

[0010] Secondly, this application also provides a method for predicting the temperature of a vehicle motor, including:

[0011] Acquire the vehicle motor's runtime timing data and coolant flow timing data at the target time;

[0012] The runtime sequence data and the coolant flow time sequence data are divided into multiple runtime subsequences and multiple coolant flow subsequences according to the preset time window.

[0013] Graph convolution is used to extract features from some data in each operation subsequence and each cold liquid flow rate subsequence to obtain the target operation convolution feature and target flow rate convolution feature corresponding to the target time.

[0014] Based on the target operation convolutional features and the target flow convolutional features, prediction processing is performed to obtain the target predicted temperature data of the vehicle motor at the target time.

[0015] Thirdly, this application also provides a vehicle motor temperature control device, the device comprising:

[0016] The data acquisition module is used to acquire the target runtime sequence data of the vehicle motor at the target time and multiple coolant flow acquisition data, wherein the target runtime sequence data includes multiple runtime acquisition data;

[0017] The flow optimization module is used to iteratively optimize the target flow control data by minimizing the output data of the motor temperature prediction model based on the target runtime sequence data and multiple coolant flow acquisition data. The motor temperature prediction model divides the input runtime sequence data and coolant flow time-series data into multiple runtime subsequences and multiple coolant flow subsequences according to a preset time window. It then uses a portion of the data from each runtime subsequence and each coolant flow subsequence for graph convolution feature extraction and temperature prediction, outputting temperature prediction data corresponding to the input runtime sequence data and coolant flow time-series data. In each iteration of optimization, the coolant flow data included in the coolant flow time-series data input to the motor temperature prediction model constitutes the candidate coolant flow data for the current iteration.

[0018] The flow control module is used to control the flow rate of coolant used to cool the vehicle's motor based on target flow control data.

[0019] The aforementioned vehicle motor temperature control method, device, and vehicle motor temperature prediction method acquire target runtime timing data and multiple coolant flow rate acquisition data corresponding to the vehicle motor at a target time. The target runtime timing data includes multiple acquisition data. Based on the target runtime timing data and multiple coolant flow rate acquisition data, iterative optimization is performed by minimizing the output data of the motor temperature prediction model to obtain target flow control data. The coolant flow rate used to cool the vehicle motor is controlled based on the target flow control data. The motor temperature prediction model divides the input runtime timing data and coolant flow rate time-series data into multiple runtime subsequences and multiple coolant flow rate subsequences according to a preset time window. Partial data from each runtime subsequence and each coolant flow rate subsequence are used for graph convolution feature extraction and temperature prediction processing. The output is temperature prediction data corresponding to the input runtime timing data and coolant flow rate time-series data. In each iterative optimization process, the partial coolant flow rate subsequences in the input coolant flow rate time-series data of the motor temperature prediction model include… The coolant flow rate data is the candidate coolant flow rate data corresponding to the current iterative optimization process. Determining the target flow control data requires multiple iterations. Each iteration requires using a motor temperature prediction model to obtain the temperature prediction data corresponding to the candidate coolant flow rate data for the current iteration. In other words, determining the target flow control data requires calling the motor temperature prediction model multiple times for calculation. The motor temperature prediction model provided in the above embodiment divides the input time-series data into multiple subsequences according to a preset time window. It uses a portion of the data from each subsequence for graph convolution feature extraction and temperature prediction. This avoids the problem in related technologies where the digital twin model of the motor temperature field requires feature extraction for each piece of input time-series data, resulting in high computational complexity and low control efficiency. Using the motor temperature prediction model provided in the above embodiment reduces the computational load for each iteration, thereby reducing the overall computational load for determining the target flow control data, increasing the speed of obtaining the target flow control data, and resulting in higher control efficiency. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a diagram illustrating the application environment of a vehicle motor temperature control method in one embodiment.

[0022] Figure 2This is a flowchart illustrating a vehicle motor temperature control method in one embodiment;

[0023] Figure 3 This is a flowchart illustrating the steps for obtaining target flow control data in one embodiment;

[0024] Figure 4 This is a flowchart illustrating the steps for obtaining candidate temperature prediction data in one embodiment.

[0025] Figure 5 This is a flowchart illustrating a vehicle motor temperature control method in another embodiment;

[0026] Figure 6 This is a schematic diagram illustrating the process of obtaining the speed diagram structure, torque diagram structure, and flow diagram structure in one embodiment.

[0027] Figure 7 This is a schematic diagram illustrating the process of obtaining intermediate rotational speed convolution features, intermediate torque convolution features, and intermediate flow rate convolution features in one embodiment.

[0028] Figure 8 This is a schematic diagram illustrating the process of obtaining target rotational speed convolutional features, target torque convolutional features, and candidate flow rate convolutional features in one embodiment.

[0029] Figure 9 This is a flowchart illustrating a vehicle motor temperature prediction method in one embodiment;

[0030] Figure 10 This is a structural block diagram of a vehicle motor temperature control device provided in one embodiment;

[0031] Figure 11 This is a structural block diagram of a vehicle motor temperature prediction device provided in one embodiment;

[0032] Figure 12 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0034] The vehicle motor temperature control method provided in this application embodiment can be applied to, for example... Figure 1In the application environment shown, the vehicle controller 102 is connected to the motor 104. The vehicle controller 102 is used to acquire real-time operating data corresponding to the motor 104 and control the coolant flow rate corresponding to the motor 104. The vehicle controller 102 may be, but is not limited to, an electronic control unit (ECU) or a microcontroller unit (MCU). In some embodiments, the motor may be a permanent magnet synchronous motor, while in other embodiments, the motor 104 may be other types of motors, such as an AC asynchronous motor.

[0035] The vehicle controller 102 acquires target runtime timing data and multiple coolant flow rate acquisition data corresponding to the vehicle motor at a target time. The target runtime timing data includes multiple operational acquisition data. Based on the target runtime timing data and multiple coolant flow rate acquisition data, it iterative optimization is performed by minimizing the output data of the motor temperature prediction model to obtain target flow control data. The motor temperature prediction model is used to divide the input runtime timing data and coolant flow rate time-series data into multiple operational subsequences and multiple coolant flow rate subsequences according to a preset time window. Partial data in each operational subsequence and each coolant flow rate subsequence are used for graph convolution feature extraction and temperature prediction processing. The output is temperature prediction data corresponding to the input runtime timing data and coolant flow rate time-series data. In each iteration optimization process, the coolant flow rate data included in the partial coolant flow rate subsequence of the coolant flow rate time-series data input to the motor temperature prediction model is the coolant flow rate candidate data corresponding to the current iteration optimization process. The coolant flow rate used to cool the vehicle motor is controlled based on the target flow control data.

[0036] In one exemplary embodiment, such as Figure 2 As shown, a vehicle motor temperature control method is provided, which is applied to... Figure 1 The following steps are used as an example to illustrate the process of using the vehicle controller 102, including steps 202 to 206. Wherein:

[0037] Step 202: Obtain the target runtime sequence data of the vehicle motor at the target time and multiple coolant flow rate acquisition data.

[0038] The target time refers to the moment when the coolant flow rate needs to be determined, which can be a periodic moment; for example, the vehicle controller updates the coolant flow rate every 10 seconds.

[0039] The target runtime sequence data includes multiple runtime acquisition data. The vehicle controller 102 continuously acquires the operating status data of the vehicle motor at a certain sampling frequency; the target runtime sequence data refers to multiple runtime acquisition data within a target time and a preset duration prior to the target time. For example, the target runtime sequence data includes a first number of runtime acquisition data; for example, the first number is determined by the acquisition frequency and the preset duration.

[0040] In one possible implementation, the operational data acquisition includes speed acquisition data and torque acquisition data. Correspondingly, the target operational timing data includes target speed timing data and target torque timing data. Here, speed refers to the motor speed, which can be acquired using a rotary encoder; torque refers to the motor output torque, which can be acquired using a torque sensor. The motor speed and output torque reflect the load and operating conditions of the motor.

[0041] In another possible implementation, the data collected during operation may include speed data, torque data, motor bus voltage data, and motor bus current data.

[0042] In another possible implementation, the operational data acquisition includes speed acquisition data and motor output power.

[0043] Here, coolant flow rate refers to the amount of coolant flowing through the coolant pipe within a unit. In this embodiment, two types of coolant flow rate data are included: one is coolant flow rate acquisition data, which can be acquired, for example, by a coolant flow rate sensor; the other is coolant flow rate control data, calculated using the motor temperature control method provided in this embodiment, and used to adjust the coolant flow rate. For example, multiple coolant flow rate acquisition data can refer to the coolant flow rate acquisition data obtained by the vehicle controller 102 within a preset time period before the target time and at the target time.

[0044] Step 204: Based on the target runtime sequence data and multiple coolant flow acquisition data, the target flow control data is obtained by iteratively optimizing the output data of the motor temperature prediction model by minimizing the output data.

[0045] The motor temperature prediction model is used to divide the input runtime sequence data and coolant flow time series data into multiple runtime subsequences and multiple coolant flow subsequences according to a preset time window. It uses a portion of the data in each runtime subsequence and each coolant flow subsequence to perform graph convolution feature extraction and temperature prediction processing, and outputs temperature prediction data corresponding to the input runtime sequence data and coolant flow time series data.

[0046] The output data of the motor temperature prediction model includes the temperature at at least one location on the vehicle motor at at least one prediction time. In one possible implementation, the output data includes the surface temperature value of the motor at a prediction time corresponding to the target time; for example, the prediction time is a time after the target time. In another possible implementation, the output data includes the temperature values ​​at multiple locations on the motor surface at the prediction time corresponding to the target time. In yet another possible implementation, the output data includes the surface temperature values ​​of the motor at multiple prediction times corresponding to the target time, wherein the multiple prediction times are multiple times after the target time; for example, the multiple prediction times are 10 prediction times, and the output data is in the form of [(prediction time 1, temperature 1), (prediction time 2, temperature 2), ..., (prediction time 10, temperature 10)]. In yet another possible implementation, the output data includes the temperature values ​​at multiple locations on the motor surface at multiple prediction times corresponding to the target time; for example, the output data includes the temperature values ​​at 5 locations on the motor surface at 10 prediction times corresponding to the target time.

[0047] In one possible implementation, where the output data of the motor temperature prediction model includes multiple temperature values, minimization during the iteration process may refer to minimizing the maximum value among the multiple temperature prediction values ​​output by the motor temperature prediction model.

[0048] In this embodiment, the runtime sequence data and the coolant flow time sequence data are first divided into multiple runtime subsequences and multiple coolant flow subsequences according to a preset time window. For example, the number of data in each runtime subsequence is the same, and the number of data in each coolant flow subsequence is the same. The amount of data in the runtime subsequence and the number of data in the coolant subsequence can be the same or different, and can be determined according to the acquisition frequency of the runtime acquisition data and the acquisition frequency of the coolant flow acquisition data.

[0049] In this embodiment, considering that the motor operating status and coolant flow rate do not change much in a very short time, for each time window, a portion of data is selected from the operating subsequence and the coolant flow rate subsequence as representative data to participate in the subsequent graph convolution feature extraction process and temperature prediction process, so as to save the computational load of the feature extraction process and temperature prediction process and improve the efficiency of temperature prediction.

[0050] In one possible implementation, the time window is less than 3 seconds. Alternatively, the time window is 2 seconds.

[0051] In each iteration of optimization, a portion of the coolant flow rate subsequence in the coolant flow rate time series data input to the motor temperature prediction model includes coolant flow rate candidate data corresponding to the current iteration of optimization.

[0052] In this process, minimizing the output data of the motor temperature prediction model is used as the objective function for iterative optimization. The candidate data of coolant flow rate during the iteration process needs to be used as part of the input of the motor temperature prediction model to determine the motor temperature prediction data corresponding to the candidate data of coolant flow rate, which is then used as the fitness value of the candidate data of coolant flow rate to determine whether the candidate data of coolant flow rate is the optimal solution of the current iteration process.

[0053] In each iteration of optimization, the runtime sequence data input to the motor temperature prediction model is the target runtime sequence data, and the input coolant flow rate time sequence data is determined based on multiple coolant flow rate acquisition data and the coolant flow rate candidate data of the current iteration. In one possible implementation, the coolant flow rate candidate data of the current iteration is selected as the coolant flow rate data in the coolant flow rate time sequence data that is temporally closest to the target time. The coolant flow rate data in the remaining coolant flow rate subsequences are the acquired coolant flow rate data, ensuring that the coolant flow rate candidate data participates in the graph convolution feature extraction process and the temperature prediction process. In another possible implementation, the coolant flow rate candidate data corresponding to the current iteration is selected as the coolant flow rate data in the two or more coolant flow rate subsequences that are temporally closest to the target time in the coolant flow rate time sequence data. The coolant flow rate data in the remaining coolant flow rate subsequences are the acquired coolant flow rate data, ensuring that the coolant flow rate candidate data participates in the graph convolution feature extraction process and the temperature prediction process.

[0054] For ease of description, each cold liquid subsequence contains m data points, and there are n cold liquid subsequences in total. The total number of data points in the entire cold liquid flow time series data is n×m. In one possible implementation, the data points closest to the target time in the cold liquid flow time series data (a total of n×m) input based on the motor temperature prediction model are filled with cold liquid flow candidate data. For example, the cold liquid flow data in the cold liquid flow subsequence closest to the target time is the cold liquid flow candidate data. Thus, the multiple cold liquid flow acquisition data points obtained in step 204 can be (n-1)×m data points that are temporally far from the target time within a preset time period before the target time.

[0055] In another possible implementation, the cold liquid flow rate data obtained in step 204 can be n×m data points. When constructing the input data for the input motor temperature prediction model, candidate cold liquid flow rate data are used to replace the cold liquid flow rate data in the corresponding cold liquid subsequence.

[0056] For example, the iterative optimization process can be described by the following formula:

[0057] Minimize max(f(x1,x2)), with the constraint: x 2L <x2<x 2H ;

[0058] Where x1 refers to runtime timing data, x2 refers to coolant timing data, and x 2L This refers to the minimum settable value for the coolant flow rate, x. 2H This refers to the maximum value that the coolant flow rate can be set, and f(x1,x2) represents the output data of the motor temperature prediction model.

[0059] Among them, the optimal coolant flow control data can be found by using an optimization algorithm based on the target operating data corresponding to the target time and multiple coolant flow acquisition data. This optimal coolant flow control data is then used as the target flow control data corresponding to the target time. When this target flow control data is used as part of the coolant flow time series data, the temperature value output by the motor temperature prediction model is minimized.

[0060] In one possible implementation, the whale optimization algorithm is used for iterative optimization to obtain the target flow control data. The whale optimization algorithm comprises three actions: surrounding the prey, spiraling attack on the prey, and randomly searching for the prey.

[0061] Here, "encircling the prey" refers to using the current best candidate solution as the target prey (i.e., the optimal solution); after knowing the prey's location, the whale pod begins to update its position based on its current relationship with the prey's position; the action of encircling the prey can be performed using Formula 1 and the formula...

[0062] Equation 2 means:

[0063]

[0064] Where t represents the current iteration number, and This indicates the position of the optimal candidate solution before generation t. and Let represent the solutions for generation t and the next generation, respectively. and The coefficient vector needs to be calculated separately. and The calculation methods are shown in Formula 3 and Formula 4 respectively:

[0065]

[0066] in, and Represents two random numbers between 0 and 1. For numbers that decrease linearly from 2 to 0 with each iteration, The calculation formula is shown in Formula 5:

[0067]

[0068] Where G represents the total number of iterations.

[0069] As a system vector, it represents a control factor, also known as the amplification factor. The amplification factor is an important parameter used to adjust the step size of whale pod members. During the update process of the whale algorithm, each whale updates its position based on the relationship between its current position and the position of the target prey (the optimal solution). Amplification Factor This process involves scaling the step size. By adjusting the magnification factor, the range and speed of the whale's movement can be controlled.

[0070] Among them, the spiral attack refers to simulating the spiral process of a humpback whale, that is, the foam net attack process; by establishing a spiral equation between the whale and its prey, the spiral movement of the humpback whale is simulated; the total spiral motion can be expressed by formulas 6 and 7:

[0071]

[0072] Where b is a constant, usually set to 1. l is a random number between [-1, 1], which can be determined according to formula 8:

[0073] l=(a2-1)·r3+1, Formula 8 where, Where r3 is a random number between 0 and 1.

[0074] The choice between surrounding the prey and spiraling attack is made randomly. Surrounding the prey is chosen when the probability p < 0.5, and spiraling attack is chosen otherwise. This can be represented by Formula 9:

[0075]

[0076] The random search action is a method used in the whale algorithm for global search and to escape local optima; the random search action can be represented by Equations 10 and 11:

[0077]

[0078] in, For the randomly selected whale (solution), it can be seen that the formula for random search is very similar to the formula for surrounding prey, the only difference being that the direction is changed from the current optimal candidate solution to a random whale. Whether the surrounding prey and random search behaviors occur depends on the value of A. Employing prey-encircling behavior; when A random search behavior is adopted; that is, in Formula 9, if p < 0.5, Then, the random search action represented by Formulas 10 and 11 is used. The action of surrounding the prey is represented by Formula 1 and Formula 2.

[0079] Among them, when When, explain The value of is close to 0 and between -1 and 1. In this case, choosing to surround the prey means adopting a strategy towards the prey or target object to approach, surround, or concentrate the target. When When, explain If the value of is greater than 1 and greater than any value close to 0, then choosing random search behavior means adopting an exploratory strategy to find new possibilities or solutions.

[0080] In one possible implementation, the target flow control data is obtained by iterative optimization using other swarm intelligence optimization algorithms such as the bat optimization algorithm or the fruit fly optimization algorithm.

[0081] Step 206: Control the coolant flow rate used to cool the vehicle motor based on the target flow control data.

[0082] Among them, the target flow control data is the coolant flow control value that minimizes the output data of the motor temperature prediction model at the target time. Controlling the coolant flow of the vehicle motor based on the target flow control data can keep the temperature of the vehicle motor at the predicted time as low as possible, thereby improving the motor's working efficiency and service life.

[0083] The vehicle motor temperature control method provided in the above embodiments includes acquiring target runtime timing data and multiple coolant flow rate acquisition data corresponding to the vehicle motor at a target time. The target runtime timing data includes multiple runtime acquisition data. Based on the target runtime timing data and the multiple coolant flow rate acquisition data, iterative optimization processing is performed by minimizing the output data of the motor temperature prediction model to obtain target flow rate control data. The coolant flow rate used to cool the vehicle motor is controlled based on the target flow rate control data. The motor temperature prediction model is used to divide the input runtime timing data and coolant flow rate time-series data into multiple runtime subsequences and multiple coolant flow rate subsequences according to a preset time window. Partial data from each runtime subsequence and each coolant flow rate subsequence are used for graph convolution feature extraction and temperature prediction processing. The output is temperature prediction data corresponding to the input runtime timing data and coolant flow rate time-series data. In each iterative optimization process, the coolant flow rate subsequences in the input coolant flow rate time-series data of the motor temperature prediction model include the coolant flow rate... The data represents the candidate coolant flow rate data corresponding to the current iterative optimization process. Determining the target flow control data requires multiple iterations. Each iteration uses a motor temperature prediction model to obtain the temperature prediction data corresponding to the candidate coolant flow rate data for the current iteration. In other words, determining the target flow control data requires calling the motor temperature prediction model multiple times for calculation. The motor temperature prediction model provided in the above embodiment divides the input time-series data into multiple subsequences according to a preset time window. It uses a portion of the data from each subsequence for graph convolution feature extraction and temperature prediction. This avoids the problem in related technologies where the digital twin model of the motor temperature field requires feature extraction for each piece of input time-series data, resulting in high computational complexity and low control efficiency. Using the motor temperature prediction model provided in the above embodiment reduces the computational load for each iteration, thereby reducing the overall computational load for determining the target flow control data, increasing the speed of obtaining the target flow control data, and resulting in higher control efficiency.

[0084] In one exemplary embodiment, based on Figure 2 The illustrated embodiment describes a process of obtaining target flow control data by iteratively optimizing the output data of a motor temperature prediction model based on target runtime sequence data and multiple coolant flow rate acquisition data. Please refer to [link / reference]. Figure 3 The process includes steps 302 to 304, wherein:

[0085] Step 302: Based on multiple cold liquid flow rate acquisition data and cold liquid flow rate candidate data in the current iteration process, candidate flow rate time series data are obtained. The flow rate data corresponding to at least one time window that is close to the target time in the candidate flow rate time series data are all cold liquid flow rate candidate data.

[0086] In one possible implementation, the flow data corresponding to the first time window closest to the target time in the candidate flow time series data are all candidate cold liquid flow data.

[0087] In one possible implementation, the flow data corresponding to the first and second time windows that are close to the target time in the candidate flow time series data are both cold liquid flow data.

[0088] Step 304: Input the target runtime timing data and candidate flow timing data into the motor temperature prediction model to obtain the candidate temperature prediction data corresponding to the candidate coolant flow rate data.

[0089] In one possible implementation, during the process of inputting the target runtime timing data and candidate flow timing data into the motor temperature prediction model, the target runtime timing data and candidate flow timing data are first normalized, resulting in normalized target runtime timing data and normalized candidate flow timing data, which are then input into the motor temperature prediction model. By normalizing the target runtime timing data and candidate flow timing data, dimensional data is converted into dimensionless data, facilitating the standardization of data scale during the subsequent feature extraction stage.

[0090] In one possible implementation, the motor temperature prediction model includes a normalization layer for normalizing the input data.

[0091] Step 306: Based on the candidate cold liquid flow rate data and the candidate temperature prediction data corresponding to the candidate cold liquid flow rate data, obtain the target flow control data.

[0092] In one possible implementation, if the current iteration process satisfies the iteration termination condition, and the candidate temperature prediction data is better than the optimal candidate solution before the current iteration process, then the cold liquid flow rate candidate data is the target flow control data; if the current iteration process satisfies the iteration termination condition, and the prediction data is better than the optimal candidate solution before the current iteration process, then the target flow control data is determined based on the optimal candidate solution. For example, the iteration termination condition includes reaching the maximum number of iterations or other preset termination conditions.

[0093] In one possible implementation, if the current iteration process does not meet the iteration termination condition, the candidate temperature prediction data is compared with the optimal candidate solution before the current iteration process, and the subsequent iteration process continues. That is, step 302 and step 304 are executed for the candidate cold liquid flow rate corresponding to the next iteration process until the iteration termination condition is met and the target flow control data is obtained.

[0094] In this embodiment, by using the flow data corresponding to at least one time window that is close to the target time in time sequence as cold liquid flow candidate data, the motor temperature prediction model divides the input candidate flow time sequence data into multiple candidate flow subsequences according to the preset time window. At least one candidate flow subsequence contains all cold liquid candidate flow data. In this way, when using some data in the candidate flow subsequence for graph convolution feature extraction processing, it is guaranteed that cold liquid candidate flow data participates in the graph convolution feature extraction processing.

[0095] In one exemplary embodiment, based on Figure 3 The embodiment shown includes an input layer, a graph convolutional layer, and a prediction layer for the motor temperature prediction model. Please refer to [reference needed]. Figure 4 This embodiment involves inputting target runtime timing data and candidate flow timing data into a motor temperature prediction model to obtain candidate temperature prediction data corresponding to the candidate coolant flow rate data. For example... Figure 4 As shown, the process includes:

[0096] Step 402: The target running time sequence data and candidate flow time sequence data are divided into multiple target running subsequences and multiple candidate flow subsequences by the input layer of the motor temperature prediction model according to the time window.

[0097] In one possible implementation, the input layer includes a normalization sublayer and a partitioning sublayer; the corresponding step 402 further includes: normalizing the target runtime timing data and candidate traffic timing data through the normalization sublayer to obtain normalized target runtime timing data and normalized candidate traffic timing data; and partitioning the normalized target runtime timing data and normalized candidate traffic timing data according to a time window through the partitioning sublayer to obtain multiple target runtime subsequences and multiple candidate traffic subsequences.

[0098] In one possible implementation, the target runtime timing data and candidate traffic timing data input to the input layer are normalized data.

[0099] Step 404: The graph convolutional layer of the motor temperature prediction model is used to extract graph convolutional features from some data in each target running subsequence and each candidate flow subsequence, so as to obtain the target running convolutional features corresponding to the target running time series data and the candidate flow convolutional features corresponding to the candidate flow time series data.

[0100] In one possible implementation, the graph convolutional layer includes a node sub-layer, a first graph convolutional sub-layer, and a second graph convolutional sub-layer, and step 402 further includes:

[0101] Step A1: Through the node sub-layers in the graph convolutional layer, at least one running data collection point is determined from each target running sub-sequence as the target running representative node, and at least one traffic data collection point or traffic candidate data is determined from each candidate traffic sub-sequence as the candidate traffic representative node.

[0102] For example, taking a target running subsequence as an example, one running data collection point is randomly selected from the target running subsequence as the target running representative node of the target running subsequence. As another example, taking a target running subsequence as an example, two running data collection points are randomly selected from the target running subsequence as the target running representative nodes of the target running subsequence.

[0103] Step A2 involves extracting graph convolutional features from each target running representative node and each candidate traffic representative node using the first graph convolutional sub-layer in the graph convolutional layer, thereby obtaining the intermediate running convolutional features corresponding to each target running representative node and the intermediate traffic convolutional features corresponding to each candidate traffic representative node.

[0104] In one possible implementation, the process of extracting graph convolutional features from each target running representative node through a first graph convolutional sub-layer to obtain the intermediate running convolutional features corresponding to each target running representative node includes:

[0105] Step A21: Based on the distance priority principle, determine a first number of cold liquid flow data from the candidate flow subsequence that belongs to the same time window as the target running representative node, wherein the first number of cold liquid flow data includes cold liquid flow acquisition data or cold liquid flow candidate data.

[0106] For example, the Euclidean distance between each cold liquid flow data in the candidate flow subsequence and the target running representative node is calculated, and a first number of cold liquid flow data are obtained by using the TOP-K nearest neighbor method, where K represents the first number.

[0107] For another example, the cosine similarity data between each cold liquid flow rate data in the candidate flow rate subsequence and the target running representative node is calculated, and the first number of cold liquid flow rate data are obtained by using the TOP-K nearest neighbor method.

[0108] Step A22: Based on the distance data between the target running representative node and the first number of cold liquid flow data, obtain the first adjacency matrix corresponding to the target running representative node.

[0109] In this embodiment, the value of each element in the first adjacency matrix is ​​the distance between the target running representative node and the corresponding chilled liquid flow data. Compared with the case where the value of each element in the adjacency matrix in related technologies is 0 or 1, using the first adjacency matrix in this embodiment for subsequent convolution aggregation processing can better represent the spatial relationship between different perspectives (running data and chilled liquid flow data) and improve the accuracy of subsequent temperature prediction.

[0110] For example, each running representative node is a running data collection point. Step A21 determines five cold liquid flow rate data points that are closest to the running data collection point. Correspondingly, the first adjacency matrix is ​​a matrix with a dimension of 11×11.

[0111] Step A23: Perform convolution aggregation processing based on the first adjacency matrix corresponding to the target running representative node, the target running representative node, and the first number of traffic data to obtain intermediate running convolution features.

[0112] For example, the convolution aggregation process can be represented by Equation 12:

[0113]

[0114] Among them, H 运行i A represents the intermediate convolutional features corresponding to the i-th target node. 运行i X is the adjacency matrix corresponding to the i-th target node. 运行i It is the node feature corresponding to the i-th target running representative node. The node feature includes the specific value corresponding to the i-th target running representative node and the specific value corresponding to the first number of cold liquid flow data. This represents a symmetric normalization operation on the adjacency matrix. W 运行i Let σ represent the weight matrix corresponding to the i-th running node, and σ be the ReLU activation function.

[0115] Correspondingly, the same method is used, similar to steps A21 to A23, to obtain the convolutional features of each intermediate flow.

[0116] Step A3 involves extracting graph convolution features from each intermediate running convolution feature and each intermediate flow convolution feature using the second graph convolution sub-layer in the graph convolution layer, thereby obtaining the target running convolution feature and the candidate flow convolution feature.

[0117] In one possible implementation, the process of extracting graph convolutional features from each intermediate running convolutional feature and each intermediate traffic convolutional feature through a second graph convolutional sublayer in the graph convolutional layer to obtain the target running convolutional feature and candidate traffic convolutional feature includes:

[0118] Step A31: The intermediate running convolutional features and intermediate flow convolutional features are concatenated according to the time window to obtain the concatenated features corresponding to each time window.

[0119] Specifically, for each time window, the intermediate running convolutional features and intermediate flow convolutional features corresponding to that time window are concatenated to obtain the concatenated features for that time window. In the convolutional sub-layer of the second graph, each time window is treated as a node, and the concatenated features corresponding to that time window are used as the node features for that node.

[0120] Step A32: Perform similarity calculation on each splicing feature to obtain the second adjacency matrix.

[0121] In this second adjacency matrix, each element represents the similarity data between the corresponding two concatenated features. For example, there are n time windows, and the second adjacency matrix is ​​an n×n matrix.

[0122] For example, the similarity data between each spliced ​​feature is calculated using cosine similarity, and used as the value of each element in the second adjacency matrix. As another example, the distance data between each spliced ​​feature is calculated using cosine similarity, and used as the value of each element in the second critical matrix.

[0123] Step A33: Perform convolution aggregation processing based on each intermediate running convolution feature, each intermediate traffic convolution feature, and the second adjacency matrix to obtain the target running convolution feature and the candidate traffic convolution feature.

[0124] Specifically, convolution aggregation is performed based on the second adjacency matrix and each intermediate running convolution feature to obtain the target running convolution feature, and convolution aggregation is performed based on the second adjacency matrix and each intermediate traffic convolution feature to obtain the candidate traffic convolution feature.

[0125] In this embodiment, the intermediate running convolutional features and intermediate flow convolutional features within each time window are concatenated using time windows as nodes. Then, the second adjacency matrix between each concatenated feature is calculated. Using the second adjacency matrix as a higher-order matrix can better capture the relationship between different time windows and the commonalities among various features. Then, by performing convolution aggregation processing on the second adjacency matrix and each intermediate running convolutional feature and each intermediate flow convolutional feature, the resulting target running convolutional feature and candidate flow convolutional feature can effectively integrate the information from different time windows and capture the commonalities and relationships between features from different perspectives, providing a more accurate and comprehensive foundation for subsequent temperature prediction processing.

[0126] In another possible implementation, the graph convolutional layer includes a node sub-layer and a first graph convolutional layer; compared with the previous implementation, the intermediate running convolutional features in the previous implementation are concatenated to obtain the target running convolutional features, and the intermediate traffic convolutional features are concatenated to obtain the candidate traffic convolutional features.

[0127] In another possible implementation, the graph convolutional layer includes a node sub-layer, a first graph convolutional layer, a second graph convolutional layer, and a third graph convolutional layer. In this implementation, the output data of the second graph convolutional layer in step A33 is input into the third graph convolutional layer for convolutional aggregation processing again to obtain the target running convolutional features and candidate traffic convolutional features.

[0128] Step 406: The target running convolutional features and candidate flow convolutional features are predicted by the prediction layer of the motor temperature prediction model to obtain candidate temperature prediction data.

[0129] In one possible implementation, the process of obtaining candidate temperature prediction data by predicting the target operating convolutional features and candidate flow convolutional features through the prediction layer of the motor temperature prediction model includes:

[0130] Step B1 involves concatenating the target running convolutional features and candidate flow convolutional features through a concatenation sub-layer in the prediction layer to obtain concatenated candidate features.

[0131] Specifically, the target running convolutional features and candidate flow convolutional features are concatenated to obtain concatenated candidate features, which are then used as a comprehensive feature representation and input into the prediction sublayer.

[0132] Step B2 involves performing prediction processing on the spliced ​​candidate features through the prediction sub-layer in the prediction layer to obtain candidate temperature prediction data, wherein the candidate temperature prediction data includes the temperature corresponding to at least one position in the vehicle motor at at least one prediction time.

[0133] For example, the prediction sublayer is a fully connected layer.

[0134] In an exemplary embodiment, the provided vehicle motor temperature control method further includes a training process for a motor temperature prediction model, comprising steps C1 to C3, wherein:

[0135] Step C1: Collect time-series data of the vehicle motor under different operating conditions to obtain training dataset and test dataset. The time-series data includes runtime acquisition time-series data, coolant flow time-series data, and motor temperature time-series data.

[0136] Step C2: Use the training dataset to iteratively learn the initial prediction model until the iterative learning termination condition is met, and obtain the target prediction model.

[0137] The loss function for iterative learning is the root mean square error, which measures the deviation between the predicted and the true values. The formula for calculating the loss function for iterative learning is: Among them, y i Indicates the predicted value; The actual value is represented by N; N represents the number of predictions. For example, the number of predictions is the temperature values ​​at 10 locations on the motor surface, with N being 10.

[0138] Step C3: Input the test dataset into the target prediction model to obtain the output data of the target prediction model. Determine whether the output data of the target prediction model meets the preset evaluation index. If it does not meet the preset evaluation index, use the target prediction model as the initial prediction model to continue iterative learning until the preset evaluation index is met. Then, use the target prediction model as the motor temperature prediction model.

[0139] In one exemplary embodiment, such as Figure 5 As shown, a vehicle motor temperature control method is provided, which is applied to... Figure 1 The method will be described using the vehicle controller 102 as an example, including steps 502 to 518. The motor temperature control method involves a motor temperature prediction model comprising an input layer, a graph convolutional layer, and a prediction layer. The graph convolutional layer includes a node sub-layer, a first graph convolutional layer, and a second graph convolutional layer. The prediction layer includes a stitching sub-layer and a prediction sub-layer.

[0140] Step 502: Obtain the target speed time-series data, target torque time-series data, and multiple coolant flow rate acquisition data corresponding to the vehicle motor at the target time. The target speed time-series data includes multiple speed acquisition data, and the target torque time-series data includes multiple torque acquisition data.

[0141] Step 504: Based on multiple cold liquid flow rate acquisition data and cold liquid flow rate candidate data in the current iteration process, candidate flow rate time series data are obtained. The flow rate data corresponding to at least one time window that is close to the target time in the candidate flow rate time series data are all cold liquid flow rate candidate data.

[0142] Step 506: The target speed time series data, target torque time series data, and candidate flow time series data are divided into multiple target speed subsequences, multiple target torque subsequences, and multiple candidate flow subsequences according to a preset time window through the input layer of the motor temperature prediction model.

[0143] Step 508: Using the node sub-layers in the graph convolutional layer of the motor temperature prediction model, at least one speed acquisition data point is determined from each target speed sub-sequence as a representative node for each target speed, at least one torque acquisition data point is determined from each target torque sub-sequence as a representative node for each target torque, and at least one flow acquisition data point or flow candidate data point is determined from each candidate flow sub-sequence as a representative node for each candidate flow.

[0144] Step 510: The first graph convolutional sub-layer in the graph convolutional layer performs graph convolutional feature extraction on each target rotational speed representative node, each target torque representative node, and each candidate flow rate representative node to obtain the intermediate rotational speed convolutional features corresponding to each target rotational speed representative node, the intermediate torque convolutional features corresponding to each target torque representative node, and the intermediate flow rate convolutional features corresponding to each candidate flow rate representative node.

[0145] Optionally, step 510 includes: determining a first number of cold liquid flow data from candidate flow subsequences belonging to the same time window as the target running representative node based on the distance priority principle, wherein the first number of cold liquid flow data includes cold liquid flow acquisition data or cold liquid flow candidate data; obtaining a first adjacency matrix corresponding to the target running representative node based on the distance data between the target running representative node and the first number of cold liquid flow data; and performing convolution aggregation processing based on the first adjacency matrix corresponding to the target running representative node, the target running representative node, and the first number of cold liquid flow data to obtain intermediate running convolution features.

[0146] Please combine Figure 6 and Figure 7 This is a schematic diagram illustrating the implementation process of steps 506 to 510 in one optional embodiment. Figure 6 As shown, taking the target speed subsequence corresponding to the i-th time window as an example, a speed acquisition data point is determined from this target speed subsequence as the representative node of the target speed. Then, the first quantity is determined from the target torque subsequence and the candidate flow rate subsequence corresponding to the i-th time window respectively through Euclidean distance or other similarity calculation methods. Figure 6 (Taking a first quantity of 5 as an example for illustration) Connect the data points to the node representing the target rotational speed to obtain the rotational speed graph structure G within the i-th time window. 转i =(A 转i ,X 转i ), where i = 1, 2, ..., n, n represents the number of time windows, A 转i The first adjacency matrix corresponding to the target rotational speed in the i-th time window, X represents the target rotational speed in the i-th time window. 转i This represents the node characteristics corresponding to all nodes in the rotational speed diagram. Please refer to... Figure 6Using the same method as the target speed subsequence, the torque diagram structure G corresponding to the i-th time window is obtained. 扭i =(A 扭i ,X 扭i The flow graph structure G corresponding to the i-th time window. 冷i =(A 冷i ,X 冷i ).

[0147] like Figure 7 As shown, graph convolutional feature fusion extraction is performed on the three graph structures in each time window. Spatial features guided by different graph structures under each time window are extracted through graph convolutional layers, thereby obtaining the relationship information between the three types of time-series data from different perspectives. This fully utilizes information from multiple graph structures to achieve feature fusion and extraction. The graph convolutional feature fusion extraction process in step 510 can be expressed by the following formula:

[0148]

[0149] Among them, H 转i H 扭i and H 冷i These represent the intermediate rotational speed convolution feature, intermediate torque convolution feature, and intermediate flow rate convolution feature corresponding to the i-th time window, respectively.

[0150] Step 512: The graph convolution features of each intermediate rotational speed, each intermediate torque, and each intermediate flow rate are extracted by the second graph convolution sub-layer in the graph convolution layer to obtain the target rotational speed convolution features, the target torque convolution features, and the candidate flow rate convolution features.

[0151] Optionally, step 512 further includes: concatenating the convolutional features of each intermediate rotational speed, each intermediate torque, and each intermediate flow rate according to the time window to obtain the concatenated features corresponding to each time window; performing similarity calculation on each concatenated feature to obtain the second adjacency matrix; and performing convolution aggregation processing based on the convolutional features of each intermediate rotational speed, each intermediate torque, each intermediate flow rate, and the second adjacency matrix to obtain the target rotational speed convolutional feature, the target torque convolutional feature, and the candidate flow rate convolutional feature.

[0152] Please refer to Figure 8 This is a schematic diagram illustrating the implementation process of step 512 in one optional implementation method. For example... Figure 8 As shown, n time windows are treated as n nodes. The intermediate rotational speed convolutional features, intermediate torque convolutional features, and intermediate flow rate convolutional features corresponding to each time window are concatenated to form the node features of each time window. Then, the similarity between each node is calculated using cosine similarity, resulting in an n×n similarity matrix, which serves as the second adjacency matrix A.H In this way, the second adjacency matrix obtained by fusing the similarity of the three features can better capture the connections between different time periods and the commonalities among the three features; then, the intermediate rotational speed convolution features corresponding to the n time windows and the second adjacency matrix are combined to form a high-order rotational speed map structure G. 转 =(A H H 转i The intermediate torque convolution features and the second adjacency matrix corresponding to the n time windows are combined to form a high-order torque graph structure G. 扭 =(A H H 扭i The intermediate flow convolution features and the second adjacency matrix corresponding to the n time windows are combined to form a higher-order flow graph structure G. 冷 =(A H H 冷i Then, graph convolution is used to aggregate the features, resulting in three types of higher-order features: target rotational speed convolution feature, target torque convolution feature, and candidate flow rate convolution feature. The process of obtaining these three types of higher-order features can be represented by the following formula:

[0153]

[0154] Where H1 represents the target rotational speed convolution feature, H2 represents the target torque convolution feature, and H3 represents the candidate flow rate convolution feature. 高1 W 高2 and W 高3 These are the weight matrices corresponding to the high-order speed diagram structure, the high-order torque diagram structure, and the high-order flow diagram structure, respectively.

[0155] Step 514: The target running convolutional features and candidate flow convolutional features are concatenated through the concatenation sub-layer in the prediction layer to obtain concatenated candidate features.

[0156] Step 516: The candidate features are processed by the prediction sub-layer in the prediction layer to obtain candidate temperature prediction data, wherein the candidate temperature prediction data includes the temperature of at least one position in the vehicle motor at at least one prediction time.

[0157] Step 518: Based on the candidate coolant flow rate data and the corresponding candidate temperature prediction data, the target flow control data is obtained. The target flow control data is the candidate coolant flow rate data that minimizes the candidate temperature prediction data output by the motor temperature prediction model during each iteration.

[0158] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0159] It is understood that the term "based on" as used in this application is used to describe one or more factors that influence the determination, but does not exclude other factors that may influence the determination. For example, the phrase "determine A based on B" means that the determination of A can be based entirely or at least partially on factor B. That is, B is a factor that influences the determination of A, but does not exclude the fact that the determination of A is also based on C.

[0160] In one exemplary embodiment, please refer to Figure 9 This paper provides a method for predicting the temperature of a vehicle motor. This method is applied to… Figure 1 The following steps are used as an example to illustrate the process of using the vehicle controller 102, including steps 902 to 906. Wherein:

[0161] Step 902: Obtain the running sequence data and coolant flow time sequence data of the vehicle motor at the target time.

[0162] Step 906: Divide the runtime sequence data and coolant flow time sequence data into multiple runtime subsequences and multiple coolant flow subsequences according to the preset time window.

[0163] Step 908: Perform graph convolution feature extraction on some data in each running subsequence and each cold liquid flow rate subsequence to obtain the target running convolution feature and target flow rate convolution feature corresponding to the target time.

[0164] Step 910: Perform prediction processing based on the target running convolutional features and the target flow convolutional features to obtain the target predicted temperature data of the vehicle motor at the target time.

[0165] The vehicle motor temperature prediction method provided in the above embodiments utilizes a data-driven approach. It divides the input time-series data into multiple sub-sequences according to a preset time window, and uses partial data from each sub-sequence for graph convolution feature extraction and temperature prediction. This avoids the need for feature extraction processing on every single data point in the input time-series data, which is computationally intensive and inefficient when using digital twin models of the motor temperature field for motor temperature prediction. The vehicle motor temperature prediction method provided in the above embodiments improves the efficiency of temperature prediction.

[0166] In one exemplary embodiment, based on Figure 9 In the illustrated embodiment, this embodiment involves performing graph convolution feature extraction processing on partial data from each operational subsequence and each chilled liquid flow rate subsequence to obtain the target operational convolution feature and target flow rate convolution feature corresponding to the target time. This process includes:

[0167] Step D1: Determine at least one running data collection point from each running subsequence as each running representative node, and determine at least one traffic data collection point from each traffic subsequence as each traffic node.

[0168] Step D2 involves performing graph convolution feature extraction on each running representative node and each traffic representative node to obtain the intermediate running convolution features corresponding to each running representative node and the intermediate traffic convolution features corresponding to each traffic representative node.

[0169] Step D3 involves performing graph convolution feature extraction on each intermediate running convolution feature and each intermediate traffic convolution feature to obtain the target running convolution feature and the target traffic convolution feature.

[0170] In an exemplary embodiment, the process involves extracting graph convolutional features from each representative node in step D2 to obtain the intermediate convolutional features corresponding to each representative node. This process includes:

[0171] Step D21: Based on the distance priority principle, determine the first number of traffic collection data from the traffic subsequence that belongs to the same time window as the running representative node.

[0172] Step D22: Based on the distance data between the running representative node and the first number of traffic collection data, obtain the first adjacency matrix corresponding to the running representative node.

[0173] Step D23: Perform convolution aggregation processing based on the first adjacency matrix corresponding to the running representative node, the running representative node, and the first number of traffic collection data to obtain intermediate running convolution features.

[0174] In an exemplary embodiment, step D3 involves performing graph convolution feature extraction on each intermediate running convolution feature and each intermediate traffic convolution feature to obtain the target running convolution feature and the target traffic convolution feature. This process includes:

[0175] Step D31: The intermediate running convolutional features and intermediate flow convolutional features are concatenated according to the time window to obtain the concatenated features corresponding to each time window.

[0176] Step D32: Perform similarity calculation on each splicing feature to obtain the second adjacency matrix.

[0177] Step D33: Perform convolution aggregation processing based on each intermediate running convolution feature, each intermediate traffic convolution feature, and the second adjacency matrix to obtain the target running convolution feature and the target traffic convolution feature.

[0178] In one exemplary embodiment, based on Figure 9 The embodiment shown here relates to step 908, which involves predicting the vehicle motor's predicted temperature data at the target time based on the target running convolutional features and the target flow convolutional features. This process includes:

[0179] Step E1 involves concatenating the target running convolutional features and the target traffic convolutional features to obtain concatenated candidate features.

[0180] Step E2: Based on the spliced ​​candidate features, perform prediction processing to obtain the target predicted temperature data, wherein the target predicted temperature data includes the temperature corresponding to at least one position in the vehicle motor at at least one prediction time.

[0181] In one exemplary embodiment, based on Figure 9 In one embodiment, the vehicle motor temperature prediction method includes speed timing data and torque timing data in the runtime timing data.

[0182] Based on the same inventive concept, this application also provides a vehicle motor temperature control device for implementing the vehicle motor temperature control method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations of one or more vehicle motor temperature control device embodiments provided below can be found in the limitations of the vehicle motor temperature control method described above, and will not be repeated here.

[0183] In one exemplary embodiment, such as Figure 10 As shown, a vehicle motor temperature control device is provided, including: a data acquisition module 1002, a flow optimization module 1004, and a flow control module 1006, wherein:

[0184] The data acquisition module 1002 is used to acquire the target running sequence data of the vehicle motor at the target time and multiple coolant flow acquisition data, wherein the target running sequence data includes multiple running acquisition data;

[0185] The flow optimization module 1004 is used to perform iterative optimization based on the target runtime sequence data and multiple cold liquid flow acquisition data by minimizing the output data of the motor temperature prediction model to obtain the target flow control data. The motor temperature prediction model divides the input runtime sequence data and cold liquid flow time-series data into multiple runtime subsequences and multiple cold liquid flow subsequences according to a preset time window. It uses a portion of the data from each runtime subsequence and each cold liquid flow subsequence for graph convolution feature extraction and temperature prediction processing, outputting temperature prediction data corresponding to the input runtime sequence data and cold liquid flow time-series data. In each iteration of optimization, the cold liquid flow data included in the portion of the cold liquid flow time-series data input to the motor temperature prediction model constitutes the candidate cold liquid flow data for the current iteration.

[0186] The flow control module 1006 is used to control the flow rate of coolant used to cool the vehicle motor based on the target flow control data.

[0187] In an exemplary embodiment, the flow optimization module 1004 is used to obtain candidate flow time series data based on multiple cold liquid flow collection data and cold liquid flow candidate data in the current iteration process. The flow data corresponding to at least one time window that is close to the target time in the candidate flow time series data are all cold liquid flow candidate data. The target running time series data and the candidate flow time series data are input into the motor temperature prediction model to obtain candidate temperature prediction data corresponding to the cold liquid flow candidate data. Based on the cold liquid flow candidate data and the candidate temperature prediction data corresponding to the cold liquid flow candidate data, the target flow control data is obtained.

[0188] In an exemplary embodiment, the flow optimization module 1004 is used to divide the target running time sequence data and candidate flow time sequence data according to the time window through the input layer of the motor temperature prediction model to obtain multiple target running subsequences and multiple candidate flow subsequences; to perform graph convolution feature extraction processing on a portion of the data in each target running subsequence and each candidate flow subsequence through the graph convolution layer of the motor temperature prediction model to obtain the target running convolution features corresponding to the target running time sequence data and the candidate flow convolution features corresponding to the candidate flow time sequence data; and to perform prediction processing on the target running convolution features and candidate flow convolution features through the prediction layer of the motor temperature prediction model to obtain candidate temperature prediction data.

[0189] In an exemplary embodiment, the traffic optimization module 1004 is configured to determine at least one running data collection point as a representative node for each target running subsequence from each target running subsequence, and at least one traffic collection point or traffic candidate data as a representative node for each candidate traffic subsequence, respectively, through the node sublayer in the graph convolutional layer; perform graph convolutional feature extraction processing on each target running representative node and each candidate traffic representative node through the first graph convolutional sublayer in the graph convolutional layer to obtain each intermediate running convolutional feature corresponding to each target running representative node and each intermediate traffic convolutional feature corresponding to each candidate traffic representative node; and perform graph convolutional feature extraction processing on each intermediate running convolutional feature and each intermediate traffic convolutional feature through the second graph convolutional sublayer in the graph convolutional layer to obtain target running convolutional features and candidate traffic convolutional features.

[0190] In an exemplary embodiment, the flow optimization module 1004 is used to determine a first number of cold liquid flow data from candidate flow subsequences belonging to the same time window as the target running representative node based on the distance priority principle, wherein the first number of cold liquid flow data includes cold liquid flow acquisition data or cold liquid flow candidate data; based on the distance data between the target running representative node and the first number of cold liquid flow data, a first adjacency matrix corresponding to the target running representative node is obtained; based on the first adjacency matrix corresponding to the target running representative node, the target running representative node and the first number of cold liquid flow data, convolution aggregation processing is performed to obtain intermediate running convolution features.

[0191] In an exemplary embodiment, the traffic optimization module 1004 is used to concatenate each intermediate running convolutional feature and each intermediate traffic convolutional feature according to a time window to obtain each concatenated feature corresponding to each time window; perform similarity calculation on each concatenated feature to obtain a second adjacency matrix; and perform convolution aggregation based on each intermediate running convolutional feature, each intermediate traffic convolutional feature and the second adjacency matrix to obtain the target running convolutional feature and the candidate traffic convolutional feature.

[0192] In an exemplary embodiment, the traffic optimization module 1004 is used to perform concatenation processing on the target running convolutional features and candidate traffic convolutional features through the concatenation sub-layer in the prediction layer to obtain concatenated candidate features; and to perform prediction processing on the concatenated candidate features through the prediction sub-layer in the prediction layer to obtain candidate temperature prediction data, wherein the candidate temperature prediction data includes the temperature corresponding to at least one location in the vehicle motor at at least one prediction time.

[0193] In an exemplary embodiment, the runtime acquisition data includes speed acquisition data and torque acquisition data, and the target runtime timing data includes target speed timing data and target torque timing data.

[0194] Based on the same inventive concept, this application also provides a vehicle motor temperature prediction device for implementing the vehicle motor temperature prediction method described above. The solution provided by this device is similar to the implementation described in the above method; therefore, the specific limitations of one or more vehicle motor temperature prediction device embodiments provided below can be found in the limitations of the vehicle motor temperature prediction method described above, and will not be repeated here.

[0195] In one exemplary embodiment, such as Figure 11 As shown, a vehicle motor temperature prediction device is provided, including: a data input module 1102, a data partitioning module 1104, a feature extraction module 1106, and a temperature prediction module 1108, wherein,

[0196] Data input module 1102 is used to acquire the running timing data and coolant flow timing data of the vehicle motor at the target time.

[0197] The data partitioning module 1104 is used to divide the runtime sequence data and the coolant flow time sequence data into multiple runtime subsequences and multiple coolant flow subsequences according to a preset time window.

[0198] The feature extraction module 1106 is used to perform graph convolution feature extraction processing on some data in each running subsequence and each cold liquid flow rate subsequence to obtain the target running convolution feature and the target flow rate convolution feature corresponding to the target time.

[0199] The temperature prediction module 1108 is used to perform prediction processing based on the target running convolutional features and the target flow convolutional features to obtain the target predicted temperature data of the vehicle motor at the target time.

[0200] The modules in the aforementioned vehicle motor temperature control device and vehicle motor temperature prediction device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0201] In one exemplary embodiment, a computer device is provided, which may be a terminal vehicle controller, and its internal structure diagram may be as follows. Figure 12As shown, the computer device includes a processor, memory, input / output interface, communication interface, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface and input device are connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a vehicle motor temperature control method or a vehicle motor temperature prediction method. The input device can be buttons, a trackball, or a touchpad located on the computer device casing, or it can be an external keyboard, touchpad, or mouse.

[0202] Those skilled in the art will understand that Figure 12 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0203] In one embodiment, an on-board controller is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0204] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0205] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0206] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0207] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0208] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0209] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for controlling the temperature of a vehicle motor, characterized in that, The method includes: Acquire target runtime timing data and multiple coolant flow rate acquisition data corresponding to the vehicle motor at a target time, wherein the target runtime timing data includes multiple runtime acquisition data; Based on the target runtime sequence data and the multiple coolant flow rate acquisition data, iterative optimization is performed by minimizing the output data of the motor temperature prediction model to obtain target flow control data. In each iteration, candidate flow rate time series data is obtained based on the multiple coolant flow rate acquisition data and the coolant flow rate candidate data in the current iteration. The flow rate data corresponding to at least one time window that is close to the target time in time series are all considered as candidate coolant flow rates. The target runtime sequence data and the candidate flow rate time series data are divided according to the time windows through the input layer of the motor temperature prediction model to obtain multiple target runtime subsequences. The model uses a graph convolutional layer of the motor temperature prediction model to extract graph convolutional features from a portion of the data in each target running subsequence and each candidate flow subsequence. This yields the target running convolutional features corresponding to the target running time series data and the candidate flow convolutional features corresponding to the candidate flow time series data. The model then uses a prediction layer of the motor temperature prediction model to predict the target running convolutional features and the candidate flow convolutional features, resulting in candidate temperature prediction data corresponding to the candidate coolant flow rate data. The target flow control data is the coolant flow rate candidate data that minimizes the candidate temperature prediction data output by the motor temperature prediction model during each iteration. The coolant flow rate used to cool the vehicle motor is controlled based on the target flow control data.

2. The method according to claim 1, characterized in that, The step of extracting graph convolutional features from partial data in each target running subsequence and each candidate flow subsequence using the graph convolutional layer of the motor temperature prediction model to obtain the target running convolutional features corresponding to the target running time series data and the candidate flow convolutional features corresponding to the candidate flow time series data includes: Through the node sub-layers in the graph convolutional layer, at least one running data collection point is determined from each of the target running sub-sequences as a representative node for each target running, and at least one traffic data collection point or traffic candidate data is determined from each of the candidate traffic sub-sequences as a representative node for each candidate traffic. The graph convolutional sub-layer in the graph convolutional layer performs graph convolutional feature extraction on each of the target running representative nodes and each of the candidate traffic representative nodes to obtain the intermediate running convolutional features corresponding to each of the target running representative nodes and the intermediate traffic convolutional features corresponding to each of the candidate traffic representative nodes. The intermediate running convolutional features and intermediate traffic convolutional features are extracted by the second graph convolutional sub-layer in the graph convolutional layer to obtain the target running convolutional features and the candidate traffic convolutional features.

3. The method according to claim 2, characterized in that, The process of extracting graph convolutional features from each of the target running representative nodes through the first graph convolutional sub-layer in the graph convolutional layer to obtain the intermediate running convolutional features corresponding to each of the target running representative nodes includes: Based on the distance priority principle, a first number of cold liquid flow data are determined from the candidate flow subsequence that belongs to the same time window as the target running representative node, wherein the first number of cold liquid flow data includes the cold liquid flow acquisition data or the cold liquid flow candidate data; Based on the distance data between the target operating representative node and the first number of cold liquid flow data, a first adjacency matrix corresponding to the target operating representative node is obtained; The intermediate operation convolutional features are obtained by performing convolutional aggregation processing based on the first adjacency matrix corresponding to the target operation representative node, the target operation representative node, and the first number of cold liquid flow data.

4. The method according to claim 2, characterized in that, The step of extracting graph convolutional features from each intermediate running convolutional feature and each intermediate traffic convolutional feature through a second graph convolutional sublayer in the graph convolutional layer to obtain the target running convolutional feature and the candidate traffic convolutional feature includes: The intermediate running convolutional features and the intermediate flow convolutional features are concatenated according to the time window to obtain the concatenated features corresponding to each time window. The similarity calculation is performed on each of the splicing features to obtain the second adjacency matrix; Convolution aggregation is performed based on the intermediate running convolution features, the intermediate traffic convolution features, and the second adjacency matrix to obtain the target running convolution features and the candidate traffic convolution features.

5. The method according to claim 2, characterized in that, The process of predicting the target operating convolutional features and the candidate flow convolutional features through the prediction layer of the motor temperature prediction model to obtain the candidate temperature prediction data includes: The target running convolutional features and the candidate traffic convolutional features are concatenated by the concatenation sub-layer in the prediction layer to obtain concatenated candidate features. The candidate features are predicted by the prediction sub-layer in the prediction layer to obtain the candidate temperature prediction data, wherein the candidate temperature prediction data includes the temperature of at least one position in the vehicle motor at at least one prediction time.

6. The method according to claim 1, characterized in that, The operational data collected includes speed data and torque data, and the target operational timing data includes target speed timing data and target torque timing data.

7. A vehicle motor temperature control device, characterized in that, The device includes: The data acquisition module is used to acquire target runtime sequence data of the vehicle motor at a target time and multiple coolant flow acquisition data, wherein the target runtime sequence data includes multiple runtime acquisition data; The flow optimization module is used to iteratively optimize the target flow control data based on the target runtime timing data and the multiple coolant flow acquisition data by minimizing the output data of the motor temperature prediction model. In each iteration, candidate flow timing data is obtained based on the multiple coolant flow acquisition data and the coolant flow candidate data in the current iteration. The flow data corresponding to at least one time window that is close to the target time in time sequence are all considered as candidate coolant flow data. The target runtime timing data and the candidate flow timing data are divided according to the time window by the input layer of the motor temperature prediction model to obtain multiple target flow control data. The target running subsequence and multiple candidate flow rate subsequences are processed by graph convolutional layer of the motor temperature prediction model to extract graph convolutional features from a portion of the data in each target running subsequence and each candidate flow rate subsequence. This yields target running convolutional features corresponding to the target running time sequence data and candidate flow rate convolutional features corresponding to the candidate flow rate time sequence data. The prediction layer of the motor temperature prediction model then performs prediction processing on these target running convolutional features and candidate flow rate convolutional features to obtain candidate temperature prediction data corresponding to the candidate coolant flow rate data. The target flow rate control data is the coolant flow rate candidate data that minimizes the candidate temperature prediction data output by the motor temperature prediction model during each iteration. A flow control module is used to control the flow rate of coolant used to cool the vehicle motor based on the target flow control data.

8. A method for predicting vehicle motor temperature in the vehicle motor temperature control device of claim 7, characterized in that, The method includes: Acquire the vehicle motor's runtime timing data and coolant flow timing data at the target time; The runtime sequence data and the coolant flow time sequence data are divided into multiple runtime subsequences and multiple coolant flow subsequences according to a preset time window by the input layer of the motor temperature prediction model. The graph convolutional layer of the motor temperature prediction model is used to extract features from some data in each of the running subsequences and each of the coolant flow subsequences through graph convolution, so as to obtain the target running convolutional features and target flow convolutional features corresponding to the target time. The prediction layer of the motor temperature prediction model performs prediction processing on the target operation convolutional features and the target flow convolutional features to obtain the target predicted temperature data of the vehicle motor at the target time.

Images