Vehicle air outlet intelligent adjusting method and system based on machine learning

Abstract

The invention relates to the field of automobile air outlet adjustment, and discloses a vehicle air outlet intelligent adjustment method and system based on machine learning, a frosting area of a glass surface is identified by analyzing a surface state image of a windshield, and current in-vehicle environment data and passenger physiological data are combined with historical data, so that the vehicle air outlet intelligent adjustment method and system based on machine learning are obtained. According to the method, the frosting state of the windshield in the future is predicted, then based on the frosting recognition result and the predicted frosting state, defrosting adjustment evaluation is conducted on the frosting area of the windshield, and priority processing is conducted on adjustment of the frosting area according to the evaluation result and the sight range of a driver. The geometrical characteristics of the frosting areas and the parameters of the air outlets are subjected to matching analysis, the corresponding air outlet control strategy is matched for each area, defrosting partition adjustment on the current windshield is achieved, cooperative adjustment among real-time sensing, prediction, decision making and control of defrosting is facilitated, the frosting dynamic state is pre-judged in real time, invalid energy consumption is avoided, and the defrosting efficiency is improved. And the defrosting safety is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of automobile air outlet adjustment, and more specifically to a vehicle air outlet intelligent adjustment method and system based on machine learning. Background Art

[0002] In recent years, the development of technologies such as artificial intelligence, big data, and cloud computing has made it possible for automobiles to become intelligent. As an important branch of artificial intelligence, machine learning has performed well in pattern recognition, predictive analysis, and other aspects, and has provided technical support for the intelligent adjustment of vehicle air outlets. Traditional vehicle air-conditioning systems usually rely on the driver to manually adjust the temperature, wind speed, and direction of the air outlets. This method has certain limitations and cannot be automatically adjusted according to the in-car environment and passenger needs, resulting in unsatisfactory comfort and energy efficiency. Therefore, a method for intelligent adjustment of vehicle air outlets has emerged. By intelligently adjusting the air outlets, the energy consumption of the air-conditioning system can be reduced while ensuring comfort, and the battery life can be extended, thereby improving the energy efficiency of the entire vehicle.

[0003] However, although the existing intelligent adjustment of vehicle air outlets has many advantages, the existing adjustment methods mainly rely on a single sensor and preset rules, lack multi-dimensional perception and prediction of frosting, and cannot predict frosting dynamics in real time. Unified defrosting is usually adopted in the entire area, which has high energy consumption and affects passenger comfort. It is impossible to accurately control defrosting and optimize energy efficiency, and it is impossible to achieve coordinated adjustment between perception, prediction, decision-making and control of defrosting, and thus it is impossible to take into account energy efficiency, comfort and safety at the same time. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides a vehicle air outlet intelligent adjustment method and system based on machine learning to solve the problems existing in the above-mentioned background technology.

[0005] The present invention provides the following technical solution: a vehicle air outlet intelligent adjustment method based on machine learning, comprising:

[0006] S1: synchronously collect current in-vehicle environment data by setting a multimodal sensor array, and monitor and collect the current surface state image of the windshield in real time through a camera, and at the same time, collect occupant physiological data, and transmit the collected data to a central control unit;

[0007] S2: Analyzing the surface state image of the current windshield to obtain a frosting recognition coefficient, which is used to identify the frosted area on the glass surface;

[0008] S3: By combining and analyzing the current in-vehicle environment data and the occupant's physiological data with historical data, a frosting prediction value is obtained to predict the frosting state of the windshield in the future;

[0009] S4: Perform a comprehensive analysis based on the frosting recognition result of the glass surface and the predicted frosting state to obtain a defrost adjustment evaluation coefficient, which is used to perform a defrost adjustment evaluation on the frosted area of ​​the windshield;

[0010] S5: Based on the calculated defrost adjustment evaluation coefficient and the driver's sight range, priority processing evaluation is performed on the adjustment of the frosted area;

[0011] S6: by matching and analyzing the dynamic distribution of geometric characteristics of the frosting area with the parameters of the air outlet, and calculating the matching adjustment control index, a corresponding air outlet control strategy is matched for each area;

[0012] S7: By transmitting the defrost control strategy and defrost adjustment priority ranking of each zone to the central control unit, the defrost zone adjustment of the current windshield is achieved.

[0013] Preferably, the S1 installs a multimodal sensor array and a camera in the vehicle, and deploys a machine learning model and data processing software in a central control unit; uses the multimodal sensor array to collect the temperature, humidity and CO in the vehicle in real time. 2 Concentration, use the camera to capture the surface status of the windshield in real time, and take images of the windshield surface status, and install physiological sensors on the car seats to collect the occupant's body surface temperature, heart rate and breathing rate. The data collected by the sensors and cameras are transmitted to the central control unit through the in-vehicle network, and the data processing software in the central control unit pre-processes the data, including data cleaning, data conversion and data standardization.

[0014] Preferably, the S2 uses a convolutional neural network to analyze the pre-processed windshield surface state image, monitor the frost and fog state of the glass surface in real time, and identify the frosted area;

[0015] The local features of the image are extracted through the convolution layer, and the extracted feature map is converted into a one-dimensional feature vector. The softmax activation function is then used to output the frosting probability of each pixel. The frosting probability of each pixel output is the frosting recognition coefficient. The frosting recognition coefficient is compared with the preset recognition threshold to determine all the pixels where frost appears. The pixels whose frosting recognition coefficient is greater than the preset recognition threshold are identified as frosted pixels and marked to form a frosted area.

[0016] When a frosted area on the windshield is detected, an analysis of the defrost control for the windshield is performed.

[0017] Preferably, the S3 collects historical in-vehicle environment data and occupant physiological data over a period of time, and combines them with the currently collected in-vehicle environment data and occupant physiological data in chronological order to form a time series data set, and uses the LSTM time series prediction model to predict and analyze the future frosting of the windshield to obtain a frosting prediction value, and compares the frosting prediction value with a preset frosting threshold to predict the potential frosting trend of the vehicle windshield. When the frosting prediction value is greater than the preset frosting threshold, it is considered that the windshield will be frosted in the future.

[0018] Preferably, the S4 divides the windshield into several areas, and is used to accumulate the differences between the frost identification coefficients of all frosted pixels screened out in the feature map of each area and the preset frost threshold, calculate the frost severity of the sub-area, and perform a comprehensive analysis with the frost prediction value to obtain the defrost adjustment evaluation coefficient, and its specific calculation formula is: Among them, S v represents the defrost adjustment evaluation coefficient of the vth zone, P(y v，k ) represents the frosting recognition coefficient of pixel k in the feature map of the vth region, θ 1 represents the preset recognition threshold, max(P(y v，k )-θ 1 , 0) means when P(y v，k ) is greater than θ 1 When , the difference is accumulated, otherwise the difference is 0, h represents the frost prediction value, θ 2 represents the preset frosting threshold, and N represents the total number of pixels.

[0019] Preferably, the S5 prioritizes the defrosting of each area divided by the windshield by combining the calculated defrosting adjustment evaluation coefficient of each area with the driver's line of sight, and the defrosting adjustment evaluation coefficient S v The larger the value, the higher the priority. The priority is then adjusted based on the area within the driver's field of vision, and the final priority adjustment result is transmitted to the central control unit.

[0020] Preferably, the S6 is used to collect parameters of the vehicle air outlet, including wind speed, wind direction and temperature, and images of frosted areas, including position, size and shape, extract geometric features of the frosted areas and parameter features of the air outlets from the collected data, dynamically generate a thermodynamic distribution map based on the geometric features of the fogged area using a gradient layered defrosting algorithm, display the temperature gradients of different areas on the windshield on the thermodynamic distribution map, analyze the air flow direction and intensity regulated by the air outlet based on the parameter features of the air outlet, match the thermodynamic distribution map with the air flow direction and intensity regulated by the air outlet, analyze the defrosting heat and wind speed required for each frosted area, calculate a matching adjustment control index, match a corresponding air outlet control strategy for each area, and transmit the assigned control strategy to the central control unit to adjust the parameters of the air outlet to perform defrost control on different windshield areas.

[0021] Preferably, the S7 adjusts the parameters of the TEC module and the fan according to the control strategy of each area, and then sets a target defrost temperature for each frosting area, heats or cools the area through the TEC module to make the area temperature close to the target value, and uses a temperature sensor to monitor the actual temperature. According to the error between the target temperature and the actual temperature, the PID controller adjusts the current of the TEC module to achieve a temperature accuracy of 0.1°C, and sets a target wind speed for each frosting area, adjusts the fan speed through a PWM signal to achieve the target wind speed, uses a wind speed sensor to monitor the actual wind speed, and adjusts the duty cycle of the PWM signal through the PID controller to achieve a wind speed accuracy of ±5% according to the error between the target wind speed and the actual wind speed; and monitors the temperature and wind speed in real time, adjusts the TEC module and PWM signal through the PID controller to continuously maintain the target temperature and target wind speed, and when the frosting area reaches the defrosting target, gradually reduces the output of the TEC module and the fan until the defrosting process is completed.

[0022] To achieve the above object, the present invention provides the following technical solution: a vehicle air outlet intelligent adjustment system based on machine learning, implementing the above vehicle air outlet intelligent adjustment method based on machine learning, including:

[0023] Data acquisition module: It collects the current in-vehicle environment data synchronously by setting a multi-modal sensor array, and monitors and collects the surface state image of the current windshield in real time through the camera. At the same time, it collects the physiological data of the occupants and transmits the collected data to the central control unit;

[0024] Image recognition module: by analyzing the current windshield surface state image, the frosting recognition coefficient is obtained to identify the frosted area on the glass surface;

[0025] Frost prediction module: By combining and analyzing the current in-vehicle environment data and occupant physiological data with historical data, the frost prediction value is obtained to predict the frosting state of the windshield in the future.

[0026] Defrost adjustment evaluation module: Based on the comprehensive analysis of the frost recognition results and the predicted frost state of the glass surface, the defrost adjustment evaluation coefficient is obtained, which is used to evaluate the defrost adjustment of the frosted area of ​​the windshield;

[0027] Priority processing module: based on the calculated defrost adjustment evaluation coefficient and the driver's line of sight, priority processing evaluation is performed on the adjustment of the frosting area;

[0028] Matching and adjustment module: by matching and analyzing the dynamic distribution of the geometric characteristics of the frosting area with the parameters of the air outlet, and calculating the matching adjustment control index, the corresponding air outlet control strategy is matched for each area;

[0029] Defrost execution module: By transmitting the defrost control strategy and defrost adjustment priority ranking of each zone to the central control unit, the defrost zone adjustment of the current windshield is realized.

[0030] Technical effects and advantages of the present invention:

[0031] The present invention synchronously collects the current in-vehicle environment data by setting a multimodal sensor array, and monitors and collects the surface state image of the current windshield in real time through a camera. At the same time, the physiological data of the occupants are collected, and the collected data is transmitted to a central control unit. By analyzing the surface state image of the current windshield, a frost recognition coefficient is obtained, which is used to identify the frosted area on the glass surface. By combining and analyzing the current in-vehicle environment data and the physiological data of the occupants with historical data, a frost prediction value is obtained to predict the frosting state of the windshield in the future. A comprehensive analysis is performed based on the frost recognition result of the glass surface and the predicted frosting state to obtain a defrost adjustment evaluation coefficient, which is used to perform defrost adjustment evaluation on the frosted area of ​​the windshield. Based on the calculated defrost adjustment evaluation coefficient and the driver's field of vision, the adjustment of the frosted area is evaluated with priority, and then the number of frosted areas is evaluated. The dynamic distribution of any characteristics is matched and analyzed with the parameters of the air outlet, and the matching adjustment control index is calculated to match the corresponding air outlet control strategy for each area. Finally, the defrost control strategy and defrost adjustment priority of each area are transmitted to the central control unit to realize the defrost zone adjustment of the current windshield; historical data and real-time data are analyzed by machine learning to predict future frosting trends, and the current in-vehicle environment data, windshield surface images and occupant physiological data are collected synchronously, which is conducive to multi-dimensional perception and prediction of frosting, real-time prediction of frosting dynamics, and precise positioning of frosting areas through image analysis to avoid ineffective energy consumption. The distribution of frosting areas is dynamically matched with the air outlet parameters, which can achieve precise zoning control of defrosting, facilitate real-time coordinated adjustment between perception, prediction, decision-making and control of defrosting, and ensure the balance of energy efficiency, comfort and safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a diagram of the method steps of the present invention.

[0033] Figure 2 It is a system structure block diagram of the present invention. DETAILED DESCRIPTION

[0034] The technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the present invention. In addition, the forms of the various structures recorded in the following embodiments are merely illustrative. The vehicle air outlet intelligent adjustment method and system based on machine learning involved in the present invention are not limited to the various structures recorded in the following embodiments. All other implementations obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention.

[0035] like Figure 1 The embodiment shown provides a vehicle air outlet intelligent adjustment method based on machine learning, including:

[0036] S1: The current in-vehicle environment data is synchronously collected by setting a multimodal sensor array, and the surface state image of the current windshield is monitored and collected in real time through a camera. At the same time, the physiological data of the occupants is collected and transmitted to the central control unit.

[0037] In this embodiment, the S1 installs a multimodal sensor array and a camera in the vehicle, and deploys a machine learning model and data processing software in the central control unit; the multimodal sensor array is used to collect the temperature, humidity and CO in the vehicle in real time. 2 Concentration, use the camera to capture the surface status of the windshield in real time, and take images of the windshield surface status, and install physiological sensors on the car seats to collect the occupant's body surface temperature, heart rate and breathing rate. The data collected by the sensors and cameras are transmitted to the central control unit through the in-vehicle network, and the data processing software in the central control unit pre-processes the data, including data cleaning, data conversion and data standardization.

[0038] It should be specified that the specific data collection process includes: when the vehicle is started or the physiological sensor senses that the occupant has entered the vehicle, the multimodal sensor array and the camera are automatically activated, and data is collected from the sensors and the camera at a fixed frequency, ensuring that the data collected by the multimodal sensor array and the camera are synchronized in time, integrating the data collected by different sensors to form a comprehensive data set, transmitting the data to the central control unit through the high-speed communication network in the vehicle, and performing pre-processing operations in the central control unit.

[0039] S2: Analyze the surface state image of the current windshield to obtain a frosting recognition coefficient for identifying the frosted area on the glass surface.

[0040] In this embodiment, the S2 uses a convolutional neural network to analyze the pre-processed windshield surface state image, monitor the frost and fog state of the glass surface in real time, and identify the frosted area;

[0041] The local features of the image are extracted through the convolution layer, and the extracted feature map is converted into a one-dimensional feature vector. The softmax activation function is then used to output the frosting probability of each pixel. The frosting probability of each pixel output is the frosting recognition coefficient. The frosting recognition coefficient is compared with the preset recognition threshold to determine all the pixels where frost appears. The pixels whose frosting recognition coefficient is greater than the preset recognition threshold are identified as frosted pixels and marked to form a frosted area.

[0042] When a frosted area on the windshield is detected, an analysis of the defrost control for the windshield is performed.

[0043] It should be specifically noted that the size of the windshield surface state image is adjusted to meet the requirements of the CNN input layer, and the pixel values ​​of the image are normalized, and the processed image is input into the trained convolutional neural network model to identify the windshield surface state;

[0044] The specific analysis method of using the convolutional neural network model to identify the surface state of the windshield is as follows:

[0045] Step S211: Use the convolutional layer of CNN to extract local features of the image. The specific extraction formula is: Among them, F(i, j) represents an element in the feature map, I represents the input image, K represents the convolution kernel, (i, j) represents the position in the feature map, and (m, n) represents the position in the convolution kernel;

[0046] Step S212: converting the feature map output by the convolutional layer into a one-dimensional feature vector through a fully connected layer;

[0047] Step S213: Use the softmax activation function to output the frost probability of each pixel. The specific function calculation formula is: Among them, P(y k ) represents the frost probability of pixel k, y k Expressed as a normalized predicted value, the frost probability P(y k ) is the frost recognition coefficient;

[0048] Step S214: Set the frost identification coefficient P(y k ) and the preset recognition threshold θ 1 Comparison is used to filter out all frosted pixels. When P(y k )>θ 1 When , the pixel k is identified as frosted, and all the pixels identified as frosted are screened out at the same time, and all the pixels identified as frosted are combined into a frosted area.

[0049] S3: By combining and analyzing the current in-vehicle environment data and the occupant's physiological data with historical data, a frosting prediction value is obtained to predict the frosting state of the windshield in the future.

[0050] In this embodiment, S3 collects historical in-vehicle environment data and occupant physiological data over a period of time, and combines them with the currently collected in-vehicle environment data and occupant physiological data in chronological order to form a time series data set, and uses the LSTM time series prediction model to predict and analyze the future frosting of the windshield to obtain a frosting prediction value, and compares the frosting prediction value with a preset frosting threshold to predict the potential frosting trend of the vehicle windshield. When the frosting prediction value is greater than the preset frosting threshold, it is considered that the windshield will be frosted in the future.

[0051] It should be noted that the specific analysis method for predicting windshield frosting through the LSTM time series prediction model is as follows:

[0052] Step S311: constructing an LSTM model, including defining the architecture of the LSTM model, selecting an appropriate loss function and optimizer to compile the model;

[0053] Step S312: Use the time series data set to train the LSTM model, and use the trained LSTM model to predict the new time series data set. The specific model prediction formula is h = f(LSTM(X t ; W)), where h represents the frost prediction value at time point t, f(LSTM(·)) represents the LSTM model, and X t represents the input feature vector at time point t, including the in-vehicle environment data and the occupant physiological data, and W represents the weight of the LSTM model;

[0054] Step S313: By comparing the predicted value h with the preset frost threshold θ 2 For comparison, if h>θ 2 , it is believed that the windshield will be frosted in the future.

[0055] S4: A comprehensive analysis is performed based on the frosting recognition result of the glass surface and the predicted frosting state to obtain a defrost adjustment evaluation coefficient, which is used to perform a defrost adjustment evaluation on the frosted area of ​​the windshield.

[0056] In this embodiment, the S4 divides the windshield into several areas, and is used to accumulate the difference between the frost identification coefficients of all frosted pixels screened out in the feature map of each area and the preset frost threshold, calculate the frost severity of the sub-area, and perform a comprehensive analysis with the frost prediction value to obtain the defrost adjustment evaluation coefficient. The specific calculation formula is: Among them, S v represents the defrost adjustment evaluation coefficient of the vth zone, P(y v，k ) represents the frosting recognition coefficient of pixel k in the feature map of the vth region, θ 1 represents the preset recognition threshold, max(P(y v，k )-θ 1 , 0) means when P(y v，k ) is greater than θ 1 When , the difference is accumulated, otherwise the difference is 0, h represents the frost prediction value, θ 2 represents the preset frosting threshold, and N represents the total number of pixels.

[0057] S5: Based on the calculated defrost adjustment evaluation coefficient and the driver's sight range, priority processing evaluation is performed on the adjustment of the frosted area.

[0058] In this embodiment, S5 prioritizes the defrosting of each area divided by the windshield by combining the calculated defrosting adjustment evaluation coefficient of each area with the driver's line of sight, and the defrosting adjustment evaluation coefficient S v The larger the value, the higher the priority. The priority is then adjusted based on the area within the driver's field of vision, and the final priority adjustment result is transmitted to the central control unit.

[0059] It should be specifically stated that the specific execution process of the defrost priority ranking is as follows:

[0060] Step S511: First adjust the evaluation coefficient S according to the defrosting v The size of the windshield is used to determine the defrosting priority of the windshield, and S v The larger the value, the more serious the frost in the area, and a value based on S v The priority list is the result of the preliminary sorting;

[0061] Step S512: marking the areas on the windshield that directly affect the driver's line of sight, which areas generally include the center of the windshield and the part directly in front of the driver's line of sight;

[0062] Step S513: According to the preliminary sorting result, check whether each area is within the driver's line of sight to adjust the order of the priority list. The specific adjustment calculation formula is Z v =S v +α×C v , where Z v represents the comprehensive priority of the vth region, S v represents the defrost adjustment evaluation coefficient of the vth area, α represents the weight coefficient within the adjustment range, C v Indicates that area v is within the driver's line of sight. If it is within the driver's line of sight, then C v =1, otherwise, C v =0;

[0063] Step S514: According to the comprehensive priority Z v The defrost zones are finally sorted to determine the defrost priority for each zone.

[0064] S6: By matching and analyzing the dynamic distribution of the geometric characteristics of the frosting area with the parameters of the air outlet, and calculating the matching adjustment control index, the corresponding air outlet control strategy is matched for each area.

[0065] In this embodiment, the S6 is used to collect parameters of the vehicle air outlet, including wind speed, wind direction and temperature, and images of frosted areas, including position, size and shape, extract geometric features of the frosted area and parameter features of the air outlet from the collected data, and dynamically generate a thermodynamic distribution map using a gradient layered defrosting algorithm based on the geometric features of the fogged area. The temperature gradients of different areas on the windshield are displayed on the thermodynamic distribution map, and the air flow direction and intensity regulated by the air outlet are analyzed based on the parameter characteristics of the air outlet. The thermodynamic distribution map is then matched with the air flow direction and intensity regulated by the air outlet, the defrosting heat and wind speed required for each frosted area are analyzed, and a matching adjustment control index is calculated. The corresponding air outlet control strategy is matched for each area, and the assigned control strategy is transmitted to the central control unit to adjust the parameters of the air outlet to perform defrost control on different windshield areas.

[0066] It should be noted that the specific analysis method for the matching adjustment control index is as follows:

[0067] Step S611: extracting geometric features of the frosted area from the image data, including the position (a, b), area A and shape G, and measuring and acquiring the wind speed V, wind direction D and temperature T of the air outlet through the sensor;

[0068] Step S612: dynamically generate a thermodynamic distribution map using a gradient layered defrosting algorithm based on the extracted geometric features, and display the temperature gradients of different areas of the windshield on the distribution map. The specific thermodynamic distribution map generation formula is Td(a, b)=G(a, b, A, G)+L(V, D, T), where Td represents the temperature on the thermodynamic distribution map, G represents a function based on the geometric features, and B represents a function based on the air outlet parameters;

[0069] Step S613: According to the temperature of the frosted area on the thermodynamic distribution diagram and the air flow direction and intensity regulated by the air outlet, the defrosting heat and wind speed required for each frosted area are analyzed, and the defrosting heat is calculated as Q v =ρ×c×A v ×(T′-Td v ), where ρ represents the air density, c represents the specific heat capacity of air, and A v represents the area of ​​the vth frosting area, T′ represents the air temperature blowing from the air outlet to the frosting area, Td v represents the temperature of the vth frosted area shown on the distribution diagram;

[0070] Step S614: Based on the analyzed defrosting heat, the matching analysis is performed with the wind speed required for defrosting, and the matching adjustment control index is calculated as Among them, Y v represents the matching regulation control index of the vth frosting area, Q vIndicates the defrosting heat required for each frosting area, V v ′ represents the defrosting wind speed required for each frosting area, r represents the wind speed influence index, which is usually greater than 1. The specific value is determined by the nonlinear effect of the increase in wind speed on the defrosting effect, β 1 and β 2 Represents the weight coefficient, which adjusts the control index Y according to the matching v , match the corresponding air outlet control strategy for each frosting area, and transmit the control strategy to the central control unit to adjust the parameters of the air outlet.

[0071] S7: By transmitting the defrost control strategy and defrost adjustment priority ranking of each zone to the central control unit, the defrost zone adjustment of the current windshield is achieved.

[0072] In this embodiment, the S7 adjusts the parameters of the TEC module and the fan according to the control strategy of each area, and then sets a target defrost temperature for each frosted area. The TEC module is heated or cooled to make the area temperature close to the target value, and a temperature sensor is used to monitor the actual temperature. According to the error between the target temperature and the actual temperature, the PID controller adjusts the current of the TEC module to achieve a temperature accuracy of 0.1°C, and sets a target wind speed for each frosted area. The fan speed is adjusted by a PWM signal to achieve the target wind speed. A wind speed sensor is used to monitor the actual wind speed. According to the error between the target wind speed and the actual wind speed, the PID controller adjusts the duty cycle of the PWM signal to achieve a wind speed accuracy of ±5%; and the temperature and wind speed are monitored in real time, and the TEC module and PWM signals are adjusted by the PID controller to continuously maintain the target temperature and target wind speed. When the frosted area reaches the defrost target, the output of the TEC module and the fan is gradually reduced until the defrosting process is ended.

[0073] like Figure 2 The present embodiment shown provides an implementation system corresponding to the vehicle air outlet intelligent adjustment method based on machine learning, including a data acquisition module, an image recognition module, a frost prediction module, a defrost adjustment evaluation module, a priority processing module, a matching adjustment module and a defrost execution module, the data acquisition module is connected to the image recognition module, the data acquisition module is connected to the frost prediction module, the image recognition module is connected to the defrost adjustment evaluation module, the frost prediction module is connected to the defrost adjustment evaluation module, the defrost adjustment evaluation module is connected to the priority processing module, the priority processing module is connected to the defrost execution module, and the matching adjustment module is connected to the defrost execution module.

[0074] The data acquisition module synchronously collects the current in-vehicle environment data by setting a multimodal sensor array, and monitors and collects the current surface state image of the windshield in real time through a camera, and at the same time, collects the physiological data of the occupants, and transmits the collected data to the central control unit;

[0075] The image recognition module obtains a frosting recognition coefficient by analyzing the current surface state image of the windshield, which is used to identify the frosted area on the glass surface;

[0076] The frosting prediction module combines and analyzes the current in-vehicle environment data and the occupant's physiological data with the historical data to obtain a frosting prediction value and predict the frosting state of the windshield within a certain period of time in the future;

[0077] The defrost adjustment evaluation module performs a comprehensive analysis based on the frosting recognition result of the glass surface and the predicted frosting state to obtain a defrost adjustment evaluation coefficient for performing a defrost adjustment evaluation on the frosted area of ​​the windshield;

[0078] The priority processing module performs a priority processing evaluation on the adjustment of the frosted area based on the calculated defrost adjustment evaluation coefficient and the driver's line of sight;

[0079] The matching and adjustment module matches the corresponding air outlet control strategy for each area by analyzing the dynamic distribution of the geometric characteristics of the frosting area and the parameters of the air outlet, and calculating the matching and adjustment control index;

[0080] The defrost execution module transmits the defrost control strategy and defrost adjustment priority ranking of each zone to the central control unit, thereby realizing defrost zone adjustment for the current windshield.

[0081] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

[0082] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A vehicle air outlet intelligent adjustment method based on machine learning, characterized in that: include: S1: synchronously collect current in-vehicle environment data by setting a multimodal sensor array, and monitor and collect the current surface state image of the windshield in real time through a camera, and at the same time, collect occupant physiological data, and transmit the collected data to a central control unit; S2: Analyzing the surface state image of the current windshield to obtain a frosting recognition coefficient, which is used to identify the frosted area on the glass surface; S3: By combining and analyzing the current in-vehicle environment data and the occupant's physiological data with historical data, a frosting prediction value is obtained to predict the frosting state of the windshield in the future; S4: Perform a comprehensive analysis based on the frosting recognition result of the glass surface and the predicted frosting state to obtain a defrost adjustment evaluation coefficient, which is used to perform a defrost adjustment evaluation on the frosted area of ​​the windshield; S5: Based on the calculated defrost adjustment evaluation coefficient and the driver's sight range, priority processing evaluation is performed on the adjustment of the frosted area; S6: by matching and analyzing the dynamic distribution of geometric characteristics of the frosting area with the parameters of the air outlet, and calculating the matching adjustment control index, a corresponding air outlet control strategy is matched for each area; S7: By transmitting the defrost control strategy and defrost adjustment priority ranking of each zone to the central control unit, the defrost zone adjustment of the current windshield is achieved.

2. The vehicle air outlet intelligent adjustment method based on machine learning according to claim 1 is characterized in that: The S1 is implemented by installing a multimodal sensor array and camera in the vehicle and deploying a machine learning model and data processing software in a central control unit; A multimodal sensor array is used to collect the temperature, humidity and CO2 concentration in the car in real time. A camera is used to capture the surface status of the windshield in real time and take images of the windshield surface status. Physiological sensors are installed on the car seats to collect the occupants' body surface temperature, heart rate and breathing rate. The data collected by the sensors and cameras are transmitted to the central control unit via the in-vehicle network, and the data processing software in the central control unit pre-processes the data, including data cleaning, data conversion and data standardization.

3. The vehicle air outlet intelligent adjustment method based on machine learning according to claim 1 is characterized in that: The S2 uses a convolutional neural network to analyze the pre-processed windshield surface state image, monitor the frost and fog state of the glass surface in real time, and identify the frosted area; The local features of the image are extracted through the convolution layer, and the extracted feature map is converted into a one-dimensional feature vector. The softmax activation function is then used to output the frosting probability of each pixel. The frosting probability of each pixel output is the frosting recognition coefficient. The frosting recognition coefficient is compared with the preset recognition threshold to determine all the pixels where frost appears. The pixels whose frosting recognition coefficient is greater than the preset recognition threshold are identified as frosted pixels and marked to form a frosted area. When a frosted area on the windshield is detected, an analysis of the defrost control for the windshield is performed.

4. The vehicle air outlet intelligent adjustment method based on machine learning according to claim 1 is characterized in that: The S3 collects historical in-vehicle environment data and occupant physiological data over a period of time, and combines them with the currently collected in-vehicle environment data and occupant physiological data in chronological order to form a time series data set, and uses the LSTM time series prediction model to predict and analyze the future frosting of the windshield to obtain a frosting prediction value, and compares the frosting prediction value with a preset frosting threshold to predict the potential frosting trend of the vehicle windshield. When the frosting prediction value is greater than the preset frosting threshold, it is considered that the windshield will be frosted in the future.

5. The vehicle air outlet intelligent adjustment method based on machine learning according to claim 1 is characterized in that: The S4 divides the windshield into several areas, and is used to accumulate the difference between the frost identification coefficients of all frost pixels screened out in the feature map of each area and the preset frost threshold, calculate the frost severity of the sub-area, and perform a comprehensive analysis with the frost prediction value to obtain the defrost adjustment evaluation coefficient. The specific calculation formula is: Among them, S v represents the defrost adjustment evaluation coefficient of the vth zone, P(y v，k ) represents the frost recognition coefficient of pixel k in the feature map of the vth region, θ1 represents the preset recognition threshold, max(P(y v，k )-θ1,0) means when P(y v，k ) is greater than θ1, the difference is accumulated, otherwise the difference is 0, h represents the frost prediction value, θ2 represents the preset frost threshold, and N represents the total number of pixels.

6. The vehicle air outlet intelligent adjustment method based on machine learning according to claim 1 is characterized in that: The S5 prioritizes the defrosting of each area divided by the windshield by combining the calculated defrosting adjustment evaluation coefficient of each area with the driver's sight range, and the defrosting adjustment evaluation coefficient S v The larger the value, the higher the priority. The priority is then adjusted based on the area within the driver's field of vision, and the final priority adjustment result is transmitted to the central control unit.

7. The vehicle air outlet intelligent adjustment method based on machine learning according to claim 1 is characterized in that: The S6 is used to collect parameters of the vehicle air outlet, including wind speed, wind direction and temperature, and images of frosted areas, including position, size and shape, extract geometric features of the frosted area and parameter features of the air outlet from the collected data, dynamically generate a thermodynamic distribution map using a gradient layered defrosting algorithm based on the geometric features of the fogged area, display the temperature gradients of different areas on the windshield on the thermodynamic distribution map, analyze the air flow direction and intensity regulated by the air outlet based on the parameter features of the air outlet, match the thermodynamic distribution map with the air flow direction and intensity regulated by the air outlet, analyze the defrosting heat and wind speed required for each frosted area, calculate a matching adjustment control index, match a corresponding air outlet control strategy for each area, and transmit the assigned control strategy to the central control unit to adjust the parameters of the air outlet to perform defrost control on different windshield areas.

8. The vehicle air outlet intelligent adjustment method based on machine learning according to claim 1 is characterized in that: The S7 adjusts the parameters of the TEC module and the fan according to the control strategy of each area, and then sets a target defrost temperature for each frosting area, heats or cools the area through the TEC module to make the area temperature close to the target value, and uses a temperature sensor to monitor the actual temperature. According to the error between the target temperature and the actual temperature, the PID controller adjusts the current of the TEC module to achieve a temperature accuracy of 0.1°C, and sets a target wind speed for each frosting area, adjusts the fan speed through a PWM signal to achieve the target wind speed, uses a wind speed sensor to monitor the actual wind speed, and adjusts the duty cycle of the PWM signal through the PID controller to achieve a wind speed accuracy of ±5% according to the error between the target wind speed and the actual wind speed; and monitors the temperature and wind speed in real time, adjusts the TEC module and PWM signal through the PID controller, and continuously maintains the target temperature and target wind speed. When the frosting area reaches the defrosting target, the output of the TEC module and the fan is gradually reduced until the defrosting process is completed.

9. A vehicle air outlet intelligent adjustment system based on machine learning, implementing a vehicle air outlet intelligent adjustment method based on machine learning as described in any one of claims 1 to 8, characterized in that: include: Data acquisition module: It collects the current in-vehicle environment data synchronously by setting a multi-modal sensor array, and monitors and collects the surface state image of the current windshield in real time through the camera. At the same time, it collects the physiological data of the occupants and transmits the collected data to the central control unit; Image recognition module: by analyzing the current windshield surface state image, the frosting recognition coefficient is obtained to identify the frosted area on the glass surface; Frost prediction module: By combining and analyzing the current in-vehicle environment data and occupant physiological data with historical data, the frost prediction value is obtained to predict the frosting state of the windshield in the future. Defrost adjustment evaluation module: Based on the comprehensive analysis of the frost recognition results and the predicted frost state of the glass surface, the defrost adjustment evaluation coefficient is obtained, which is used to evaluate the defrost adjustment of the frosted area of ​​the windshield; Priority processing module: based on the calculated defrost adjustment evaluation coefficient and the driver's line of sight, priority processing evaluation is performed on the adjustment of the frosting area; Matching and adjustment module: by matching and analyzing the dynamic distribution of the geometric characteristics of the frosting area with the parameters of the air outlet, and calculating the matching adjustment control index, the corresponding air outlet control strategy is matched for each area; Defrost execution module: By transmitting the defrost control strategy and defrost adjustment priority ranking of each zone to the central control unit, the defrost zone adjustment of the current windshield is realized.

Images