In-vehicle air quality adjusting system and method based on machine learning

Abstract

The invention discloses an in-vehicle air quality adjusting system and method based on machine learning, and belongs to the technical field of vehicle-mounted air conditioning, the adjusting system comprises a data collecting and preprocessing module, an in-vehicle CO2 concentration real-time prediction module and an in-vehicle environment control module, and the data collecting and preprocessing module collects in-vehicle data in real time; the in-train CO2 concentration real-time prediction module obtains in-train data, predicts the in-train CO2 concentration of a next long tunnel or a tunnel group in real time, and generates a corresponding control signal based on a prediction result. The in-vehicle environment control module obtains the control signal and automatically adjusts the opening degree of the air conditioner air duct valve according to the control signal. By adopting the system and the method, real-time prediction of the CO2 concentration in the vehicle is realized based on machine learning, and the opening degree of the air duct valve is dynamically adjusted in combination with a control strategy, so that the air quality in the vehicle is in a suitable range, and the health and comfort of sensitive people in the vehicle are guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle air conditioning, and particularly to an in-vehicle air quality regulation system and regulation method based on machine learning. Background Art

[0002] CO 2 As a colorless and odorless gas, it is an important indicator for evaluating the ventilation efficiency and oxygen content in a closed space. In a closed environment, due to the increase in the concentration of CO 2 usually being difficult to detect, its potential impact on the in-vehicle air quality cannot be ignored. Although CO 2 itself is non-toxic and is generally considered harmless to humans at low concentrations, under specific conditions, especially when a high-speed train passes through a very long tunnel or a tunnel group, the accumulation of its concentration may have an adverse impact on sensitive populations (such as individuals with respiratory diseases, cardiovascular diseases, the elderly, children, and pregnant women), and may even induce acute or chronic health problems.

[0003] During the operation of a high-speed train, the stability of the in-vehicle air environment is crucial. Especially when the train passes through a tunnel, the sudden change in external air pressure may have an adverse impact on the in-vehicle environment. To cope with this change, most of the existing air conditioning systems are equipped with passive pressure protection devices. This device closes the fresh air door and the exhaust air door to make the vehicle interior free from the interference of external pressure changes to achieve the stability of the in-vehicle air environment. When the train is about to enter the tunnel, the air conditioning system will immediately switch to the return air mode, suspend the introduction of external air, keep the in-vehicle air pressure constant, and ensure passenger comfort and equipment safety. At the same time, the pressure protection valve is closed to ensure that the in-vehicle air pressure has a relatively small fluctuation range during the tunnel crossing until the train exits the tunnel area. However, when external air cannot be introduced, the concentration of CO 2 in the vehicle will rise sharply, affecting the in-vehicle air quality. Currently, there is a lack of real-time collection and intelligent regulation based on the change in the concentration of CO 2 in the train, and an efficient and energy-saving method has not been provided to maintain the air quality within an appropriate range.

[0004] Therefore, a new solution needs to be provided for the above problems. Summary of the Invention

[0005] The object of the present invention is to provide an in-vehicle air quality regulation system and regulation method based on machine learning, which realizes the real-time prediction of the concentration of CO 2 in the vehicle based on machine learning, and combines control strategies to dynamically adjust the opening degree of the air duct valve, so that the concentration of CO 2 in the vehicle is within an appropriate range, thereby controlling the in-vehicle air quality and ensuring the health and comfort of sensitive populations in the vehicle.

[0006] To achieve the above object, the present invention provides an in-vehicle air quality regulation system based on machine learning, including a data acquisition and preprocessing module, a real-time prediction module for in-vehicle CO 2 concentration, and an in-vehicle environment control module. The data acquisition and preprocessing module is connected to the real-time prediction module for in-vehicle CO 2 concentration. Both the data acquisition and preprocessing module and the real-time prediction module for in-vehicle CO 2 concentration are connected to the in-vehicle environment control module;

[0007] The data acquisition and preprocessing module is used to obtain the in-vehicle CO 2 concentration, the number of people in the train, the train speed, and the tunnel length in real time, and clean and preprocess the data;

[0008] The real-time prediction module for in-vehicle CO 2 concentration obtains the data from the data acquisition and preprocessing module, and predicts the in-vehicle CO 2 concentration when the train passes through the next long tunnel or tunnel group based on the obtained data, and generates corresponding control signals based on the prediction results;

[0009] The in-vehicle environment control module obtains the control signals generated by the real-time prediction module for in-vehicle CO 2 concentration, automatically adjusts the opening degree of the air-conditioning duct valve according to the control signals to optimize the air quality, and feeds back the regulation information to the data acquisition and preprocessing module.

[0010] Preferably, the data acquisition and preprocessing module includes a trigger, a sensor unit, a real-time data transmission unit, and a data cleaning and standardization unit; the sensor unit includes a CO 2 dynamic sensor assembly, a speed sensor, and an infrared thermal imaging sensor assembly;

[0011] The trigger controls the start and stop of data acquisition according to the speed sensor signal;

[0012] The CO 2 dynamic sensor assembly is used to detect the in-vehicle CO 2 concentration in real time, and transmits it to the data cleaning and standardization unit through the real-time data transmission unit;

[0013] The infrared thermal imaging sensor unit counts the number of passengers in the carriage and transmits it to the data cleaning and standardization unit through the real-time data transmission unit;

[0014] The data cleaning and standardization unit cleans and standardizes the data, and the processed data is automatically transmitted to the real-time prediction module for in-vehicle CO 2 concentration.

[0015] Preferably, the train speed data is obtained through the train operation control system; the tunnel length data is calculated by using the track database and the train positioning system, in combination with inertial navigation and odometer data.

[0016] Preferably, the in-vehicle CO 2 concentration real-time prediction module includes a machine learning prediction model unit, a model evaluation and feedback unit, an online update and incremental learning unit, and a real-time prediction and decision-making unit; the machine learning prediction model unit is respectively connected to the online update and incremental learning unit, the model evaluation and feedback unit, and the real-time prediction and decision-making module;

[0017] The machine learning prediction model unit embeds a prediction model based on machine learning, and real-time predicts the in-vehicle CO 2 concentration when the train passes through the next long tunnel or tunnel group according to the data transmitted by the data acquisition and preprocessing module;

[0018] The model evaluation and feedback unit monitors the performance of the prediction model and evaluates the model through the mean absolute percentage error MAPE and the mean square error MSE;

[0019] The online update and incremental learning unit realizes the online learning and dynamic update of the prediction model according to the evaluation information of the model evaluation and feedback unit, and performs incremental training or correction on the model by using the newly collected data;

[0020] The real-time prediction and decision-making unit obtains the CO 2 prediction result, compares the CO 2 prediction result with the set CO 2 threshold, and generates a corresponding control signal to be sent to the in-vehicle environment control module.

[0021] Preferably, the in-vehicle environment control module includes a ventilation equipment unit, the ventilation equipment unit is connected to the real-time prediction and decision-making unit, the ventilation equipment unit automatically adjusts the opening degree of the air duct valve according to the result of the real-time prediction and decision-making unit to optimize the air quality, and after the regulation is completed, feeds back the regulation information to the data acquisition and preprocessing module.

[0022] Preferably, the real-time prediction and decision-making unit divides the CO 2 prediction result into three states, and controls the opening and closing degree of the air duct valve according to this rule;

[0023] The first state: the predicted CO 2 concentration X i > 5000 ppm, the air duct valve opens at 100% opening;

[0024] The second state is: 1500 ppm ≤ predicted CO 2Concentration X i ≤5000 ppm, the air duct valve is opened at 60% opening;

[0025] The third state is: predicted CO 2 Concentration X i <1500 ppm, the air duct valve maintains normal opening.

[0026] Preferably, the prediction model uses a BP neural network model, and it is necessary to set the hidden layer and neurons of the network structure in advance, select an activation function, reasonably set the learning rate, and perform feature selection.

[0027] The present invention also provides an in-vehicle air quality adjustment method based on machine learning, and the steps include:

[0028] S1. Obtain the CO 2 concentration, the number of people in the train, the train speed, and the tunnel length in real time, and clean and preprocess the data;

[0029] S2. Based on the obtained data, use the BP neural network prediction model to predict the CO 2 concentration;

[0030] S3. Monitor the performance of the trained neural network prediction model, evaluate the model through error analysis indicators. If the mean absolute percentage error MAPE < 10%, predict the CO 2 concentration in the train when passing through the next long tunnel or tunnel group. If the mean absolute percentage error MAPE > 10%, perform incremental training, continuously adjust the hyperparameters according to the training results, and make predictions according to the adjusted neural network prediction model;

[0031] S4. Compare the CO 2 prediction result with the set CO 2 threshold, and generate corresponding control signals according to the preset rules;

[0032] S5. The ventilation equipment unit obtains the control signal in real time, adjusts the CO 2 concentration according to the control signal, optimizes the air quality, and after the regulation is completed, feeds back the regulation information to the data acquisition and preprocessing module.

[0033] Therefore, the present invention adopts the above-mentioned in-vehicle air quality adjustment system and adjustment method based on machine learning, and has the following beneficial effects:

[0034] (1) It can not only monitor the CO 2 concentration in the high-speed train carriage in real time, realize the precise dynamic regulation of the CO 2 concentration, but also combine intelligent prediction to adjust the opening of the air duct valve in advance, optimize the ventilation strategy, and effectively reduce CO2 The risk of concentration accumulation;

[0035] (2) Based on CO 2 Dynamic monitoring and prediction can achieve intelligent optimization of the air quality inside the vehicle compartment, thereby improving the comfort of passengers and the safety of sensitive groups.

[0036] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Description of the Drawings

[0037] Figure 1 is the system module diagram of the embodiment of the present invention;

[0038] Figure 2 is the structural diagram of the BP neural network prediction model of the embodiment of the present invention;

[0039] Figure 3 is the flow chart of adjusting the opening degree of the air duct valve of the embodiment of the present invention;

[0040] Figure 4 is the flow chart of the adjustment method of the embodiment of the present invention;

[0041] Figure 5 is the CO 2 concentration adjustment process display diagram of the embodiment of the present invention. Specific Embodiments

[0042] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Components of the embodiments of the present invention usually described and illustrated herein can be arranged and designed in various different configurations. In the description of the present invention, it should be noted that the orientation or positional relationship indicated by terms such as "upper", "lower", "inner", "outer", etc. is based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship in which the product of the invention is usually placed during use. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as a limitation to the present invention.

[0043] Embodiment

[0044] Referring to Figures 1 - 3 , the present invention provides an in-vehicle air quality adjustment system based on machine learning, including a data acquisition and preprocessing module, an in-vehicle CO 2 concentration real-time prediction module, and an in-vehicle environment control module. The data acquisition and preprocessing module is connected to the in-vehicle CO 2The real-time concentration prediction module is connected to the data acquisition and preprocessing module and the in-vehicle CO 2 The real-time concentration prediction module is also connected to the in-vehicle environment control module.

[0045] The data acquisition and preprocessing module is used to obtain the in-train CO 2 concentration, the number of people in the train, the train speed, and the tunnel length in real time, and clean and preprocess the data.

[0046] In this embodiment, the data acquisition and preprocessing module is connected to the train operation control system, and the train speed data is obtained through the train operation control system. The tunnel length data is calculated by using the track database and the train positioning system, combined with inertial navigation and odometer data.

[0047] The data acquisition and preprocessing module includes a trigger, a sensor unit, a real-time data transmission unit, and a data cleaning and standardization unit. The sensor unit includes a CO 2 dynamic sensor assembly, a speed sensor, and an infrared thermal imaging sensor assembly.

[0048] The trigger controls the start and stop of data acquisition according to the speed sensor signal. The trigger receives the speed sensor signal when the train starts or stops, and uses this signal to generate an output signal as the start or stop signal of the CO 2 sensor and the data acquisition system.

[0049] CO 2 The dynamic sensor assembly is used to detect the in-vehicle CO 2 concentration in real time and transmit it to the data cleaning and standardization unit through the real-time data transmission unit. CO 2 The dynamic sensor uses a T6615CO 2 detection sensor with a detection frequency of 1 / 60 Hz and a detection range of 0 - 10,000 ppm.

[0050] The infrared thermal imaging sensor unit counts the number of passengers in the carriage and transmits it to the data cleaning and standardization unit through the real-time data transmission unit.

[0051] The data cleaning and standardization unit cleans and standardizes the data, and the processed data is automatically transmitted to the in-vehicle CO 2 concentration real-time prediction module.

[0052] In-vehicle CO 2 The concentration real-time prediction module obtains the data from the data acquisition and preprocessing module, predicts the in-vehicle CO 2 concentration when the train passes through the next long tunnel or tunnel group based on the obtained data, and generates corresponding control signals based on the prediction results.

[0053] In this embodiment, the in-vehicle CO2 The real-time concentration prediction module includes a machine learning prediction model unit, a model evaluation and feedback unit, an online update and incremental learning unit, and a real-time prediction and decision-making unit; the machine learning prediction model unit is respectively connected to the online update and incremental learning unit, the model evaluation and feedback unit, and the real-time prediction and decision-making module.

[0054] The machine learning prediction model unit embeds a prediction model based on machine learning. The BP neural network model is adopted to predict the CO concentration inside the train when passing through the next long tunnel or tunnel group in real time according to the data of CO concentration, the number of passengers, the train speed, and the tunnel length collected by the data acquisition and preprocessing module. 2 Concentration, number of passengers, train speed, and tunnel length data are used to predict the CO concentration inside the train when passing through the next long tunnel or tunnel group in real time. Among them, the influence of each parameter on the CO concentration inside the train is mainly reflected in the following aspects: 2 Concentration of 2 is mainly reflected in the following aspects:

[0055] Number of passengers: CO is released during the breathing process of passengers. An increase in the number of passengers leads to an increase in the CO concentration. Train speed: The train speed affects the air exchange efficiency. At a higher speed, the air circulation inside and outside the train speeds up, which helps the diffusion of CO. Tunnel length: When the train runs in the tunnel, the ventilation conditions are limited, and the supply of fresh air from the outside decreases, resulting in an accelerated accumulation rate of CO concentration. This effect is more significant in the environment of extra-long tunnels. 2 , and an increase in the number leads to an increase in the CO 2 concentration. Train speed: The train speed affects the air exchange efficiency. At a higher speed, the air circulation inside and outside the train speeds up, which helps the diffusion of CO. Tunnel length: When the train runs in the tunnel, the ventilation conditions are limited, and the supply of fresh air from the outside decreases, resulting in an accelerated accumulation rate of CO 2 concentration. When the train runs in the tunnel, the ventilation conditions are limited, and the supply of fresh air from the outside decreases, resulting in an accelerated accumulation rate of CO 2 concentration, and this effect is more significant in the environment of extra-long tunnels.

[0056] When building a BP neural network model, it is necessary to set the hidden layer and neurons of the network structure in advance, select an activation function, reasonably set the learning rate, and perform feature selection. Its training process is as Figure 2 shown in the figure. In the figure, both w 1 and w 2 represent weights, indicating the influence degree of the input on the output. The input layer x 1 , x 2 …x d are input variables, where d represents the dimension of the input features; the hidden layer h 1 , h 2 ,…, h n are hidden layer neurons, where n is the number of hidden layer neurons; the output layer part f 1 , f 2 ,…, f c are output layer neurons, where c represents the dimension of the output. f represents the calculation result of the output layer of the neural network, that is, the final predicted value of the prediction model.

[0057] The model evaluation and feedback unit monitors the performance of the prediction model and evaluates the model through the mean absolute percentage error (MAPE) and the mean square error (MSE). If the MAPE of the model is less than 10%, the concentration of CO inside the train passing through the next long tunnel or tunnel group is predicted. If the MAPE is greater than 10%, the online update and incremental learning unit performs online learning and dynamic update, uses the newly collected data to perform incremental training or calibration on the model, and continuously adjusts the hyperparameters (including the learning rate, batch size, regularization coefficient, etc.) and the network structure according to the training results. 2 Specifically, fine-tuning is performed using the newly collected data. The Adam optimizer is selected and a relatively small learning rate (such as 1e-4) is set to avoid drastic changes in the model. During the training process, the mean square error (MSE) is used as the loss function, and 5 rounds of training are performed to avoid overfitting. In each round of training, the gradients are cleared, CO prediction is performed, the loss is calculated, and backpropagation is executed. Finally, the model weights are updated. After the fine-tuning is completed, the optimized model is saved.

[0058] Specifically, fine-tuning is performed using the newly collected data. The Adam optimizer is selected and a relatively small learning rate (such as 1e-4) is set to avoid drastic changes in the model. During the training process, the mean square error (MSE) is used as the loss function, and 5 rounds of training are performed to avoid overfitting. In each round of training, the gradients are cleared, CO 2 prediction is performed, the loss is calculated, and backpropagation is executed. Finally, the model weights are updated. After the fine-tuning is completed, the optimized model is saved.

[0059] The real-time prediction and decision-making unit obtains the CO 2 prediction result, compares the CO 2 prediction result with the set CO 2 threshold, and generates the corresponding control signal to send to the in-vehicle environment control module. The real-time prediction and decision-making unit classifies the CO 2 prediction result into three states and controls the opening and closing degree of the air duct valve according to this rule to optimize the air quality;

[0060] The first state: when the predicted CO 2 concentration X i > 5000 ppm, the air duct valve is opened at 100% opening.

[0061] The second state is: 1500 ppm ≤ predicted CO 2 concentration X i ≤ 5000 ppm, the air duct valve is opened at 60% opening.

[0062] The third state is: when the predicted CO 2 concentration X i < 1500 ppm, the air duct valve maintains the normal opening.

[0063] As Figure 3 shown, the in-vehicle environment control module obtains the in-vehicle CO 2The control signal generated by the real-time prediction module of the concentration automatically adjusts the opening degree of the air duct valve of the air conditioner according to the control signal, and feeds back the regulation information to the data acquisition and preprocessing module. The data acquisition and preprocessing module records the opening and closing degree of the current air duct valve, providing basic data for system optimization. Combining with the CO 2 concentration change trend, analyze the influence of different valve opening degrees on the CO 2 concentration inside the vehicle. Adjust the air duct valve opening strategy through historical data to improve the air circulation efficiency and optimize the air quality.

[0064] In this embodiment, the in-vehicle environment control module includes a ventilation equipment unit. The ventilation equipment unit is connected to the real-time prediction and decision-making unit. The ventilation equipment unit automatically adjusts the opening degree of the air duct valve according to the result of the real-time prediction and decision-making unit to optimize the air quality inside the vehicle. The regulated CO 2 concentration continues to be compared with the set threshold. If the CO 2 concentration after regulation exceeds the threshold, continue to generate corresponding control signals according to the set state rules and execute cyclically until the CO 2 concentration drops within the qualified range (X i <1500 ppm) to optimize the air quality.

[0065] Referring to Figure 4 , the present invention also provides a method for adjusting the air quality inside a vehicle based on machine learning. The steps include:

[0066] S1. The data acquisition and preprocessing module real-time obtains the CO 2 concentration inside the train, the number of people inside the train, the train speed, and the tunnel length, and cleans and preprocesses the data;

[0067] S2. Based on the obtained data, use the machine learning prediction model unit to predict the CO 2 concentration. In this embodiment, the machine learning prediction model unit adopts a BP neural network prediction model. However, in practical applications, the selection of the prediction model should not be limited to this.

[0068] S3. The model evaluation and feedback unit monitors the performance of the trained neural network prediction model, evaluates the model through error analysis indicators. If the mean absolute percentage error MAPE < 10%, predict the CO 2 concentration inside the vehicle when the train passes through the next long tunnel or tunnel group. If the mean absolute percentage error MAPE > 10%, the online update and incremental learning unit performs incremental training on the prediction model, continuously adjusts the hyperparameters according to the training results, and makes predictions according to the adjusted neural network prediction model.

[0069] S4. The real-time prediction and decision-making unit compares the CO 2 prediction result with the set CO2 Compare with the threshold value and generate corresponding control signals according to the preset rules.

[0070] S5. The ventilation equipment unit obtains the control signal in real time and adjusts the CO 2 concentration, optimizes the air quality. After the regulation is completed, feedback the regulation information to the data acquisition and preprocessing module.

[0071] As Figure 5 shown, it is a display diagram of the CO 2 concentration adjustment process, which aims at the CO 2 concentration change in the high-speed train carriage, and the threshold value is set at 2000 ppm.

[0072] In the normal air-conditioning mode, the CO 2 concentration continues to rise after entering the tunnel and exceeds the set threshold value of 2000 ppm after a period of accumulation. In the regulated air supply mode, the in-vehicle CO 2 real-time prediction system makes an early prediction at t = 10 s and triggers the regulation measures to increase the opening degree of the air duct valve.

[0073] After the regulation, the CO 2 concentration drops in the initial stage, then maintains a low level, and finally effectively avoids the CO 2 concentration exceeding the upper limit of 2000 ppm, thus improving the air quality in the vehicle.

[0074] Therefore, the present invention adopts the above-mentioned in-vehicle air quality regulation system and regulation method based on machine learning, uses the machine learning method to predict the CO 2 concentration, adjusts the opening degree of the air duct valve in advance, optimizes the ventilation strategy, effectively reduces the risk of CO 2 concentration accumulation, thereby optimizing the air quality and improving the comfort of passengers and the safety of sensitive people.

[0075] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions of the present invention or make equivalent replacements, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A vehicle air quality control system based on machine learning, characterized in that: It includes a data acquisition and preprocessing module, a real-time prediction module for in-vehicle CO2 concentration and an in-vehicle environment control module, wherein the data acquisition and preprocessing module is connected to the real-time prediction module for in-vehicle CO2 concentration, and both the data acquisition and preprocessing module and the real-time prediction module for in-vehicle CO2 concentration are connected to the in-vehicle environment control module; The data acquisition and preprocessing module is used to obtain the CO2 concentration in the train, the number of people in the train, the train speed and the tunnel length in real time, and clean and preprocess the data; The in-vehicle CO2 concentration real-time prediction module obtains the data from the data acquisition and preprocessing module, predicts the in-vehicle CO2 concentration when the train passes through the next long tunnel or tunnel group according to the acquired data, and generates a corresponding control signal based on the prediction result; The in-vehicle environment control module obtains the control signal generated by the in-vehicle CO2 concentration real-time prediction module, automatically adjusts the opening of the air-conditioning duct valve according to the control signal to optimize the air quality, and feeds back the control information to the data acquisition and preprocessing module.

2. The in-vehicle air quality control system based on machine learning according to claim 1, characterized in that: The data acquisition and preprocessing module includes a trigger, a sensor unit, a real-time data transmission unit and a data cleaning and standardization unit; the sensor unit includes a CO2 dynamic sensor component, a speed sensor and an infrared thermal imaging sensor component; The trigger controls the start and stop of data collection according to the speed sensor signal; The CO2 dynamic sensor assembly is used to detect the CO2 concentration in the vehicle in real time, and transmit it to the data cleaning and standardization unit via the real-time data transmission unit; The infrared thermal imaging sensor unit counts the number of passengers in the carriage and transmits the count to the data cleaning and standardization unit via the real-time data transmission unit; The data cleaning and standardization unit cleans and standardizes the data, and the processed data is automatically transmitted to the in-vehicle CO2 concentration real-time prediction module.

3. The in-vehicle air quality control system based on machine learning according to claim 1, characterized in that: The train speed data is obtained through the train operation control system; the tunnel length data is calculated using the track database and the train positioning system in combination with inertial navigation and odometer data.

4. The in-vehicle air quality control system based on machine learning according to claim 1, characterized in that: The in-vehicle CO2 concentration real-time prediction module includes a machine learning prediction model unit, a model evaluation and feedback unit, an online update and incremental learning unit, and a real-time prediction and decision unit; the machine learning prediction model unit is respectively connected to the online update and incremental learning unit, the model evaluation and feedback unit, and the real-time prediction and decision module; The machine learning prediction model unit is embedded with a prediction model based on machine learning, and predicts the CO2 concentration inside the train when the train passes through the next long tunnel or tunnel group in real time according to the data transmitted by the data acquisition and preprocessing module; The model evaluation and feedback unit monitors the performance of the prediction model and evaluates the model through mean absolute percentage error (MAPE) and mean square error (MSE); The online update and incremental learning unit realizes online learning and dynamic update of the prediction model according to the evaluation information of the model evaluation and feedback unit, and performs incremental training or correction on the model using the newly collected data; The real-time prediction and decision-making unit obtains the CO2 prediction result, compares the CO2 prediction result with the set CO2 threshold, and generates a corresponding control signal to send to the in-vehicle environment control module.

5. The in-vehicle air quality control system based on machine learning according to claim 4, characterized in that: The in-vehicle environment control module includes a ventilation equipment unit, which is connected to the real-time prediction and decision-making unit. The ventilation equipment unit automatically adjusts the opening of the air duct valve according to the results of the real-time prediction and decision-making unit to optimize the air quality. After the control is completed, the control information is fed back to the data acquisition and preprocessing module.

6. The in-vehicle air quality control system based on machine learning according to claim 5, characterized in that: The real-time prediction and decision-making unit divides the CO2 prediction results into three states and controls the opening and closing degree of the air duct valve according to the rules; First state: predict CO2 concentration X i >5000ppm, the air duct valve is opened 100%; The second state is: 1500ppm≤predicted CO2 concentration X i ≤5000ppm, air duct valve is opened to 60%; The third state is: predicted CO2 concentration X i <1500ppm, the air duct valve maintains normal opening.

7. The in-vehicle air quality control system based on machine learning according to claim 4, characterized in that: The prediction model adopts BP neural network model.

8. A method for adjusting the in-vehicle air quality based on machine learning, using the in-vehicle air quality adjustment system based on machine learning according to any one of claims 1 to 7, characterized in that the steps include: S1. Obtain the CO2 concentration in the train, the number of people in the train, the train speed and the tunnel length in real time, and clean and pre-process the data; S2. Based on the acquired data, the BP neural network prediction model is used to predict the CO2 concentration; S3. Monitor the performance of the trained neural network prediction model and evaluate the model through error analysis indicators. If the mean absolute percentage error (MAPE) is less than 10%, predict the CO2 concentration inside the train when it passes through the next long tunnel or tunnel group. If the mean absolute percentage error (MAPE) is greater than 10%, perform incremental training, continuously adjust the hyperparameters according to the training results, and perform predictions based on the adjusted neural network prediction model. S4, comparing the CO2 prediction result with the set CO2 threshold, and generating a corresponding control signal according to a preset rule; S5. The ventilation equipment unit obtains the control signal in real time, adjusts the CO2 concentration according to the control signal, optimizes the air quality, and after the control is completed, feeds back the control information to the data acquisition and preprocessing module.

Images