Demisting method, model training deployment method, demisting device and vehicle

Abstract

The invention discloses a demisting method, a model training deployment method, a demisting device and a vehicle. The automatic demisting method comprises the steps of obtaining model parameters of a side window temperature prediction model; input parameters of the side window temperature prediction model are obtained, wherein the input parameters comprise the outside humidity, the outside temperature, the inside humidity, the inside temperature and air conditioner real-time parameters; obtaining a real-time predicted temperature based on the side window temperature prediction model and the input parameters; obtaining a dew point temperature; and the real-time predicted temperature is compared with the dew point temperature, and when the real-time predicted temperature is not higher than the dew point temperature, air conditioner real-time parameters are adjusted so that the real-time predicted temperature can be higher than the dew point temperature. According to the demisting method, the real-time parameters of the air conditioner can be adjusted in real time according to the driving environment and the working condition, so that the vehicle can provide the best comfort, prevent the side window from fogging and automatically demist after fogging in different weathers, and the riding experience is improved.

Description

Technical Field

[0001] The present invention relates to vehicle intelligent control, and in particular to a defogging method, a model training and deployment method, a defogging device and a vehicle. Background Art

[0002] With the popularization of automobile products, automobile functions are becoming more and more abundant. How to provide more advanced intelligent experience for old cars, second-hand cars and other affordable cars has become one of the directions that everyone is considering. Among them, automatic defogger is the most important and direct function in the driving process. At present, the air conditioning sensor, external temperature and humidity sensor of new energy vehicles are the most basic sensors. How to use these basic sensors and air conditioning control system to predict the temperature and fogging of the side windows in real time has become the key point to bring direct experience to users. Summary of the invention

[0003] In view of this, the purpose of the present invention is to provide a defog method, a model training and deployment method, a defog device and a vehicle. The defog method can adjust the real-time parameters of the air conditioner in real time according to the driving environment and working conditions, so that the vehicle can provide optimal comfort in different weather conditions, prevent the side windows from fogging, and automatically defog after fogging, thereby improving the riding experience; the model training and deployment method can continuously optimize the parameters through the vehicle-side training mechanism, reduce problems such as manual calibration, reduce labor costs, and provide higher reliability and prediction accuracy.

[0004] A defogging method in the present invention, the automatic defogging method comprises: Obtaining model parameters of a side window temperature prediction model; Acquire input parameters of the side window temperature prediction model, wherein the input parameters include outside humidity, outside temperature, inside humidity, inside temperature, and air conditioning real-time parameters; Based on the side window temperature prediction model and the input parameters, obtaining a real-time predicted temperature; Get the dew point temperature; The real-time predicted temperature is compared with the dew point temperature, and when the real-time predicted temperature is not higher than the dew point temperature, the real-time parameters of the air conditioner are adjusted to make the real-time predicted temperature higher than the dew point temperature.

[0005] Further, obtaining the dew point temperature includes: Calculate the dew point temperature outside the vehicle based on the humidity outside the vehicle and the temperature outside the vehicle; The dew point temperature inside the vehicle is calculated based on the humidity and temperature inside the vehicle.

[0006] Further, the comparing the real-time predicted temperature with the dew point temperature, and when the real-time predicted temperature is not higher than the dew point temperature, adjusting the real-time parameters of the air conditioner to make the real-time predicted temperature higher than the dew point temperature, comprises: The real-time predicted temperature includes the real-time predicted temperature of the outer surface of the side window and the real-time predicted temperature of the inner surface of the side window; Comparing the real-time predicted temperature of the outer surface of the side window with the dew point temperature outside the vehicle, and comparing the real-time predicted temperature of the inner surface of the side window with the dew point temperature inside the vehicle; When the real-time predicted temperature of the outer surface of the side window is not higher than the dew point temperature outside the vehicle or / and the real-time predicted temperature of the inner surface of the side window is not higher than the dew point temperature inside the vehicle, adjusting the real-time parameters of the air conditioner so that the real-time predicted temperature of the outer surface of the side window is higher than the dew point temperature outside the vehicle and the real-time predicted temperature of the inner surface of the side window is higher than the dew point temperature inside the vehicle; Among them, the real-time air conditioning parameters include air conditioning start and stop status, air conditioning outlet temperature, air conditioning air volume, air conditioning wind speed, air conditioning gear and air conditioning outlet direction.

[0007] Further, the formulas for calculating the dew point temperature outside the vehicle and the dew point temperature inside the vehicle are both: Td=(A*t+B*U) / (C*U+D); Wherein, when Td is the dew point temperature outside the vehicle, t is the temperature outside the vehicle, and U is the humidity outside the vehicle; when Td is the dew point temperature inside the vehicle, t is the temperature inside the vehicle, and U is the humidity inside the vehicle; The unit of Td and t is ℃, the unit of U is %, A is a constant between 0.1780 and 0.2180, B is a constant between 0.0015 and 0.0019, C is a constant between 0.7400 and 0.9400, and D is a constant between 650 and 658.

[0008] A model training and deployment method in the present invention is used to train and deploy model parameters of the above-mentioned side window temperature prediction model; the model training and deployment method comprises: Based on the cyclic training of the model training vehicle, the model parameters of the final side window temperature prediction model are obtained; The model training vehicle exports the model parameter data of the trained side window temperature prediction model and uploads the data to the cloud, so that the model application vehicle can obtain the data from the cloud, parse and verify it, and obtain the model parameters of the side window temperature prediction model.

[0009] Furthermore, the model training vehicle exports the model parameters of the trained side window temperature prediction model and uploads the data to the cloud, so that the model application vehicle can obtain the data from the cloud and perform analysis and verification to obtain the model parameters of the side window temperature prediction model, specifically including: The model training vehicle exports the model parameters of the trained side window temperature prediction model in the form of APDU commands and uploads them to the cloud through the machine learning deployment module, so that the model application vehicle can obtain the APDU commands from the cloud, parse and verify them, and obtain the model parameters of the side window temperature prediction model.

[0010] Furthermore, the model parameters of the final side window temperature prediction model are obtained by the cyclic training based on the model training vehicle end, including: The model training vehicle obtains and processes the actual temperature of the side window through the data acquisition and calculation module, and the actual temperature of the side window includes the actual temperature of the outer surface of the side window and the actual temperature of the inner surface of the side window; The model training vehicle obtains and processes input parameters through the data acquisition and calculation module, and the input parameters include the humidity outside the vehicle, the temperature outside the vehicle, the humidity inside the vehicle, the temperature inside the vehicle, and the real-time parameters of the air conditioner; The data acquisition and calculation module sends the acquired and processed data to the buffer, which stores the data and sends the stored data in batches to the machine learning training module; The machine learning training module obtains the model parameters of the trained side window temperature prediction model through cyclic training and iterative training; Write the model parameters of the trained side window temperature prediction model into the machine learning deployment module.

[0011] Furthermore, the machine learning training module obtains the model parameters of the final side window temperature prediction model through cyclic training and iterative training, including: The machine learning training module uses the PCA principal component analysis method to screen weights, uses the normal equation to derive the parameters of the linear regression equation, and obtains the prediction results of the model, wherein the prediction results are the real-time predicted temperature of the outer surface of the side window and the real-time predicted temperature of the inner surface of the side window; The machine learning training module performs cyclic training and iterative training based on the reward function, designs a score of the reward function based on the degree of proximity between the predicted result and the actual temperature of the side window, and averages the training scores of different batches respectively; When the number of samples exceeds a preset number and the average score exceeds a preset value, the model parameters of the trained side window temperature prediction model are obtained; otherwise, the cyclic training and iterative training are continued.

[0012] A demisting device in the present invention comprises: A machine learning deployment module for obtaining model parameters of a side window temperature prediction model; A data acquisition and calculation module is used to obtain input parameters of the side window temperature prediction model, wherein the input parameters include outside humidity, outside temperature, inside humidity, inside temperature and real-time air conditioning parameters; An output module, used for obtaining a real-time predicted temperature based on the side window temperature prediction model and the input parameters; And a control module, which is used to obtain the dew point temperature and compare the real-time predicted temperature with the dew point temperature. When the real-time predicted temperature is not higher than the dew point temperature, the real-time parameters of the air conditioner are adjusted to make the real-time predicted temperature higher than the dew point temperature.

[0013] A vehicle in the present invention uses the above-mentioned automatic defog method, uses the above-mentioned model training deployment method, or includes the above-mentioned defog device.

[0014] The beneficial effects of the present invention are: (1) The automatic defog method of the present invention can adjust the real-time parameters of the air conditioner in real time according to the driving environment and working conditions, so that the vehicle can provide the best comfort in different weather conditions, prevent the side windows from fogging, and automatically defog after fogging, thereby improving the riding experience. In addition, by predicting the side window temperature through a machine learning algorithm, it is possible to reduce the need to arrange temperature sensors on the side windows, thereby reducing the cost of the entire vehicle.

[0015] (2) The model training and deployment method of the present invention adopts the multi-sensor fusion technology of the machine learning algorithm during training, so that the system has stronger adaptability and can handle complex driving scenarios, including different road conditions, weather and driving behaviors, thereby improving the robustness of the system.

[0016] (3) The model training and deployment method of the present invention can continuously optimize parameters through the vehicle-side training mechanism, reduce problems such as manual calibration, reduce labor costs, and provide higher reliability and prediction accuracy. And it can realize the rapid use of one vehicle training and multi-vehicle deployment applications through TBOX, OTA and other programs. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to make the purpose, technical solution and beneficial effects of the present invention clearer, the present invention provides the following drawings for illustration: Figure 1 A schematic diagram of the flow of the model training deployment method of the present invention; Figure 2 Schematic diagram of the process of the automatic demisting method of the present invention; Figure 3 This is an architecture diagram of the model training vehicle side, cloud side, and model application vehicle side of the present invention. DETAILED DESCRIPTION

[0018] The technical solution of the present invention is described in detail below in conjunction with the accompanying drawings and embodiments.

[0019] like Figure 1As shown, in this embodiment, the model parameters of the trained side window temperature prediction model are first obtained through the model training vehicle end, and then the model training vehicle end exports the model parameters of the trained side window temperature prediction model and uploads the data to the cloud. The model application vehicle end obtains the data from the cloud and parses and verifies it to obtain the model parameters of the side window temperature prediction model, so as to implement the automatic defog method. Among them, the model training vehicle end refers to the vehicle end used for training and uploading model parameters, which can be the manufacturer's test vehicle. The model application vehicle end can be a new car or a used car that can obtain model parameters from the cloud OTA program.

[0020] Specifically, the model parameters of the side window temperature prediction model are trained and deployed through the model training and deployment method, and the model training and deployment method includes the following steps: S1. Based on the cyclic training of the model training vehicle, the model parameters of the final side window temperature prediction model are obtained. This step specifically includes: S101, the model training vehicle side obtains and processes the actual temperature of the side window through the data acquisition and calculation module, and the actual temperature of the side window includes the actual temperature of the outer surface of the side window and the actual temperature of the inner surface of the side window.

[0021] During the training phase, the model training vehicle arranges a temperature sensor on the inner and outer surfaces of the side windows to measure the actual temperature of the outer and inner surfaces of the side windows, so as to facilitate the training and verification of the model parameters.

[0022] S102, the model training vehicle side obtains and processes input parameters through the data acquisition and calculation module, and the input parameters include the humidity outside the vehicle, the temperature outside the vehicle, the humidity inside the vehicle, the temperature inside the vehicle and the real-time parameters of the air conditioner.

[0023] The data acquisition and calculation module collects the outside humidity, outside temperature, inside humidity, inside temperature and real-time parameters of air conditioning through sensors and CAN signals. In order to adapt the linear regression algorithm and improve the algorithm accuracy, the data acquisition and calculation module can write a matlab function and convert it into a specified language using the matlab toolbox Embedded Coder. The function is a method for processing input parameters and performing feature processing, including but not limited to calculating variance, calculating average value, and calculating time information; these processed data and the input parameters obtained by the data acquisition and calculation module can be used as the input X of the machine learning training module.

[0024] S103: The data acquisition and calculation module sends the acquired and processed data to the cache. The cache stores the data and sends the stored data in batches to the machine learning training module.

[0025] The machine learning training module builds a machine learning algorithm model. The data acquisition and calculation module obtains and processes input parameters as input X. The actual temperature of the outer surface of the side window and the actual temperature of the inner surface of the side window measured by the temperature sensor are used as the Y of the model.

[0026] There is a buffer between the machine learning training module and the input, which is used to store data for a period of time and transmit it to the machine learning training module in batches in real time. In order to solve the problem of too much data for the same working condition and the priority of vehicle-side storage space, a switch button is provided. When the button is on, the buffer samples and stores data at a specified frequency with a sampling length of N lines each time; when the button is off, the buffer caches data for a long time without interruption.

[0027] S104, the machine learning training module obtains the model parameters of the trained side window temperature prediction model through cyclic training and iterative training.

[0028] The machine learning training module uses the PCA principal component analysis method to screen weights, and uses the normal equation to derive the parameters of the linear regression equation to obtain the prediction results of the model, which are the real-time predicted temperature of the outer surface of the side window and the real-time predicted temperature of the inner surface of the side window. First, because the sensor temperature is relatively sensitive, the temperature in the input X and the Y of the model is rounded to the single digit. Then, the time series features are constructed, including the calculation of the duration through the instructions such as starting the air conditioner and turning off the air conditioner, and the effective characteristic temperature is constructed using the effective temperature = the temperature in the car - a*air conditioner outlet temperature*air conditioner air volume-b*air conditioner outlet temperature*humidity in the car. Both a and b are constants, and the value range of a can be 0.2-0.5, and the value range of b can be 0.2-0.5. The variance, standard deviation, covariance and other features are calculated, and finally the PCA principal component analysis model is used to screen the feature data with a correlation coefficient greater than 0.2. Finally, the linear regression model is constructed as: y=Xθ+ϵ; where X is the m×n feature (m is the number of samples, n is the number of features), θ is the n×1 parameter vector, y is the m×1 temperature, and ϵ is the error term. Based on the linear regression equation, the loss function (residual sum of squares) is defined as: J(θ)= θ ⊤ X ⊤ Xθ−2y ⊤ Xθ+y ⊤ y; Derivative of the parameter θ of the J(θ) equation, where the quadratic term θ ⊤ X ⊤ The derivative of Xθ is 2X ⊤ Xθ, linear term −2y ⊤ The derivative of Xθ is −2X ⊤ y ; constant term y ⊤ The y derivative is 0; the final gradient is: ∇ θ J=2X⊤ Xθ−2X ⊤ y, use the normal equation to derive and solve, let ∇ θ J = 0, so that 2X ⊤ Xθ−2X ⊤ y=0, finally θ=(X ⊤ X) −1 X ⊤ y.

[0029] The machine learning training module performs cyclic training and iterative training based on the reward function, designs the score of the reward function based on the degree of proximity between the predicted result and the actual temperature of the side window, and averages the training scores of different batches.

[0030] The prediction result is obtained through the linear regression model y=Xθ+ϵ and input X: temperature y, which is compared with the actual temperature of the outer surface of the side window and the actual temperature of the inner surface of the side window Y. The reward function of the machine learning training module unit is designed so that the closer the model's prediction result is to the actual temperature of the outer surface of the side window and the actual temperature of the inner surface of the side window Y, the higher the score given by the reward function; the reward function is divided into three levels: when the model prediction result exceeds the actual temperature by 20%, the score is -400; when the model prediction result error is between 10% and 20%, the score is 100; when the model prediction result error is between 5% and 10%, the score is 200; when the model prediction result error is between 0% and 5%, the score is 300.

[0031] When the number of samples exceeds the preset number and the average score exceeds the preset value, the model parameters of the trained side window temperature prediction model are obtained; otherwise, the cycle training and iterative training are continued. If the average score of each batch of training exceeds 250 points and the number of samples exceeds the preset number, the training can be considered reliable and sufficient.

[0032] S105. Writing the model parameters of the trained side window temperature prediction model into the machine learning deployment module.

[0033] When the training is completed, the model training vehicle copies the model parameters into the space corresponding to the machine learning deployment module of the model training vehicle, and only copies the parameters. The model parameters can also be shared and uploaded to the designated model application vehicle (which can be a used car or other new car) through TBOX, cloud and other networks to achieve the purpose of fast OTA program.

[0034] S2. The model training vehicle exports the model parameter data of the trained side window temperature prediction model and uploads the data to the cloud, so that the model application vehicle can obtain the data from the cloud, parse and verify it, and obtain the model parameters of the side window temperature prediction model.

[0035] The model training vehicle exports the model parameters of the trained side window temperature prediction model in the form of APDU commands and uploads them to the cloud through the machine learning deployment module, so that the model application vehicle can obtain the APDU commands from the cloud, parse and verify them, and obtain the model parameters of the side window temperature prediction model.

[0036] When the training is completed, in order to achieve the reproducibility of the model parameters, the model training vehicle will export the model parameters in the machine learning deployment module in the form of APDU commands. For this reason, both the model training vehicle and the model application vehicle are equipped with machine learning deployment modules, and both have toolkits for parsing APDU commands. The format includes 5 parts, identification command, operation command, length, data, and response command. In order to ensure the security of model parameters transmitted over the network, the AES algorithm is also used for symmetric encryption and hash verification functions to encrypt and decrypt on the vehicle and cloud to ensure the transmission of parameters. Specifically, the vehicle passes the parameters through the HASH256 algorithm to calculate the summary information, and at the same time, the parameters are encrypted through the AES encryption algorithm. The AES key is composed of the last 8 digits of the vehicle frame number concatenated with 8 zeros. Finally, the value derived from hkdf is used to encrypt the parameters + summary information together through AES and transmit them to the cloud.

[0037] In order to ensure that the models of used cars or other models can also be deployed and loaded, the algorithm loading module and parameter decryption module are mainly used to carry out the work, and the OTA and other Internet transmissions are carried out in the APDU protocol. The specific measures include the following: First, after the cloud receives the data from the model training vehicle, it uses the frame number to decrypt the key information derived from the HDKF algorithm and saves it to the MySQL database. Then, the cloud passes it to the model application vehicle through TBOX (which can be a used car or other new car), and uses the frame number of the model application vehicle to derive the key of the specified vehicle using the HDKF algorithm, and inserts the parameters and hash values ​​into the APDU instruction framework. Then, after the model application vehicle receives the information, it decrypts it in the same way and verifies the hash value to determine whether the information has been tampered with. If it has been tampered with, it sends the information to the cloud, otherwise it continues to replace the model parameters; the model code is similarly replaced. Subsequently, after the model application vehicle parses the parameters, it fills the model parameters in sequence according to the instruction sequence, and after the call is completed, calls the test data to determine whether the parameter loading is successful. Among them, the replacement process is mainly to execute the data in the specified starting space in sequence according to the parameter size. The parameter order and number can also be transmitted using the APDU protocol to ensure consistency in different vehicles; each operation command of APDU can distinguish whether the data is a parameter or other information.

[0038] like Figure 2 As shown, when the model is applied to the vehicle side to implement the automatic defogging method, the following steps are included: A1. The model parameters of the side window temperature prediction model are obtained on the vehicle side of the model application.

[0039] After the model training vehicle exports the model parameter data of the trained side window temperature prediction model and uploads the data to the cloud, the model application vehicle can obtain the APDU command from the cloud and parse and verify it to obtain the model parameters of the side window temperature prediction model.

[0040] A2. Obtain input parameters of the side window temperature prediction model, the input parameters including outside humidity, outside temperature, inside humidity, inside temperature and air conditioning real-time parameters. These input parameters can be used as input X of the model. The air conditioning real-time parameters include air conditioning start / stop status, air conditioning outlet temperature, air conditioning air volume, air conditioning wind speed, air conditioning gear position and air conditioning outlet direction.

[0041] A3. Based on the side window temperature prediction model and the input parameters, a real-time predicted temperature is obtained. A prediction result: temperature y is obtained by a linear regression model y=Xθ+ϵ and input X. Since the model has been trained, temperature y can be used as the real-time predicted temperature, and the real-time predicted temperature includes the real-time predicted temperature of the outer surface of the side window and the real-time predicted temperature of the inner surface of the side window.

[0042] A4. Obtaining the dew point temperature; specifically including: calculating the dew point temperature outside the vehicle based on the humidity outside the vehicle and the temperature outside the vehicle; and calculating the dew point temperature inside the vehicle based on the humidity inside the vehicle and the temperature inside the vehicle.

[0043] The formulas for calculating the dew point temperature outside the vehicle and the dew point temperature inside the vehicle are: Td=(A*t+B*U) / (C*U+D).

[0044] Among them, when Td is the dew point temperature outside the vehicle, t is the temperature outside the vehicle, and U is the humidity outside the vehicle; when Td is the dew point temperature inside the vehicle, t is the temperature inside the vehicle, and U is the humidity inside the vehicle; the units of Td and t are both ℃, the unit of U is %, A is a constant between 0.1780 and 0.2180, B is a constant between 0.0015 and 0.0019, C is a constant between 0.7400 and 0.9400, and D is a constant between 650 and 658. Preferably, A = 0.1980, B = 0.0017, C = 0.8400, and D = 654.

[0045] A5. Compare the real-time predicted temperature with the dew point temperature. When the real-time predicted temperature is not higher than the dew point temperature, adjust the real-time parameters of the air conditioner so that the real-time predicted temperature is higher than the dew point temperature.

[0046] Comparing the real-time predicted temperature of the outer surface of the side window with the dew point temperature outside the vehicle, and comparing the real-time predicted temperature of the inner surface of the side window with the dew point temperature inside the vehicle; When the real-time predicted temperature of the outer surface of the side window is not higher than the dew-point temperature outside the vehicle or / and the real-time predicted temperature of the inner surface of the side window is not higher than the dew-point temperature inside the vehicle, adjust the real-time parameters of the air conditioner so that the real-time predicted temperature of the outer surface of the side window is higher than the dew-point temperature outside the vehicle and the real-time predicted temperature of the inner surface of the side window is higher than the dew-point temperature inside the vehicle.

[0047] Step A4 and step A5 are both implemented through the control module of the thermal management system. The real-time predicted temperature obtained in step A3 is written into the control module. After the control module calculates the dew point temperature, the real-time predicted temperature is compared with the dew point temperature. When the real-time predicted temperature of the outer surface of the side window is not higher than the dew point temperature outside the vehicle or / and the real-time predicted temperature of the inner surface of the side window is not higher than the dew point temperature inside the vehicle, the real-time parameters of the air conditioner are adjusted. The adjustment methods can be: changing the air conditioner gear, changing the air conditioner outlet temperature, changing the air conditioner air volume, changing the direction of the air conditioner outlet, etc.

[0048] like Figure 3 As shown, this embodiment also provides a demisting device, including: A machine learning deployment module for obtaining model parameters of a side window temperature prediction model; A data acquisition and calculation module is used to obtain input parameters of the side window temperature prediction model, wherein the input parameters include outside humidity, outside temperature, inside humidity, inside temperature and real-time air conditioning parameters; An output module, used for obtaining a real-time predicted temperature based on the side window temperature prediction model and the input parameters; And a control module, which is used to obtain the dew point temperature and compare the real-time predicted temperature with the dew point temperature. When the real-time predicted temperature is not higher than the dew point temperature, the real-time parameters of the air conditioner are adjusted to make the real-time predicted temperature higher than the dew point temperature.

[0049] Among them, the machine learning deployment module can upload or receive data to the cloud through TBOX. The machine learning deployment module of the model training vehicle can export the model parameters of the trained side window temperature prediction model in the form of APDU commands and upload them to the cloud through TBOX. The model training vehicle (which can be a used car or other new car) uses the machine learning deployment module to enable the model application vehicle to obtain APDU commands from the cloud through TBOX and then parse and verify them, so as to obtain the model parameters of the side window temperature prediction model, and enable the model training vehicle to share and upload the model parameters to the specified model application vehicle (which can be a used car or other new car), so as to achieve the purpose of fast OTA program.

[0050] This embodiment also provides a vehicle, which uses the above-mentioned automatic defogger method, uses the above-mentioned model training and deployment method, or includes the above-mentioned defogger device.

[0051] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

Claims

1. An automatic defogging method, characterized in that: The automatic defogging method comprises: Obtaining model parameters of a side window temperature prediction model; Obtaining input parameters of the side window temperature prediction model, wherein the input parameters include outside humidity, outside temperature, inside humidity, inside temperature, and air conditioning real-time parameters; Based on the side window temperature prediction model and the input parameters, obtaining a real-time predicted temperature; Get the dew point temperature; The real-time predicted temperature is compared with the dew point temperature, and when the real-time predicted temperature is not higher than the dew point temperature, the real-time parameters of the air conditioner are adjusted to make the real-time predicted temperature higher than the dew point temperature.

2. The automatic defogging method according to claim 1, characterized in that: The obtaining of the dew point temperature comprises: Calculate the dew point temperature outside the vehicle based on the humidity outside the vehicle and the temperature outside the vehicle; The dew point temperature inside the vehicle is calculated based on the humidity and temperature inside the vehicle.

3. The automatic defogging method according to claim 2, characterized in that: The step of comparing the real-time predicted temperature with the dew point temperature and, when the real-time predicted temperature is not higher than the dew point temperature, adjusting the real-time parameters of the air conditioner so that the real-time predicted temperature is higher than the dew point temperature comprises: The real-time predicted temperature includes the real-time predicted temperature of the outer surface of the side window and the real-time predicted temperature of the inner surface of the side window; Comparing the real-time predicted temperature of the outer surface of the side window with the dew point temperature outside the vehicle, and comparing the real-time predicted temperature of the inner surface of the side window with the dew point temperature inside the vehicle; When the real-time predicted temperature of the outer surface of the side window is not higher than the dew point temperature outside the vehicle or / and the real-time predicted temperature of the inner surface of the side window is not higher than the dew point temperature inside the vehicle, adjusting the real-time parameters of the air conditioner so that the real-time predicted temperature of the outer surface of the side window is higher than the dew point temperature outside the vehicle and the real-time predicted temperature of the inner surface of the side window is higher than the dew point temperature inside the vehicle; Among them, the real-time air conditioning parameters include air conditioning start and stop status, air conditioning outlet temperature, air conditioning air volume, air conditioning wind speed, air conditioning gear and air conditioning outlet direction.

4. The automatic defogging method according to claim 2, characterized in that: The formula for calculating the dew point temperature outside the vehicle and the dew point temperature inside the vehicle are both: Td=(A*t+B*U) / (C*U+D); Wherein, when Td is the dew point temperature outside the vehicle, t is the temperature outside the vehicle, and U is the humidity outside the vehicle; when Td is the dew point temperature inside the vehicle, t is the temperature inside the vehicle, and U is the humidity inside the vehicle; The unit of Td and t is ℃, the unit of U is %, A is a constant between 0.1780 and 0.2180, B is a constant between 0.0015 and 0.0019, C is a constant between 0.7400 and 0.9400, and D is a constant between 650 and 658.

5. A model training deployment method, characterized in that: Used to train and deploy model parameters of the side window temperature prediction model as claimed in any one of claims 1 to 4; the model training and deployment method comprises: Based on the cyclic training of the model training vehicle, the model parameters of the final side window temperature prediction model are obtained; The model training vehicle exports the model parameter data of the trained side window temperature prediction model and uploads the data to the cloud, so that the model application vehicle can obtain the data from the cloud, parse and verify it, and obtain the model parameters of the side window temperature prediction model.

6. The model training deployment method according to claim 5, characterized in that: The model training vehicle exports the model parameters of the trained side window temperature prediction model and uploads the data to the cloud, so that the model application vehicle can obtain the data from the cloud and perform analysis and verification to obtain the model parameters of the side window temperature prediction model, specifically including: The model training vehicle exports the model parameters of the trained side window temperature prediction model in the form of APDU commands and uploads them to the cloud through the machine learning deployment module, so that the model application vehicle can obtain the APDU commands from the cloud, parse and verify them, and obtain the model parameters of the side window temperature prediction model.

7. The model training deployment method according to claim 5, characterized in that: The model parameters of the final side window temperature prediction model are obtained by the cyclic training based on the model training vehicle end, including: The model training vehicle obtains and processes the actual temperature of the side window through the data acquisition and calculation module, and the actual temperature of the side window includes the actual temperature of the outer surface of the side window and the actual temperature of the inner surface of the side window; The model training vehicle obtains and processes input parameters through the data acquisition and calculation module, and the input parameters include the humidity outside the vehicle, the temperature outside the vehicle, the humidity inside the vehicle, the temperature inside the vehicle, and the real-time parameters of the air conditioner; The data acquisition and calculation module sends the acquired and processed data to the buffer, which stores the data and sends the stored data in batches to the machine learning training module; The machine learning training module obtains the model parameters of the trained side window temperature prediction model through cyclic training and iterative training; Write the model parameters of the trained side window temperature prediction model into the machine learning deployment module.

8. The model training deployment method according to claim 7, characterized in that: The machine learning training module obtains the model parameters of the final side window temperature prediction model through cyclic training and iterative training, including: The machine learning training module uses the PCA principal component analysis method to screen weights, uses the normal equation to derive the parameters of the linear regression equation, and obtains the prediction results of the model, wherein the prediction results are the real-time predicted temperature of the outer surface of the side window and the real-time predicted temperature of the inner surface of the side window; The machine learning training module performs cyclic training and iterative training based on the reward function, designs a score of the reward function based on the degree of proximity between the predicted result and the actual temperature of the side window, and averages the training scores of different batches respectively; When the number of samples exceeds a preset number and the average score exceeds a preset value, the model parameters of the trained side window temperature prediction model are obtained; otherwise, the cyclic training and iterative training are continued.

9. A demisting device, characterized in that: include: A machine learning deployment module for obtaining model parameters of a side window temperature prediction model; A data acquisition and calculation module is used to obtain input parameters of the side window temperature prediction model, wherein the input parameters include outside humidity, outside temperature, inside humidity, inside temperature and real-time air conditioning parameters; An output module, used for obtaining a real-time predicted temperature based on the side window temperature prediction model and the input parameters; And a control module, which is used to obtain the dew point temperature and compare the real-time predicted temperature with the dew point temperature. When the real-time predicted temperature is not higher than the dew point temperature, the real-time parameters of the air conditioner are adjusted to make the real-time predicted temperature higher than the dew point temperature.

10. A vehicle, characterized in that: The vehicle uses the automatic defog method as described in any one of claims 1-4, uses the model training deployment method as described in any one of claims 5-8, or includes the defog device as described in claim 9.

Images