Temperature prediction method and device, equipment and storage medium

Abstract

The invention discloses a temperature prediction method and device, equipment and a storage medium. The temperature prediction method comprises the following steps: acquiring target working condition characteristics related to a target object in traffic equipment, wherein the target working condition characteristics are obtained by performing preset processing on a plurality of initial working condition data collected in a preset time period; inputting the target working condition characteristics into a gating module in a temperature prediction model to obtain candidate weights of at least one preset temperature interval output by the gating module; the target working condition features are input into at least one prediction module in a temperature prediction model, candidate prediction temperatures output by all the prediction modules are obtained, and all the prediction modules correspond to different preset temperature intervals; and determining a temperature prediction result of the target object based on each candidate weight and each candidate prediction temperature, the temperature prediction result comprising the temperature of the target object at the current moment. According to the scheme, the accuracy of temperature prediction of the target object can be improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular, to a temperature prediction method, apparatus, device, and storage medium. Background Art

[0002] Currently, the method for detecting the temperature of a target object (such as devices like motors and batteries) inside a traffic device is to directly obtain the temperature of the target object to be temperature-detected using a sensor. This places high requirements on the installation location, reliability, and accuracy of the sensor. If the installation location is inappropriate or the sensor has errors, the temperature detection result will be inaccurate. In some application scenarios, a general temperature prediction model can be used to detect the temperature of devices such as motors and batteries inside a traffic device. The current general temperature prediction models do not consider the situation when the target object reaches the limit temperature, and the prediction effect of these general temperature prediction models for the limit temperature is often poor.

[0003] In view of the existing technical deficiencies, how to provide an effective temperature prediction solution is a technical problem that those skilled in the art urgently need to solve. Summary of the Invention

[0004] This application provides at least a temperature prediction method, apparatus, device, and storage medium.

[0005] This application provides a temperature prediction method, including: obtaining target operating condition features related to a target object in a traffic device, where the target operating condition features are obtained by performing a preset process on a plurality of initial operating condition data collected within a preset time period; inputting the target operating condition features into a gating module in a temperature prediction model to obtain candidate weights for at least one preset temperature range output by the gating module; inputting the target operating condition features into at least one prediction module in the temperature prediction model respectively to obtain candidate predicted temperatures output by each prediction module, where each prediction module corresponds to a different preset temperature range; and determining a temperature prediction result of the target object based on the candidate weights and the candidate predicted temperatures, where the temperature prediction result includes the temperature of the target object at the current moment.

[0006] In some embodiments, inputting the target operating condition features into a gating module in a temperature prediction model to obtain candidate weights for at least one preset temperature range output by the gating module includes: performing a preset feature extraction process on the target operating condition features to obtain initial features corresponding to the target operating condition features; performing a first fusion process on the initial features and historical advanced features to obtain advanced features corresponding to the target operating condition features, where the historical advanced features are advanced features corresponding to historical target operating condition features; and performing a first mapping process on the advanced features corresponding to the target operating condition features to obtain each candidate weight.

[0007] In some embodiments, the step of performing preset feature extraction processing on the target operating condition features to obtain the initial features corresponding to the target operating condition features is executed by a preset feature extraction module. The preset feature extraction module includes a number of target modules arranged in cascade. The features output by the last target module are used as the initial features. The step of performing preset feature extraction processing on the target operating condition features to obtain the initial features corresponding to the target operating condition features includes: using the current features as the input of the current target module for convolution processing to obtain convolution features, where the current features are the target operating condition features or the features output by the previous target module; performing normalization processing on the convolution features to obtain the normalized convolution features; performing max pooling processing on the normalized convolution features to obtain pooling features, and using the pooling features as the features output by the current target module.

[0008] In some embodiments, inputting the target operating condition features into at least one prediction module in the temperature prediction model respectively to obtain the candidate predicted temperatures output by each prediction module includes: for each prediction module, performing the following steps: performing second fusion processing on the target operating condition features and the historical fusion features to obtain the fusion features corresponding to the target operating condition features, where the historical fusion features are the fusion features corresponding to the historical target operating condition features; performing second mapping processing on the fusion features corresponding to the target operating condition features to obtain the candidate predicted temperatures output by the prediction module.

[0009] In some embodiments, determining the temperature prediction result of the target object based on each candidate weight and each candidate predicted temperature includes: performing weighted fusion on each candidate weight and each candidate predicted temperature to obtain the target temperature; using the target temperature as the temperature prediction result.

[0010] In some embodiments, when there are multiple preset operating condition types to which a number of initial operating condition data belong, obtaining the target operating condition features related to the target object in the transportation equipment includes: filtering the number of initial operating condition data according to a preset rule to obtain a number of candidate operating condition data, where the preset rule is related to the power signal of the target object and / or the rotational speed and torque of the target object; performing standardization processing on the data group to be processed to obtain the processed data group corresponding to the data group to be processed, where the data group to be processed includes the data information in each candidate operating condition data belonging to the same preset operating condition type; determining the target operating condition features based on the processed data group.

[0011] In some embodiments, the temperature prediction model is pre-trained, and the training steps of the temperature prediction model include: obtaining sample target operating condition features related to a sample object and the true temperature of the sample object, where the sample target operating condition features are obtained by performing preset processing on a number of sample initial operating condition data collected from the sample object in a traffic device within a preset time period; inputting the sample target operating condition features into the temperature prediction model to obtain a temperature prediction result of the sample object output by the temperature prediction model, where the temperature prediction result of the sample object includes a predicted temperature of the sample object; determining a target loss based on the difference between the predicted temperature of the sample object and the true temperature of the sample object; and adjusting the parameters in the temperature prediction model using the target loss to obtain a trained temperature prediction model.

[0012] The present application provides a temperature prediction device, including: an acquisition module, a first determination module, a second determination module, and a third determination module; the acquisition module is configured to acquire target operating condition features related to a target object in a traffic device, where the target operating condition features are obtained by performing preset processing on a number of initial operating condition data collected within a preset time period; the first determination module is configured to input the target operating condition features into a gating module in the temperature prediction model to obtain candidate weights for at least one preset temperature range output by the gating module; the second determination module is configured to input the target operating condition features into at least one prediction module in the temperature prediction model respectively to obtain candidate predicted temperatures output by each prediction module, where each prediction module corresponds to a different preset temperature range; and the third determination module is configured to determine a temperature prediction result of the target object based on the candidate weights and the candidate predicted temperatures, where the temperature prediction result includes the temperature of the target object at the current moment.

[0013] The present application provides an electronic device, including a memory and a processor, where the processor is configured to execute program instructions stored in the memory to implement the above temperature prediction method.

[0014] The present application provides a computer-readable storage medium, on which program instructions are stored, and when the program instructions are executed by a processor, the above temperature prediction method is implemented.

[0015] In the above solution, by acquiring the target operating condition features related to the target object in the traffic device, the candidate weights for each preset temperature range obtained by inputting the target operating condition features into the gating module in the temperature prediction model can reflect the importance of each preset temperature range, and the candidate predicted temperatures output by each prediction module obtained by inputting the target operating condition features into the prediction modules corresponding to different preset temperature ranges can have higher accuracy in each preset temperature range. Then, using the different candidate weights and the different candidate predicted temperatures to determine the temperature prediction result of the target object can adapt to different preset temperature ranges, thereby improving the accuracy of the temperature prediction result.

[0016] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, rather than limiting the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings herein are incorporated into and constitute a part of this specification. These drawings show embodiments consistent with the present application and, together with the specification, are used to explain the technical solutions of the present application.

[0018] Figure 1 is a flowchart illustration of an embodiment of the temperature prediction method of the present application Figure 1 ;

[0019] Figure 2 is a flowchart illustration of an embodiment of the temperature prediction method of the present application Figure 2 ;

[0020] Figure 3 is a flowchart illustration of an embodiment of the temperature prediction method of the present application Figure 3 ;

[0021] Figure 4 is a flowchart illustration of an embodiment of the temperature prediction method of the present application Figure 4 ;

[0022] Figure 5 is a schematic structural diagram of an embodiment of the temperature prediction device of the present application;

[0023] Figure 6 is a schematic structural diagram of an embodiment of the electronic device of the present application;

[0024] Figure 7 is a schematic structural diagram of an embodiment of the computer-readable storage medium of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0025] The following describes the solutions of the embodiments of the present application in detail with reference to the accompanying drawings of the specification.

[0026] In the following description, specific details such as specific system architectures, interfaces, and technologies are set forth for purposes of illustration rather than limitation, in order to provide a thorough understanding of the present application.

[0027] As used herein, the term "and / or" is merely a description of the relationship between associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. Additionally, the character " / " in this text generally indicates that the associated objects before and after are in an "or" relationship. Furthermore, "plurality" in this text means two or more than two. Additionally, the term "at least one" in this text means any one of a plurality or any combination of at least two of a plurality. For example, including at least one of A, B, and C can mean including any one or more elements selected from the set composed of A, B, and C.

[0028] This application provides some temperature prediction methods and temperature prediction devices. The application scenarios of the temperature prediction method include, but are not limited to, temperature prediction of a target object. The execution subject of the temperature prediction method can be a temperature prediction device. For example, the temperature prediction device can be set in a terminal device, a server, or other processing devices. Among them, the terminal device can be a device for temperature prediction, a user equipment (UE), a mobile device, a user terminal, a terminal, a cellular phone, a cordless phone, a personal digital assistant (PDA), a handheld device, a computing device, a vehicle-mounted device, etc. In some possible implementation manners, the temperature prediction method can be implemented by a processor calling computer-readable instructions stored in a memory.

[0029] Please refer to Figure 1 , Figure 1 which is a schematic flowchart of an embodiment of the temperature prediction method of this application. Figure 1 .

[0030] Specifically, the temperature prediction method may include the following steps:

[0031] Step S11: Obtain target operating condition characteristics related to the target object in the transportation device.

[0032] The transportation device can be an airplane, a ship, or a vehicle. The target object can be an electric motor. This application takes the transportation device as a vehicle as an example. The target operating condition characteristics are obtained by performing preset processing on a number of initial operating condition data collected within a preset time period.

[0033] The preset time period can be a time period before the current moment or a time period including the current moment. The several initial operating condition data can be temperature reference information collected using vehicle sensors. Each of the initial operating condition data includes at least one temperature reference information. Each temperature reference information in each initial operating condition data can include one or more of the following: the current and / or voltage feedback of each axis of the motor, oil-cooled electric drive oil pump related information, motor speed, actual torque feedback, motor inlet temperature, motor controller temperature, motor water pump speed, etc. Specifically, the current and / or voltage feedback of each axis of the motor includes at least one of the following: the current of the D-axis of the motor, the current of the Q-axis of the motor, the voltage of the D-axis of the motor, the voltage of the Q-axis of the motor. The oil-cooled electric drive oil pump related information includes at least one of the following: motor water pump speed, oil-cooled electric drive oil pump current, oil-cooled electric drive oil pump bus voltage, oil-cooled electric drive oil pump speed.

[0034] In some application scenarios, the target object is the motor, and the initial operating condition data of the target object within the preset time period is collected using vehicle on-board sensors, including: obtaining several initial operating condition data on the vehicle on-board sensors, and the initial operating condition data is collected synchronously at a certain frequency (for example, synchronously collected every three seconds) to ensure that the moments corresponding to each initial operating condition data collected each time are the same. Exemplarily, each piece of initial operating condition data includes at least one of the following: vehicle identification ID, sampling time, D-axis current feedback of the motor, Q-axis current feedback of the motor, D-axis voltage of the motor, Q-axis voltage of the motor, motor speed, actual torque feedback, automotive power switch signal, motor inlet temperature, motor controller temperature, motor water pump speed, oil-cooled electric drive oil pump current, oil-cooled electric drive oil pump bus voltage, oil-cooled electric drive oil pump speed, and other information.

[0035] The preset processing may include, but is not limited to, at least one of screening a number of initial operating condition data and data combination. In some application scenarios, screening a number of initial operating condition data may be based on a screening reference factor to screen each initial operating condition data to obtain the screened initial operating condition data, and feature construction is performed on the screened initial operating condition data to obtain the target operating condition feature. The screening reference factor may be at least one of the automotive power switch signal, motor speed, and actual torque feedback in each initial operating condition data. Feature construction may be to splice the screened initial operating condition data to obtain an input sequence, and use this input sequence as the target operating condition feature. In other application scenarios, data combination may be to combine at least part of the temperature reference information in each initial operating condition data to obtain advanced temperature reference information. The advanced temperature reference information includes one or more of the following: the total current of the motor, the total voltage, the first total power, and the second total power. The first total power is obtained based on the total current and the total voltage. The second total power is the sum of the first initial power and the second initial power. The first initial power is determined based on the current of the D-axis of the motor and the voltage of the D-axis of the motor. The second initial power is determined based on the current of the Q-axis of the motor and the voltage of the Q-axis of the motor. In some application scenarios, data combination of a number of initial operating condition data may be at least one temperature reference information in each of the number of initial operating condition data and the advanced temperature reference information at each moment to obtain the target operating condition feature. In other application scenarios, the preset processing may be to screen and combine a number of initial operating condition data, where the screening and data combination may respectively refer to the above content and will not be elaborated here.

[0036] Step S12: Input the target operating condition feature into the gating module in the temperature prediction model to obtain the candidate weights of at least one preset temperature range output by the gating module.

[0037] The temperature prediction model includes at least one module. Among them, each module in the temperature prediction model may be a network module or an algorithm module. Specifically, the temperature prediction model includes a gating module and at least one prediction module corresponding to different preset temperature ranges. The gating module is used to determine the importance of at least one preset temperature range for the input target operating condition feature. Each prediction module is used to determine the predicted temperature of the input target operating condition feature in each preset temperature range.

[0038] The preset temperature intervals can be obtained by pre-dividing the temperature range (i.e., the full temperature range) in which the target object can be located according to the requirements of temperature prediction accuracy. The number of each preset temperature interval can be multiple. The temperature change values in the temperature ranges corresponding to each preset temperature interval can be the same or different. When the temperature change values in the temperature ranges corresponding to each preset temperature interval are the same, the number of each preset temperature interval is equal to the number of candidate weights that the gating module can output and the number of prediction modules. For the purpose of reducing the complexity of the temperature prediction model in this application, take the example where the temperature change values in the temperature ranges corresponding to each temperature interval are different. Exemplarily, the number of each preset temperature interval can be three, including the first preset temperature interval, the second preset temperature interval, and the third preset temperature interval respectively. The first preset temperature interval, the second preset temperature interval, and the third preset temperature interval can cover the full temperature range in which the target object can be located. Among them, the first preset temperature interval is the temperature less than the first preset value. The second preset temperature interval is greater than or equal to the first preset value but less than the second preset value. The third preset temperature interval is greater than or equal to the second preset value. Among them, the first preset value is less than the second preset value. For example, the first preset value can be 80 °C. The second preset value can be 120 °C.

[0039] The gating module in the above step S12 may include at least one network layer or algorithm module. For example, the gating module may at least include a preset feature extraction network and a fully connected layer. The preset feature extraction network may be a feature extraction network such as a convolutional neural network, a recurrent neural network, a long short-term memory network, a gated recurrent unit, a BiGRU neural network, or an attention mechanism.

[0040] Each candidate weight is used to represent the possibility that the temperature prediction result of the target object is in each preset temperature interval. Specifically, the gating module predicts the fuzzy interval corresponding to the target operating condition characteristics, that is, the probability values of each preset temperature interval, corresponding to the weights of the candidate temperatures output by each prediction module.

[0041] Step S13: Input the target operating condition characteristics into at least one prediction module in the temperature prediction model respectively to obtain the candidate predicted temperatures output by each prediction module.

[0042] Among them, each prediction module corresponds to a different preset temperature range. Among them, each prediction module can be a network module or an algorithm module with the same or different structures. In some application scenarios, each prediction module can be a different network module or algorithm module. The historical temperature prediction results of the target object are statistically analyzed, and in the preset temperature range where the historical temperature is relatively concentrated or dispersed, a preset network layer is added to the prediction module corresponding to the preset temperature range to improve the accuracy of the candidate predicted temperature output by the prediction module corresponding to the preset temperature range. The preset network layer can be a convolutional neural network or a gated recurrent unit, etc., which are feature extraction network structures that increase the module complexity. In some other application scenarios, each prediction module can be the same network module or algorithm module. Specifically, the structures of each prediction module are the same, but the network complexities are not the same. For the purpose of reducing the computational complexity of the temperature prediction model in this application, each prediction module being the same network module or algorithm module is taken as an example.

[0043] Each prediction module can at least include a preset feature extraction network and a fully connected layer. The preset feature extraction network can be a feature extraction network such as a convolutional neural network, a recurrent neural network, a long short-term memory network, a gated recurrent unit, a BiGRU neural network, or an attention mechanism. Exemplarily, the preset feature extraction network in each prediction module can be the long short-term memory network (LSTM, Long Short-Term Memory) in the recurrent neural network.

[0044] In some other application scenarios, the structures between the gating module and each prediction module are not the same, where the complexity of the gating module is higher than that of each prediction module.

[0045] Step S14: Determine the temperature prediction result of the target object based on each candidate weight and each candidate predicted temperature.

[0046] The temperature prediction result includes the temperature of the target object at the current moment.

[0047] In some other application scenarios, the temperature prediction result includes the temperature of the target object at the next moment. That is, this application can be used to predict the temperature of the target object at the current moment or can also be used to predict the temperature of the target object at a future moment.

[0048] The above step S14 may be to determine the temperature prediction result of the target object by combining each candidate weight and each candidate predicted temperature. In some application scenarios, the present application may select the largest candidate weight from each candidate weight, and use the preset temperature range corresponding to the largest candidate weight as the target temperature range. The candidate temperature output by the prediction module corresponding to the target temperature range is used as the target temperature. And the target temperature is used as the temperature prediction result of the target object. In other application scenarios, the above step S14 may be to directly perform weighted fusion on each candidate weight and each candidate predicted temperature to obtain the target temperature. And the target temperature is used as the temperature prediction result of the target object.

[0049] In the above solution, the target operating condition characteristics related to the target object in the transportation device are obtained. The candidate weights of each preset temperature range obtained by inputting the target operating condition characteristics into the gating module in the temperature prediction model can reflect the importance of each preset temperature range. The candidate predicted temperatures output by each prediction module obtained by inputting the target operating condition characteristics into different preset temperature range corresponding prediction modules can have higher accuracy in each preset temperature range. Then, using different candidate weights and different candidate predicted temperatures to determine the temperature prediction result of the target object can adapt to different preset temperature ranges, thereby improving the accuracy of the temperature prediction result.

[0050] In some embodiments, there are multiple types of preset operating conditions to which several initial operating condition data belong. The above step S11 may include the following steps: filtering the several initial operating condition data according to a preset rule to obtain several candidate operating condition data, and the preset rule is related to the power signal of the target object and / or the rotational speed and torque of the target object. Performing a normalization process on the data group to be processed to obtain a processed data group corresponding to the data group to be processed, and the data group to be processed includes the data information in each candidate operating condition data belonging to the same type of preset operating condition. Based on the processed data group, the target operating condition characteristics are determined.

[0051] The preset operating condition type can be the type of the above temperature reference information in each initial operating condition data. One preset operating condition type corresponds to one temperature reference information. A number of candidate operating condition data are the filtered initial operating condition data. Specifically, at least some of the initial operating condition data among the number of initial operating condition data are each candidate operating condition data. Among them, the temperature reference information in each initial operating condition data at least includes an automotive power switch signal related to the power signal of the target object and / or a motor speed and actual torque feedback related to the speed and torque of the target object. The preset rule can be that for each initial operating condition data, if the power signal of the target object and / or the speed and torque of the target object meet the preset conditions, then this initial operating condition data is retained. If the power signal of the target object and / or the speed and torque of the target object do not meet the preset conditions, then this initial operating condition data is discarded. The preset conditions can be a first preset condition and / or a second preset condition. Specifically, for the preset rule, meeting the first preset condition can be that this initial operating condition data is the driving segment corresponding to when the power signal of the target object is on. Specifically, for the preset rule, meeting the second preset condition can be that in the driving segment corresponding to this initial operating condition data, neither the speed nor the torque of the target object is 0 or the time length when the speed and torque of the target object are 0 is less than a certain time threshold. For example, the time threshold is 30 seconds. It can be understood that in the case where the preset conditions are the first preset condition and the second preset condition, the content of each preset condition can refer to the above content and will not be elaborated here. It can be understood that in the preset rule, the filtering corresponding to the first preset condition and the filtering corresponding to the second preset condition can be executed serially or in parallel. In the case where the filtering corresponding to the first preset condition and the filtering corresponding to the second preset condition in the preset rule are executed serially, the present application does not limit the sequence of their serial execution.

[0052] Exemplarily, a number of initial operating condition data may be the raw data collected by vehicle-mounted sensors. Considering that the raw data includes all driving states of the vehicle, when the vehicle is parked, the motor is not in a working state and is not within the scope of the research on the motor temperature prediction model. Therefore, it is necessary to extract the continuous driving segments of the vehicle from the collected raw data (a number of initial operating condition data). The initial operating condition data that meets the above first preset condition is retained to obtain the retained initial operating condition data. Specifically, the sub-segments in the raw data where the vehicle power supply signal is on are extracted. Then, it is determined whether the retained initial operating condition data meets the above second preset condition to obtain the filtered initial operating condition data. The second preset condition may be to extract the continuous driving segments where both the rotational speed and torque are greater than a certain value from the sub-segments, or the continuous driving segments where both the rotational speed and torque are 0 and the time length is less than a certain threshold. Specifically, the continuous driving segments where both the rotational speed and torque are not 0 are extracted from the above sub-segments. Note that the time length where both the rotational speed and torque are 0 in these continuous driving segments is allowed to be less than a certain time threshold, that is, the vehicle is allowed to pause during driving. In this way, a series of continuous driving segments can be obtained for the vehicle as the filtered initial operating condition data. The filtered initial operating condition data may be a number of candidate operating condition data. It can be understood that, in the case where the filtering corresponding to the first preset condition and the filtering corresponding to the second preset condition in the preset rules are executed serially in the present application, the first preset condition filtering is performed on a number of initial operating condition data according to the preset rules first, and then the second preset condition filtering is performed as an example.

[0053] The above-mentioned standardization process for the data group to be processed to obtain the processed data group corresponding to the data group to be processed may include: for each candidate operating condition data in each preset operating condition type, according to the average value and standard deviation of the candidate operating condition data of each preset operating condition type, determine the standardized result corresponding to the candidate operating condition data. Specifically, the standardized result corresponding to the candidate operating condition data may be expressed as

[0054] Exemplarily, according to the raw data obtained by vehicle-mounted sensors and the energy transfer mechanism of the motor, each candidate operating condition data selected in the present application includes at least one of the following as data information: motor D-axis current feedback, motor Q-axis current feedback, motor D-axis voltage feedback, motor Q-axis voltage feedback, motor speed, actual torque feedback, motor inlet temperature, motor controller temperature, motor water pump speed, oil-cooled electric drive oil pump current, oil-cooled electric drive oil pump bus voltage, oil-cooled electric drive oil pump speed. In the present application, it is taken as an example that each candidate operating condition data includes a total of 12 pieces of data information as above.

[0055] Exemplarily, the data information in each candidate operating condition data is classified according to various preset operating condition types, and data groups to be processed corresponding to various preset operating condition types are obtained. Among them, each data group to be processed includes the data information in each candidate operating condition data belonging to the same preset operating condition type. For example, when the preset operating condition type is the D-axis current feedback of the motor, the data group to be processed is the data group to which the D-axis current feedback of the motor belongs. The data group to be processed includes the D-axis current feedback in each candidate operating condition data. That is, the data information in the data group to be processed is the D-axis current feedback in each candidate operating condition data.

[0056] In some application scenarios, the standardization process can be Z-score standardization (standard normal distribution), Min-Max standardization, etc. Specifically, the process of standardizing each data information in each data group to be processed can refer to formula (1):

[0057]

[0058] where μ i is the average value of the preset operating condition type statistically obtained. σ i is the standard deviation of the preset operating condition type statistically obtained. x i is used to represent each data information in the data group to be processed of this type. As in the above example, each data information x i can be the D-axis current feedback in each candidate operating condition data. is used to represent the standardized result corresponding to the candidate operating condition data. Specifically, is used to represent the processed data information obtained after standardizing each data information in each data group to be processed. The processed data group corresponding to each data group to be processed includes the processed data information obtained after standardizing each data information in the data group to be processed of this preset operating condition type.

[0059] It can be understood that the number of data groups to be processed of each type is the same as the number of types of preset operating condition types. The data groups to be processed correspond to the number of types of preset operating condition types, and each data group to be processed only contains the data information corresponding to one preset operating condition type of each candidate operating condition data. In addition, the number of processed data groups of each type also corresponds to the number of types of preset operating condition types, and each processed data group only contains all the processed data information corresponding to one preset operating condition type.

[0060] In some application scenarios, the steps of determining the target operating condition characteristics based on the processed data groups may include: directly concatenating the processed data information in all the processed data groups to obtain the target operating condition characteristics. In other application scenarios, after obtaining each processed data group, according to a preset time window length (for example, the preset time window length is window size equal to 64), each continuous driving segment is divided into multiple segments of the same length to obtain each segmented length segment. The processed data information corresponding to each segmented length segment is concatenated to obtain the target operating condition characteristics.

[0061] It can be considered that in this application, filtering a number of initial operating condition data according to a preset rule can ensure the validity of the remaining number of candidate operating condition data. Then, performing a normalization process on the data group to be processed can ensure that the data distribution in the processed data group is more uniform, reduce the influence of extreme values on the temperature prediction result, and thus improve the accuracy of the temperature prediction model.

[0062] Please refer to Figure 2 , Figure 2 which is a flowchart illustration of an embodiment of the temperature prediction method of this application. Figure 2 .

[0063] In some embodiments, the above step S12 may include the following steps: Step S21: Perform a preset feature extraction process on the target operating condition characteristics to obtain the initial characteristics corresponding to the target operating condition characteristics. Step S22: Perform a first fusion process on the initial characteristics and the historical advanced characteristics to obtain the advanced characteristics corresponding to the target operating condition characteristics. The historical advanced characteristics are the advanced characteristics corresponding to the historical target operating condition characteristics. Step S23: Perform a first mapping process on the advanced characteristics corresponding to the target operating condition characteristics to obtain each candidate weight.

[0064] The above step S21 is executed by a preset feature extraction module. The preset feature extraction module may use the above-mentioned preset feature extraction network, which will not be elaborated here. The preset feature extraction module includes a number of target modules arranged in cascade, and the feature output by the last target module is used as the initial feature. Specifically, each target module may be provided with a convolution module, a normalization module, and a pooling module. Among them, each target module may be a network module with the same structure but different complexities. It can be understood that the gating module includes a number of target modules arranged in cascade. Exemplarily, the number of a number of target modules is 2.

[0065] In some embodiments, the above step S21 is executed by a preset feature extraction module, which includes a number of target modules arranged in cascade. The feature output by the last target module is used as the initial feature. The above step S21 may include the following steps: taking the current feature as the input of the current target module for convolution processing to obtain a convolution feature, where the current feature is the target working condition feature or the feature output by the previous target module. Performing normalization processing on the convolution feature to obtain a normalized convolution feature. Performing max pooling processing on the normalized convolution feature to obtain a pooled feature, and taking the pooled feature as the feature output by the current target module.

[0066] For the first target module, the first target module includes a convolution module, a normalization module, and a pooling module. Specifically, the target working condition feature is sequentially input into the convolution module, the normalization module, and the pooling module in the first target module, and the feature output by the pooling module is taken as the feature output by the first target module. For non-first target modules, the non-first target modules include a convolution module, a normalization module, and a pooling module. Specifically, the feature output by the previous target module is sequentially input into the convolution module, the normalization module, and the pooling module in the first target module, and the feature output by the pooling module is taken as the feature output by the non-first target module.

[0067] In some application scenarios, the target working condition feature is used as the input of the first target module. The target working condition feature is input into the convolution module in the first target module for convolution processing to obtain the convolution feature corresponding to the first target module. The convolution feature corresponding to the first target module is input into the normalization module in the first target module for normalization processing to obtain the normalized convolution feature corresponding to the first target module. The normalized convolution feature corresponding to the first target module is input into the pooling module in the first target module for max pooling processing to obtain the pooled feature corresponding to the first target module, and the pooled feature corresponding to the first target module is taken as the feature output by the first target module.

[0068] In some application scenarios, the target working condition feature is used as the input of a non-first target module. The target working condition feature is input into the convolution module in the non-first target module for convolution processing to obtain the convolution feature corresponding to the non-first target module. The convolution feature corresponding to the non-first target module is input into the normalization module in the non-first target module for normalization processing to obtain the normalized convolution feature corresponding to the non-first target module. The normalized convolution feature corresponding to the non-first target module is input into the pooling module in the non-first target module for max pooling processing to obtain the pooled feature corresponding to the non-first target module, and the pooled feature corresponding to the non-first target module is taken as the feature output by the non-first target module.

[0069] It can be understood that the features output by the pooling module in the last target module are used as the features output by the last target module, and then the features output by the last target module are used as the above-mentioned initial features.

[0070] The historical advanced features are the advanced features corresponding to the historical target operating condition features. The historical target operating condition features are obtained by performing preset processing on a number of historical initial operating condition data collected within a historical preset time period. Specifically, the historical advanced features are obtained by performing a first fusion process on the historical target operating condition features to obtain the advanced features corresponding to the historical target operating condition features. The historical preset time period is a preset time period before the above-mentioned preset time period. It can be understood that, for example, the historical preset time period can be the time period when the historical temperature prediction result output by the temperature prediction model was obtained in the last temperature prediction.

[0071] The above step S22 can be that the gating module includes a first fusion module. The first fusion process can be used to perform feature fusion on the initial features and the historical advanced features. The first fusion module can be a network module provided with a long short-term memory network (LSTM) or a gated recurrent unit (GRU). Specifically, the above step S22 can be to input the initial features and the historical advanced features into the first fusion module for the first fusion process to obtain the advanced features corresponding to the target operating condition features. In this application, the first fusion module is taken as an example of a long short-term memory network.

[0072] The above step S23 can be that the gating module includes a first mapping module. The first mapping module can be used to perform a mapping process on the advanced features corresponding to the target operating condition features. The first mapping module can be a network module provided with a fully connected layer or a multi-layer perceptron. Specifically, the above step S23 can be to input the advanced features corresponding to the target operating condition features into the first mapping module for the first mapping process to obtain the candidate weights for each preset temperature range. In this application, the first mapping module is taken as an example of a fully connected layer.

[0073] Exemplarily, considering that the temperature prediction model of the target object is complex and unstable, a deep learning approach is adopted here to construct the temperature prediction model. The temperature prediction model of this application includes at least one prediction module and a gating module. Each prediction module is responsible for predicting the temperature of its respective preset temperature range to obtain each candidate temperature. The gating module predicts different preset temperature ranges based on the input target operating condition characteristics and shows the appropriate weights for each prediction module. This application uses the weighted sum of the candidate temperatures output by each prediction module as the output of the temperature prediction model. It can be considered that the structure of the temperature prediction model of this application enables each prediction module to focus on predicting a small range of temperatures, so that a certain prediction module can be more focused on the temperature prediction of its respective preset temperature range. This application can achieve the prediction of extreme temperatures, enabling the entire temperature model to cover the prediction of the full temperature range of the target object.

[0074] Exemplarily, for the gating module, which is the core of the entire temperature prediction model and determines the performance of the entire temperature prediction model, it is necessary to determine the fuzzy interval of the temperature of the target object according to the input target operating condition characteristics. Therefore, the gating module adopts a time series classification model, and each category corresponds to its own preset temperature interval. Specifically, the gating module uses a CNN_LSTM_FC architecture, including a convolutional neural network, a long short-term memory neural network (Long Short-term Memory Networks, LSTM), and a fully connected neural layer (Fully Connected laye, FC). Specifically, the dimension of the input target operating condition characteristics is batch size × window size × feature dimension. The input needs to be transposed in two and three dimensions and then passed through two cascaded target modules. Each of the target modules is a convolutional layer. Each convolutional layer consists of a one-dimensional convolution (conv1d) with a convolutional kernel of 3×3, a batch normalization layer (BatchNormalization, BN), and a max pooling layer with a size of 2. At the same time, the hidden layer dimension of each convolutional layer is set to 125. This convolutional layer can expand the receptive field of the model, fuse the features of multiple moments, and alleviate the lag effect between features. Then, after transposing in two and three dimensions, it passes through an LSTM, that is, the above-mentioned first fusion module, whose hidden layer dimension is set to 100. This layer can predict the present through historical data and avoid the long-term dependence problem of time series prediction. Finally, the last bit of the window dimension output by the LSTM layer is input into the fully connected layer and the Softmax layer to fuse the hidden layer dimensions, and finally the probability values of each category are output. Among them, the above-mentioned first mapping module can be set as a fully connected layer and a Softmax layer. Each category corresponds to a preset temperature interval. In this application, three categories, that is, three preset temperature intervals, are taken as an example. The three probability values output by the temperature prediction model are subsequently used as the weight values of the three prediction modules.

[0075] Please refer to Figure 3 , Figure 3 which is a schematic flowchart of an embodiment of the temperature prediction method of the present application Figure 3 . In some embodiments, the above step S13 may include the following steps: For each prediction module, perform the following steps as Figure 3 shown: Step S31: Perform a second fusion process on the target operating condition characteristics and the historical fusion characteristics to obtain the fusion characteristics corresponding to the target operating condition characteristics. The historical fusion characteristics are the fusion characteristics corresponding to the historical target operating condition characteristics. Step S32: Perform a second mapping process on the fusion characteristics corresponding to the target operating condition characteristics to obtain the candidate predicted temperature output by the prediction module.

[0076] The historical fusion feature is the fusion feature corresponding to the historical target operating condition feature. The historical target operating condition feature is obtained by performing preset processing on a number of historical initial operating condition data collected within a historical preset time period. Specifically, the historical fusion feature is obtained by performing a second fusion process on the historical target operating condition feature to obtain the fusion feature corresponding to the historical target operating condition feature. The historical preset time period is a preset time period before the above-mentioned preset time period. It can be understood that, for example, the historical preset time period can be the time period when the historical temperature prediction result was output by the temperature prediction model in the previous temperature prediction.

[0077] The above step S31 may be that each prediction module includes a second fusion module. The second fusion process may be used to perform feature fusion on the target operating condition feature and the historical fusion feature. The second fusion module may be a network module provided with a long short-term memory network (LSTM, Long Short-Term Memory) or a gated recurrent unit (GRU, Gated Recurrent Units). Specifically, the above step 31 may be to input the target operating condition feature and the historical fusion feature into the second fusion module for the second fusion process to obtain the fusion feature corresponding to the target operating condition feature. In this application, the second fusion module is taken as an example of a long short-term memory network.

[0078] The above step S32 may be that each prediction module includes a second mapping module. The second mapping module may be used to perform mapping processing on the fusion feature corresponding to the target operating condition feature. The second mapping module may be a network module provided with a fully connected layer or a multi-layer perceptron. Specifically, the above step S32 may be to input the fusion feature corresponding to the target operating condition feature into the second mapping module for the second mapping process to obtain the candidate weights for each preset temperature range. In this application, the second mapping module is taken as an example of a fully connected layer.

[0079] Exemplarily, for each prediction module, it can accurately predict different preset temperature ranges, determining the upper limit of the entire temperature prediction model. For example, the three prediction modules in this application adopt the same model structure, and each temperature prediction module includes a second fusion module and a second mapping module. Among them, the second fusion module consists of one layer of LSTM, and the second mapping module consists of a fully connected layer FC. In some application scenarios, the second fusion module and the second mapping module in each prediction module can have the same structure as the LSTM in the first fusion module and the FC in the first mapping module in the gating module respectively. Since the prediction module accurately predicts a smaller temperature range, this application can adopt feature fusion without using a convolutional layer to avoid detail loss during the fusion process. For example, an input with a dimension of batch size × window size × feature dimension of the target operating condition features passes through a second fusion module with a hidden layer dimension of 100, and the last digit of the window dimension output by this layer passes through a fully connected layer, that is, the second mapping module, and the output of the second mapping module is used as the candidate temperature output by each prediction module.

[0080] In some embodiments, the above step S14 may include the following steps: performing weighted fusion on each candidate weight and each candidate predicted temperature to obtain a target temperature, and using the target temperature as the temperature prediction result.

[0081] The target temperature may be the weighted sum of each candidate weight and each candidate predicted temperature. The above step S14 may directly perform weighted fusion on each candidate weight and each candidate predicted temperature to obtain a target temperature, and use the target temperature as the temperature prediction result of the target object.

[0082] It can be considered that different prediction modules may perform better in different preset temperature ranges. Through weighted fusion, the advantages of multiple prediction modules can be comprehensively utilized, thereby improving the overall performance of the temperature prediction model.

[0083] Please refer to Figure 4 , Figure 4 which is a schematic flowchart of an embodiment of the temperature prediction method of this application. Figure 4 .

[0084] In some embodiments, the temperature prediction model is pre-trained, and the training steps of the temperature prediction model include: Step S41: Obtain sample target operating condition features related to the sample object and the true temperature of the sample object. The sample target operating condition features are obtained by performing preset processing on a number of sample initial operating condition data collected from the sample object in the traffic equipment within a preset time period. Step S42: Input the sample target operating condition features into the temperature prediction model to obtain the temperature prediction result of the sample object output by the temperature prediction model. The temperature prediction result of the sample object includes the predicted temperature of the sample object. Step S43: Determine the target loss based on the difference between the predicted temperature of the sample object and the true temperature of the sample object. Step S44: Use the target loss to adjust the parameters in the temperature prediction model to obtain the trained temperature prediction model.

[0085] Obtain a number of sample initial operating condition data on the in-vehicle sensors of the vehicle. The sample initial operating condition data is collected synchronously at a certain frequency (for example, synchronously collected every three seconds) to ensure that the moments corresponding to each sample initial operating condition data collected each time are the same. Among them, each sample initial operating condition data includes at least one sample temperature reference information. Exemplarily, each sample temperature reference information in each piece of sample initial operating condition data includes at least one of the following: vehicle identification ID, sampling time, motor D-axis current feedback, motor Q-axis current feedback, rear motor D-axis voltage, rear motor Q-axis voltage, motor speed, actual torque feedback, motor temperature, vehicle power switch signal, motor inlet water temperature, motor controller temperature, motor water pump speed, oil-cooled electric drive oil pump current, oil-cooled electric drive oil pump bus voltage, oil-cooled electric drive oil pump speed, and other information. Among them, the motor temperature is the true temperature of the sample object.

[0086] The step of obtaining the sample target operating condition features by performing preset processing on a number of sample initial operating condition data can refer to the above step of obtaining the target operating condition features by performing preset processing on a number of initial operating condition data, which will not be elaborated here. Among them, the sample object can be the same object as the above target object or a similar object of the above target object.

[0087] Exemplarily, after the structure of the temperature prediction model is constructed, the temperature prediction model is trained. First, a data set is prepared. The data set contains window segment data of different vehicles. The dimensions of these segments can be respectively represented as the number of segments × window size × feature dimension. Here, the window size is set to 64, the input feature dimension is 12, and the motor temperature of the last bit of the 64-bit window is predicted, that is, the output dimension is the number of segments × 1. The entire data set is randomly divided into a training set, a validation set, and a test set according to different vehicles, with a ratio of 7:2:1. The model is trained on the training set, and during the training process, it is judged whether the training ends according to the effect on the validation set. It can be understood that the data set includes sample target condition features related to the sample object and the true temperature of the sample object.

[0088] After the data set is prepared, the training method of the temperature prediction model is determined. First, the temperature prediction model is composed of multiple modules. If end-to-end training is directly performed, it is difficult to obtain the optimal solutions of all modules in the temperature prediction model at the same time. Therefore, each module needs to be pre-trained first, and then the entire temperature prediction model is fine-tuned. The target loss includes a first loss related to the gating module and a second loss related to each prediction module.

[0089] Among them, for the pre-training of the gating module, generally the training settings are that the optimizer is Adam, the scheduler is ReduceLROnPlateau, the initial learning rate is 0.01, and the batch size is 256. The loss function for training is selected as FocalLoss. Specifically, the process of determining the first loss can refer to formula (2):

[0090] FL = -(1 - p) γ log(p) Formula (2);

[0091] Among them, γ is the hyperparameter weight factor. p is the probability of a certain category predicted in the temperature prediction model, that is, the sample candidate weight of any preset temperature range obtained by inputting the above sample target condition features into the gating module. For example, γ can be set to 0.2. FL can represent the first loss function. It can be understood that this first loss function improves the problem of the multi-sample dominant temperature prediction model result caused by the sample imbalance problem through the weight factor γ. The gating module is trained on the entire full-temperature training set, and the early stopping strategy for training is that the classification accuracy of the validation set does not increase in 10 rounds. Obtain the true weights of the preset temperature ranges. Based on the difference between the sample candidate weights of each preset temperature range obtained by inputting the sample target condition features into the gating module and the true weights, determine the first loss. Based on the first loss, adjust the parameters of the gating module in the temperature prediction model to obtain the trained gating module.

[0092] For the pre-training of the prediction module, the general training settings are the same as the early stopping strategy. The MSE can be used as the training loss function. Specifically, refer to formula (3) for the process of determining the second loss:

[0093]

[0094] where N is the number of segments, and y i is the true temperature. is the predicted value of the predicted temperature of the sample object output by the temperature prediction model. Each predicted value needs to be trained on the data within the respective preset temperature range. Specifically, first, the entire dataset is divided into three subsets according to the full temperature range of the target object. The division method is as follows: the first subset is obtained when the maximum segment temperature is less than 80, the second subset is obtained when the minimum segment temperature is greater than or equal to 80 and less than 120, and the third subset is obtained when the minimum segment temperature is greater than or equal to 120. A prediction module is trained on each training set, and the effects of these prediction modules are better than those on the entire dataset. Training stops early when the MSE of the validation set does not decrease after 10 rounds. Based on the difference between the predicted temperature of the sample object and the true temperature of the sample object, the target loss and the second loss are determined. Based on the second loss, the parameters of each prediction module in the temperature prediction model are adjusted to obtain each trained prediction module. The above target loss can be the second loss.

[0095] In some application scenarios, step S43 above can be directly using the second loss as the target loss. In some other application scenarios, step S44 above can be training the gating module and each prediction module using the first loss and the second loss respectively, and using the trained gating module and the trained prediction modules as the gating module and each prediction module in the trained temperature prediction model. In some other application scenarios, step S44 above can be, after training the gating module and each prediction module using the first loss and the second loss respectively, jointly training the gating module and each prediction module in the temperature prediction model to obtain the trained temperature prediction model.

[0096] Exemplarily, after training the gating module and each prediction module using the first loss and the second loss respectively, after pre-training the gating module and each prediction module, they are combined into the entire temperature prediction model framework. Note that each prediction module needs to correspond to each temperature range result output by the gating module. Then, the entire temperature prediction model is trained on the dataset of the entire full temperature range. The initial weights of the gating module are obtained by loading the pre-trained temperature prediction model. Generally, the training settings, early stopping strategy, and loss function are the same as those in the pre-training stage of the prediction module. After pre-training and fine-tuning of the temperature prediction model, the effect of the temperature prediction model can be verified on the test set. The effect standard is set to have a temperature range every 20 degrees, and the final loss MSE between the predicted value of the predicted temperature and the true value of the true temperature of the sample object of the temperature prediction model is calculated for each temperature range. In response to the final loss meeting the requirements, it indicates that the fine-tuning of the temperature prediction model is completed.

[0097] Table 1: Comparative example table between the first reference loss and the second reference loss

[0098]

[0099] Among them, each preset temperature range includes the first temperature range to the eighth temperature range. The temperature ranges of the first temperature range to the eighth temperature range are 0-20, 20-40, 40-60, 60-80, 80-100, 100-120, 120-140, 140-160 respectively, where the unit of each temperature value is degree Celsius (℃).

[0100] Exemplarily, in order to verify the performance of the temperature prediction model, the present application may select vehicles that have experienced failures as sample objects, and specifically select the driving data of these vehicles in the two weeks before and after the failure as the above dataset. In order to verify the superiority of the temperature prediction model, the present application also trains a comparative temperature prediction model on the entire dataset, and the comparative temperature prediction model only includes one prediction module. The comparison of the effects between the temperature prediction model of the present application and the comparative temperature prediction model is shown in Table 1 above. The final loss of the comparative temperature prediction model is the above first reference loss. The final loss of the temperature prediction model of the present application in each preset temperature range is the above second reference loss. As can be seen from Table 1, the temperature prediction model of the present application has better effects in each preset temperature range than the comparative temperature prediction model, and the effects above 120°C and below 20°C are significantly improved. At the same time, in order to more significantly display the real-time prediction effect of the vehicle, after the real-time prediction of the same vehicle by the comparative temperature prediction model and the temperature prediction model of the present application, when the temperature value in the preset temperature range of the present application is higher, the effect of the temperature prediction model of the present application is better, and it can be seen that the effect of the temperature prediction model of the present application is more stable. The comparative temperature prediction model jitters more violently and is farther from the actual temperature curve.

[0101] In the above solution, by obtaining the target working condition characteristics related to the target object in the traffic equipment, the candidate weights of each preset temperature range obtained by inputting the target working condition characteristics into the gating module in the temperature prediction model can reflect the importance of each preset temperature range. Inputting the target working condition characteristics into the prediction modules corresponding to different preset temperature ranges can respectively obtain candidate predicted temperatures output by each prediction module with relatively high accuracy in each preset temperature range. Then, using different candidate weights and different candidate predicted temperatures to determine the temperature prediction result of the target object can adapt to different preset temperature ranges, thereby improving the accuracy of the temperature prediction result.

[0102] Please refer to Figure 5 , Figure 5It is a schematic structural diagram of an embodiment of the temperature prediction device of the present application. The temperature prediction device 50 includes an acquisition module 51, a first determination module 52, a second determination module 53, and a third determination module 54. The acquisition module 51 is configured to acquire target operating condition features related to a target object in a traffic device, where the target operating condition features are obtained by performing preset processing on a plurality of initial operating condition data collected within a preset time period; the first determination module 52 is configured to input the target operating condition features into a gating module in the temperature prediction model to obtain candidate weights of at least one preset temperature range output by the gating module; the second determination module 53 is configured to input the target operating condition features into at least one prediction module in the temperature prediction model respectively to obtain candidate predicted temperatures output by each prediction module, where each prediction module corresponds to a different preset temperature range; the third determination module 54 is configured to determine a temperature prediction result of the target object based on the candidate weights and the candidate predicted temperatures, and the temperature prediction result includes the temperature of the target object at the current moment.

[0103] In the above solution, the candidate weights of each preset temperature range obtained by inputting the target operating condition features related to the target object in the traffic device into the gating module in the temperature prediction model can reflect the importance degree of each preset temperature range, and the candidate predicted temperatures output by each prediction module obtained by inputting the target operating condition features into the prediction modules corresponding to different preset temperature ranges can have relatively high accuracy in each preset temperature range. Then, using different candidate weights and different candidate predicted temperatures to determine the temperature prediction result of the target object can adapt to different preset temperature ranges, thereby improving the accuracy of the temperature prediction result.

[0104] For the functions executed by each module, please refer to the temperature prediction method, which will not be elaborated here.

[0105] Please refer to Figure 6 , Figure 6 It is a schematic structural diagram of an embodiment of an electronic device of the present application. The electronic device 60 includes a memory 61 and a processor 62. The processor 62 is configured to execute program instructions stored in the memory 61 to implement the steps in the embodiment of the above temperature prediction method. In a specific implementation scenario, the electronic device 60 may include, but is not limited to, a microcomputer, a server. In addition, the electronic device 60 may also include mobile devices such as a laptop computer and a tablet computer, which will not be limited here.

[0106] Specifically, the processor 62 is used to control itself and the memory 61 to implement the steps in the above embodiments of the temperature prediction method. The processor 62 can also be referred to as a CPU (Central Processing Unit). The processor 62 may be an integrated circuit chip with signal processing capabilities. The processor 62 can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. Additionally, the processor 62 can be implemented jointly by integrated circuit chips.

[0107] In the above solution, the target operating condition features related to the target object in the transportation device are obtained. The candidate weights of each preset temperature range obtained by inputting the target operating condition features into the gating module in the temperature prediction model can reflect the importance of each preset temperature range. The candidate predicted temperatures output by each prediction module obtained by inputting the target operating condition features into the corresponding prediction modules of different preset temperature ranges can have relatively high accuracy in each preset temperature range. Then, using different candidate weights and different candidate predicted temperatures to determine the temperature prediction result of the target object can adapt to different preset temperature ranges, thereby improving the accuracy of the temperature prediction result.

[0108] Please refer to Figure 7 , Figure 7 is a schematic structural diagram of an embodiment of the computer-readable storage medium of the present application. The computer-readable storage medium 70 has program instructions 701 stored thereon. When the program instructions 701 are executed by the processor, the steps in any of the above embodiments of the temperature prediction method are implemented.

[0109] In the above solution, the target operating condition features related to the target object in the transportation device are obtained. The candidate weights of each preset temperature range obtained by inputting the target operating condition features into the gating module in the temperature prediction model can reflect the importance of each preset temperature range. The candidate predicted temperatures output by each prediction module obtained by inputting the target operating condition features into the corresponding prediction modules of different preset temperature ranges can have relatively high accuracy in each preset temperature range. Then, using different candidate weights and different candidate predicted temperatures to determine the temperature prediction result of the target object can adapt to different preset temperature ranges, thereby improving the accuracy of the temperature prediction result.

[0110] In some embodiments, the functions or modules included in the system provided by the embodiments of the present disclosure can be used to execute the methods described in the above method embodiments. The specific implementation can refer to the description of the above method embodiments. For the sake of brevity, it will not be repeated here.

[0111] The descriptions of the above embodiments tend to emphasize the differences between the embodiments. The similarities or similarities between them can be referred to each other. For the sake of brevity, they will not be repeated here.

[0112] In several embodiments provided in the present application, it should be understood that the disclosed methods and apparatuses can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of modules or units is only a logical function division. In actual implementation, there may be other division methods. For example, units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the couplings or direct couplings or communication connections shown or discussed with each other can be through some interfaces. The indirect couplings or communication connections of devices or units can be in electrical, mechanical or other forms.

[0113] In addition, in each embodiment of the present application, each functional unit can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0114] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods in various embodiments of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs that can store program codes.

Claims

1. A temperature prediction method, characterized in that: The method comprises: Acquire a target operating condition feature related to a target object in a traffic device, wherein the target operating condition feature is obtained by performing preset processing on a number of initial operating condition data collected within a preset time period; Inputting the target operating condition characteristics into a gating module in a temperature prediction model to obtain a candidate weight of at least one preset temperature interval output by the gating module; Inputting the target operating condition characteristics into at least one prediction module in the temperature prediction model respectively to obtain candidate prediction temperatures output by each prediction module, wherein each prediction module corresponds to a different preset temperature range; Based on each of the candidate weights and each of the candidate predicted temperatures, a temperature prediction result of the target object is determined, where the temperature prediction result includes the temperature of the target object at a current moment.

2. The method according to claim 1, characterized in that The step of inputting the target operating condition characteristic into a gating module in a temperature prediction model to obtain a candidate weight of at least one preset temperature interval output by the gating module comprises: Performing preset feature extraction processing on the target operating condition feature to obtain an initial feature corresponding to the target operating condition feature; Performing a first fusion process on the initial feature and the historical advanced feature to obtain an advanced feature corresponding to the target operating condition feature, wherein the historical advanced feature is an advanced feature corresponding to the historical target operating condition feature; A first mapping process is performed on the advanced features corresponding to the target operating condition features to obtain the candidate weights.

3. The method according to claim 2, characterized in that The step of performing preset feature extraction processing on the target operating condition feature to obtain the initial feature corresponding to the target operating condition feature is performed by a preset feature extraction module, the preset feature extraction module includes a plurality of target modules arranged in cascade, and the feature output by the last target module is used as the initial feature. The step of performing preset feature extraction processing on the target operating condition feature to obtain the initial feature corresponding to the target operating condition feature includes: Performing convolution processing on the current feature as the input of the current target module to obtain a convolution feature, wherein the current feature is the target operating condition feature or the feature output by the previous target module; Normalizing the convolutional features to obtain normalized convolutional features; The normalized convolutional features are subjected to maximum pooling processing to obtain pooled features, and the pooled features are used as features output by the current target module.

4. The method according to any one of claims 1 to 3, characterized in that The step of inputting the target operating condition characteristics into at least one prediction module in the temperature prediction model to obtain the candidate prediction temperature output by each prediction module includes: For each prediction module, perform the following steps: Performing a second fusion process on the target operating condition feature and the historical fusion feature to obtain a fusion feature corresponding to the target operating condition feature, wherein the historical fusion feature is a fusion feature corresponding to the historical target operating condition feature; A second mapping process is performed on the fusion feature corresponding to the target operating condition feature to obtain a candidate predicted temperature output by the prediction module.

5. The method according to any one of claims 1 to 3, characterized in that: The step of determining the temperature prediction result of the target object based on each of the candidate weights and each of the candidate predicted temperatures includes: Performing weighted fusion on each of the candidate weights and each of the candidate predicted temperatures to obtain a target temperature; The target temperature is used as the temperature prediction result.

6. The method according to any one of claims 1 to 3, characterized in that: The preset operating condition types to which the plurality of initial operating condition data belong are multiple, and the target operating condition characteristics related to the target object in the traffic equipment are obtained, including: Filtering the plurality of initial operating condition data according to a preset rule to obtain a plurality of candidate operating condition data, wherein the preset rule is related to the power signal of the target object and / or the speed and torque of the target object; Performing standardization processing on the data group to be processed to obtain a processed data group corresponding to the data group to be processed, wherein the data group to be processed includes data information in each candidate operating condition data of the same type of preset operating condition; Based on the processed data set, the target operating condition characteristics are determined.

7. The method according to any one of claims 1 to 3, characterized in that The temperature prediction model is obtained through pre-training, and the training steps of the temperature prediction model include: Acquire a sample target operating condition characteristic associated with a sample object and a real temperature of the sample object, wherein the sample target operating condition characteristic is obtained by performing preset processing on a plurality of sample initial operating condition data collected from the sample object in the traffic equipment within a preset time period; Inputting the sample target operating condition characteristics into the temperature prediction model to obtain a temperature prediction result of the sample object output by the temperature prediction model, wherein the temperature prediction result of the sample object includes a predicted temperature of the sample object; determining a target loss based on a difference between the predicted temperature of the sample object and the actual temperature of the sample object; The target loss is used to adjust the parameters in the temperature prediction model to obtain a trained temperature prediction model.

8. A temperature prediction device, characterized in that: include: An acquisition module, used to acquire target operating condition characteristics related to a target object in a traffic device, wherein the target operating condition characteristics are obtained by performing preset processing on a number of initial operating condition data collected within a preset time period; A first determination module, configured to input the target operating condition characteristics into a gating module in a temperature prediction model, and obtain a candidate weight of at least one preset temperature interval output by the gating module; A second determination module is used to input the target operating condition characteristics into at least one prediction module in the temperature prediction model to obtain candidate prediction temperatures output by each prediction module, wherein each prediction module corresponds to a different preset temperature range; The third determination module is used to determine the temperature prediction result of the target object based on each of the candidate weights and each of the candidate predicted temperatures, wherein the temperature prediction result includes the temperature of the target object at a current moment.

9. An electronic device, characterized in that: include: A memory and a processor, wherein the memory stores program instructions, and the processor retrieves the program instructions from the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having program instructions stored thereon, characterized in that: When the program instructions are executed by a processor, they are used to implement the method according to any one of claims 1 to 7.

Images