Intelligent automobile seat system control method based on multi-sensor fusion

Abstract

An intelligent automobile seat system control method based on multi-sensor fusion relates to an intelligent automobile control method, and comprises the following steps: S1, system initialization and identity recognition; s2, data acquisition: a seat pressure sensor acquires weight change information on a seat, a steering wheel fingerprint sensor acquires fingerprint information of a driver, a seat backrest light sensor detects sitting posture information, a heart rate and blood oxygen saturation degree information of the driver is monitored in real time by a heart rate and blood oxygen sensor, and temperature information of each part of the driver is monitored in real time by a temperature sensor; s3, data fusion, processing and analysis: performing data fusion, processing and analysis on the data acquired by each sensor; and S4, the automobile intelligent cabin area controller controls and adjusts the positions of a seat and a steering wheel or carries out fatigue early warning or temperature regulation and control. According to the invention, automatic identification of driver information, automatic adjustment of a sitting posture, real-time monitoring of a health state and personalized control of a seat temperature can be realized, and a more comfortable, safer and intelligent driving environment can be provided for a driver.

Description

Technical Field

[0001] The invention relates to an intelligent automobile control method, in particular to an intelligent automobile seat system control method based on multi-sensor fusion. Background Art

[0002] With the continuous development of intelligent vehicle technology, the humanized design of vehicles and the comfort and safety needs of drivers are increasingly valued. Most traditional car seat systems only provide basic adjustment functions and cannot meet the personalized needs of different drivers for comfort and safety. Therefore, it is particularly important to develop an intelligent car seat system that can automatically identify driver information, adjust seat status and monitor driver health.

[0003] With the development of automobile intelligence, the humanized design of vehicles and the comfort and safety needs of drivers are increasingly valued, and consumers who buy cars have gradually increased their requirements for driving experience. Most traditional car seat systems only provide basic adjustment functions and cannot meet the personalized needs of different drivers for comfort and safety. In the prior art, although some intelligent seat systems can achieve simple position adjustment or temperature control, they lack comprehensive intelligent recognition and adjustment capabilities and cannot meet the personalized needs of different drivers for comfort and safety. The specific shortcomings are as follows:

[0004] 1. Lack of personalized adjustment: Traditional car seat systems often only have basic manual adjustment functions, such as forward and backward movement, seat tilt, etc. These adjustments are usually manual mechanical adjustments or semi-automatic adjustments, and cannot be personalized according to the body shape, driving habits and comfort needs of different drivers.

[0005] 2. Unable to automatically identify the driver: The current seat system does not have the function of automatically identifying the driver. Every time the driver is changed, the seat and steering wheel need to be manually adjusted, which is not only troublesome but also cannot ensure that it is adjusted to the state that best suits the driver.

[0006] 3. Lack of health monitoring function: Traditional seat systems do not have the function of monitoring the driver's physiological state. During long-term driving, the driver may experience fatigue, discomfort or even health problems, and the seat system cannot provide timely feedback or warnings.

[0007] 4. Temperature adjustment is not intelligent: Although some advanced car seats have heating and ventilation functions, these functions usually require manual control by the driver and cannot intelligently adjust the temperature according to the driver's physical needs and external environment.

[0008] 5. Safety needs to be improved: Traditional seat systems that do not integrate biometric technology and real-time health monitoring have limitations in terms of safety. They cannot provide timely safety warnings or take preventive measures when the driver is fatigued or has health problems. Summary of the invention

[0009] The technical problem to be solved by the present invention is to provide an intelligent automobile seat system control method based on multi-sensor fusion to realize automatic recognition of driver information, automatic adjustment of sitting posture, real-time monitoring of health status and personalized control of seat temperature.

[0010] The technical solution to the above technical problem is: a method for controlling an intelligent automobile seat system based on multi-sensor fusion, the method comprising the following steps:

[0011] S1. System initialization and identity identification;

[0012] S2. Data collection, including:

[0013] S21. The seat pressure sensor installed at the bottom of the seat collects the weight change information on the seat, and the seat back light sensor installed on the seat back detects the driver's sitting posture information.

[0014] S22. Multiple temperature sensors distributed on different parts of the seat surface monitor the temperature information of various parts of the driver in real time;

[0015] S23. The heart rate and blood oxygen sensor installed on the steering wheel monitors the driver's heart rate information and blood oxygen saturation information in real time;

[0016] S3. Data fusion, processing and analysis, including:

[0017] S31. The intelligent cockpit domain controller of the car fuses and processes the electrical signals of the seat back light sensor and the seat pressure sensor, and performs driver posture analysis;

[0018] S32. The intelligent cockpit domain controller of the automobile fuses and processes the electrical signals collected by multiple temperature sensors distributed on different parts of the seat surface, and performs optimal seat temperature analysis;

[0019] S33. The intelligent cockpit domain controller of the car fuses and processes the electrical signal data collected by the heart rate and blood oxygen sensor, and performs fatigue assessment and analysis;

[0020] S4. The car's intelligent cockpit domain controller controls the adjustment of seat and steering wheel positions, temperature control, or fatigue warning based on the analysis results.

[0021] A further technical solution of the present invention is: the step S1. system initialization and identity recognition comprises:

[0022] S11. The vehicle is ignited, the intelligent car seat system is started and initialized, the driver's weight is detected through the seat pressure sensor, and the fingerprint data is collected through the steering wheel fingerprint sensor;

[0023] S12. After receiving the weight data and fingerprint data, the car intelligent cockpit domain controller determines whether the weight data, fingerprint data and pre-stored data in the database match; if yes, proceed to step S13, if not, proceed to step S14;

[0024] S13. Load the corresponding driver's personalized seat and steering wheel parameters, and proceed to step S15;

[0025] S14. Display that the driver information cannot be matched, and wait for the fingerprint to be re-entered or the default driver data to be selected, and then proceed to step S15;

[0026] S15. The vehicle intelligent cockpit domain controller controls the execution and adjustment system to adjust the front and rear position, pitch position and steering wheel position of the seat respectively.

[0027] A further technical solution of the present invention is that the step S21 comprises: intercepting the electrical signals of the seat back light sensor and the seat pressure sensor for a preset time length, and obtaining N signal sampling points from a plurality of electrical signal sampling points of the seat back light sensor and one electrical signal sampling point of the seat pressure sensor;

[0028] The N signal sampling points are u(1), u(2), ..., u(N); based on the N signal sampling points, m sampling points are sequentially intercepted with u(1), u(2), ..., u(N-m+1) as starting points to construct N-m+1 m-dimensional vectors; for each of the N-m+1 m-dimensional vectors, the average value of the number of vectors whose distance between the m-dimensional vector and each other vector is less than r is calculated, and the average value of the obtained N-m+1 average values ​​is calculated to obtain a first average value. value; based on the N signal sampling points, respectively, taking u(1), u(2), ..., u(Nm) as starting points, sequentially intercepting m+1 sampling points to construct Nm m+1-dimensional vectors; for each of the Nm m+1-dimensional vectors, calculating the average value of the number of vectors whose distance between the m+1-dimensional vector and each other vector is less than r, and calculating the average value of the Nm average values ​​to obtain a second average value; calculating the value of sample entropy based on the ratio of the first average value to the second average value.

[0029] A further technical solution of the present invention is: the specific content of step S31 includes: obtaining the driver's sitting posture and weight distribution data based on the machine learning model, monitoring the driver's weight distribution and sitting posture in real time, and then using the central processing unit to run the machine learning SVM classifier model to perform machine learning to obtain a segmentation value; judging the driver's sitting posture based on the segmentation value and the sample entropy value generated by the historical driving habit data of each driver in the sample library in the database.

[0030] A further technical solution of the present invention is: in step S22, the plurality of temperature sensors distributed on different parts of the seat surface are distributed according to the corner points of the rectangle and the intersection points of the diagonal lines, and all data affecting the temperature change of each part of the seat are collected in real time;

[0031] The specific contents of step S32 include:

[0032] The temperature data collected by the temperature sensor of the preset time length is used as the input of the neural network model and as the training sample of the neural network model. By collecting multiple sets of training samples, a sample set X is produced. p , where subscript p = 1, 2, 3, ..., Y, using the sample set X p The neural network model is trained to analyze the current temperature status of various parts of the seat; all the collected data that affect the temperature changes of various parts of the seat are input into the trained neural network model, and the neural network model is used to calculate the difference between the current temperature of various parts of the seat and the driver's perceived comfort temperature. The central processing unit outputs the temperature adjustment value based on the current temperature of various parts of the seat.

[0033] A further technical solution of the present invention is: in the neural network model, the number of input layer nodes is 24, the number of hidden layer nodes is 1, and the number of hidden layer nodes is 3, which are respectively the heat transfer Q to the driver's body caused by the temperature difference between the inside and outside of the seat a , Heat transfer to the driver's body caused by the temperature difference between the backrest and the lumbar support b , Temperature change caused by driver's body heat Q c , the number of output layer nodes is 1, which is the comfort of the driver's seat.

[0034] A further technical solution of the present invention is: the use of sample set X p Training a neural network model involves:

[0035] 1) Forward learning;

[0036] ①Start from the input layer of the neural network model to the hidden layer:

[0037] The total input of the jth node in the hidden layer from the input layer is Where i represents the node number of the input layer, x i The specific values ​​of the 24 input parameters corresponding to the input layer in turn, Wij is the weight between the i-th node of the input layer and the j-th node of the hidden layer, and the output of the j-th node of the hidden layer is Vj=f(hj), where the function is the sigmoid function, j represents the hidden layer node number;

[0038] ② Calculate from the hidden layer to the output layer of the neural network model;

[0039] The input of the output layer from the hidden layer is: Substituting Vj into the above formula, the input of the output layer is: Among them, Wik represents the weight between the hidden layer and the output layer, k is the number of nodes in the output layer, that is, k = 1;

[0040] The output of the output layer is:

[0041] 2) Error reverse calculation;

[0042] ①First, define the model error as: δ is the square of the difference between the neural network output and the output layer sample signal, T i is the comfort of the driver's seat in the given learning sample. The error formula can be customized according to the specific characteristics of the model. Yk represents the output of the output layer;

[0043] ② Define the reverse learning efficiency β of the model as β1, β2 and β3, and β3>β2>β1, define three threshold values ​​of the model error δ as A, B, C, and A>B>C, where C≤γ is defined, and γ is the minimum model error allowed after the model training is completed;

[0044] ③ Determine whether δ≥A is true. If so, define the learning efficiency as β1, use β1 to adjust the initial weights Wjk and Wij, and restart the forward learning process. If not, determine whether δ≥B is true. If so, define the learning efficiency as β2, use β2 to adjust the initial weights Wjk and Wij, and restart the forward learning process. If not, determine whether δ≥C is true. If so, define the learning efficiency as β3, use β3 to adjust the initial weights Wjk and Wij, and restart the forward learning process. If not, it is considered that the error meets the requirements, update the weights, stop training, and the model training is completed;

[0045] Among them, the gradient descent method is used to adjust the initial weights Wjk and Wij.

[0046] Represents the negative partial derivative of the error function δ with respect to Wjk;

[0047] Represents the negative partial derivative of the error function δ with respect to Wij;

[0048] 3) Weight update calculation:

[0049] Wlq n =Wjk n-1 +ΔWjk, where Wjk n Indicates the result of this operation of Wjk, Wjk n-1 represents the result of the last operation of Wjk, ΔWjk represents the negative partial derivative of the error function δ with respect to Wjk;

[0050] Wij n =Wij n-1 +ΔWij, where Wij n Indicates the result of Wij's current operation. n-1 represents the result of one operation of Wij, ΔWij represents the negative partial derivative of the error function δ with respect to Wij;

[0051] According to the above logic, the model adjusts the values ​​of weight coefficients Wjk and Wij each time it learns.

[0052] A further technical solution of the present invention is: the step S23 comprises:

[0053] Intercepting the electrical signal of the heart rate and blood oxygen sensor at the driver's steering wheel grip position for a preset time length, and obtaining Z signal sampling points from multiple electrical signal sampling points of the heart rate and blood oxygen sensor at the driver's steering wheel grip position for the preset time length;

[0054] The Z signal sampling points are u(1), u(2), ..., u(Z); based on the Z signal sampling points, m sampling points are sequentially intercepted with u(1), u(2), ..., u(N-m+1) as starting points to construct Z-m+1 m-dimensional vectors; for each of the Z-m+1 m-dimensional vectors, the average value of the number of vectors whose distance between the m-dimensional vector and each other vector is less than r is calculated, and the average value of the obtained Z-m+1 average values ​​is calculated to obtain a first average value. value; based on the Z signal sampling points, respectively, taking u(1), u(2), ..., u(Zm) as starting points, sequentially intercepting m+1 sampling points to construct Zm m+1-dimensional vectors; for each of the Zm m+1-dimensional vectors, calculating the average value of the number of vectors whose distance between the m+1-dimensional vector and each other vector is less than r, and calculating the average value of the Zm average values ​​to obtain a second average value; calculating the value of sample entropy based on the ratio of the first average value to the second average value.

[0055] A further technical solution of the present invention is: the specific contents of step S33 include:

[0056] The driver's heart rate and blood oxygen saturation data are processed by a machine learning model, the driver's heart rate value and blood oxygen saturation are monitored in real time, and then the central processing unit is used to run the machine learning SVM classifier model to obtain a segmentation value through machine learning; according to the segmentation value and the sample entropy value generated by the historical driving habit data of each driver in the sample library in the database, according to the fast and slow heart rate and the high and low changes in blood oxygen saturation fatigue characteristics, the driver's fatigue state is judged, and the driver is reminded in time whether he is driving fatigued to ensure safe driving;

[0057] The fatigue characteristics of the heart rate and the changes in blood oxygen saturation include:

[0058] When the heart rate sensor monitors the driver's hand heart rate, the heart rate value in the fatigue state is lower than the heart rate value when awake;

[0059] When the heart rate sensor monitors the blood oxygen saturation of the driver's hand, the blood oxygen saturation in the fatigue state is lower than the blood oxygen saturation in the awake state;

[0060] The heart rate and blood oxygen sensor is used to collect electrical signals of heart rate and blood oxygen saturation from the back of the steering wheel where both hands usually grip the steering wheel; and is used to obtain the driver's fatigue status from the electrical signals of the heart rate and blood oxygen sensor on the back of the steering wheel.

[0061] A further technical solution of the present invention is: the specific contents of step S4 include:

[0062] S41. Control and adjust the seat and steering wheel position:

[0063] According to the analysis results of the driver's sitting posture, the central processing unit of the car's intelligent cockpit domain controller sends control instructions to control the electric motor and transmission mechanism to automatically adjust the front and back, height, tilt angle of the seat, and the telescopic and tilt angle of the steering wheel;

[0064] S42. Temperature control:

[0065] S421. According to the optimal seat temperature analysis results, if the temperature distribution characteristics match the appropriate temperature range of each part of the seat in the database, then maintain the status quo, if not, proceed to step S422;

[0066] S422. Cool or heat the seat parts that do not match the temperature range;

[0067] S43. Fatigue warning:

[0068] S431. According to the fatigue assessment result, if it is fatigue, proceed to step S432, if not, no prompt;

[0069] S432. Preliminary reminder: slight vibration of the seat or steering wheel;

[0070] S433. Determine again whether you are still tired, if yes, proceed to step S434, if no, no prompt;

[0071] S434. Issue a voice warning through the car intercom, suggest taking a break and automatically plan a route to the nearest service area.

[0072] Due to the adoption of the above structure, the intelligent automobile seat system control method based on multi-sensor fusion of the present invention has the following beneficial effects compared with the prior art:

[0073] 1. Personalized adjustment of sitting posture is possible

[0074] The present invention utilizes a seat pressure sensor at the bottom of the seat and a seat back light sensor at the seat back, and adopts intelligent algorithms such as machine learning to detect the driver's weight distribution and sitting posture in real time. It then uses the vehicle's intelligent cockpit domain controller to run the algorithm, combined with each driver's historical driving habit data, to achieve automatic adjustment of the seat and steering wheel, provide the most optimized driving posture and control interface to meet the personalized needs of different drivers, and improve driving comfort and pleasure.

[0075] 2. Automatic recognition of driver information is possible

[0076] The present invention uses the steering wheel fingerprint sensor on the steering wheel combined with the seat pressure sensor at the bottom of the car seat, and compares it with the fingerprint, weight and other data stored in the system to automatically identify the driver's identity. Once the identification is successful, the system will automatically call the driver's personalized settings, greatly improving the convenience of use.

[0077] 3. Real-time monitoring of the driver's health status can be achieved to enhance driving safety

[0078] Long-term driving may cause fatigue and health problems for the driver, which cannot be monitored or reminded by traditional seat systems. The present invention installs a heart rate and blood oxygen sensor on the steering wheel, which continuously monitors the driver's heart rate, blood oxygen saturation and other physiological indicators by contacting the driver's palm. The car's intelligent cockpit domain controller analyzes this data in real time, and roughly estimates the driver's condition through the fluctuation of the seat pressure sensor, and judges the driver's health status in combination with the algorithm. When abnormalities or signs of fatigue are detected, the early warning system reminds the driver to rest, which can effectively avoid traffic accidents caused by fatigue driving, thereby effectively reducing the safety risks caused by fatigue driving and enhancing driving safety.

[0079] 4. Personalized control of seat temperature is possible

[0080] The present invention uses temperature sensors distributed throughout the seat to monitor the temperature of the human body in various parts of the seat in real time, and feeds back the data to the car's intelligent cockpit domain controller. The car's intelligent cockpit domain controller combines with a neural network algorithm to use semiconductor refrigeration plates distributed throughout the seat for precise local cooling or electric heating modules for local heating, thereby realizing automatic adjustment of the seat temperature, ensuring that the driver can enjoy a suitable seat temperature in any environment and improving driving comfort.

[0081] Below, the technical features of the intelligent automobile seat system control method based on multi-sensor fusion of the present invention are further described in conjunction with the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 : A flowchart of step S1 in Embodiment 1,

[0083] Figure 2 : A flowchart of step S2 in Embodiment 1,

[0084] Figure 3 : A flowchart of temperature control in step S42 of the first embodiment,

[0085] Figure 4 : A flowchart of fatigue warning in step S43 of the first embodiment,

[0086] Figure 5 : In the first embodiment, a top view of the distribution of the temperature sensor, the heating wire and the semiconductor cooling sheet of the driver's seat cushion,

[0087] Figure 6 : In the first embodiment, the front view of the distribution of the temperature sensors, heating wires and semiconductor cooling sheets on the driver's seat back and the lumbar support part,

[0088] exist Figure 5 , Figure 6 Among them, 1-temperature sensor, 2-semiconductor cooling sheet, 3-heating wire. DETAILED DESCRIPTION

[0089] Embodiment 1:

[0090] A method for controlling an intelligent automobile seat system based on multi-sensor fusion, the method comprising the following steps:

[0091] S1. System initialization and identity identification;

[0092] S2. Data collection, including:

[0093] S21. The seat pressure sensor installed at the bottom of the seat collects the weight change information on the seat, and the seat back light sensor installed on the seat back detects the driver's sitting posture information.

[0094] S22. Multiple temperature sensors distributed on different parts of the seat surface monitor the temperature information of various parts of the driver in real time;

[0095] S23. The heart rate and blood oxygen sensor installed on the steering wheel monitors the driver's heart rate information and blood oxygen saturation information in real time;

[0096] S3. Data fusion, processing and analysis, including:

[0097] S31. The intelligent cockpit domain controller of the car fuses and processes the electrical signals of the seat back light sensor and the seat pressure sensor, and performs driver posture analysis;

[0098] S32. The intelligent cockpit domain controller of the automobile fuses and processes the electrical signals collected by multiple temperature sensors distributed on different parts of the seat surface, and performs optimal seat temperature analysis;

[0099] S33. The intelligent cockpit domain controller of the car fuses and processes the electrical signal data collected by the heart rate and blood oxygen sensor, and performs fatigue assessment and analysis;

[0100] S4. The car intelligent cockpit domain controller controls the adjustment of the seat and steering wheel position, temperature control, or fatigue warning according to the analysis results;

[0101] S5. Feedback and continuous optimization.

[0102] The step S1. system initialization and identity recognition includes:

[0103] S11. The vehicle is ignited, the intelligent car seat system is started and initialized, the driver's weight is detected through the seat pressure sensor, and the fingerprint data is collected through the steering wheel fingerprint sensor;

[0104] S12. After receiving the weight data and fingerprint data, the car intelligent cockpit domain controller determines whether the weight data, fingerprint data and pre-stored data in the database match; if yes, proceed to step S13, if not, proceed to step S14;

[0105] S13. Load the corresponding driver's personalized seat and steering wheel parameters, and proceed to step S15;

[0106] S14. Display that the driver information cannot be matched, and wait for the fingerprint to be re-entered or the default driver data to be selected, and then proceed to step S15;

[0107] S15. The vehicle intelligent cockpit domain controller controls the execution and adjustment system to adjust the front and rear position, pitch position and steering wheel position of the seat respectively.

[0108] The step S21 includes: intercepting the electrical signals of the seat back light sensor and the seat pressure sensor for a preset time length, and obtaining N signal sampling points from a plurality of electrical signal sampling points of the seat back light sensor and one electrical signal sampling point of the seat pressure sensor;

[0109] The N signal sampling points are u(1), u(2), ..., u(N); based on the N signal sampling points, m sampling points are sequentially intercepted with u(1), u(2), ..., u(N-m+1) as starting points to construct N-m+1 m-dimensional vectors; for each of the N-m+1 m-dimensional vectors, the average value of the number of vectors whose distance between the m-dimensional vector and each other vector is less than r is calculated, and the average value of the obtained N-m+1 average values ​​is calculated to obtain a first average value. value; based on the N signal sampling points, respectively, taking u(1), u(2), ..., u(Nm) as starting points, sequentially intercepting m+1 sampling points to construct Nm m+1-dimensional vectors; for each of the Nm m+1-dimensional vectors, calculating the average value of the number of vectors whose distance between the m+1-dimensional vector and each other vector is less than r, and calculating the average value of the Nm average values ​​to obtain a second average value; calculating the value of sample entropy based on the ratio of the first average value to the second average value.

[0110] The specific contents of step S31 include: obtaining the driver's sitting posture and weight distribution data based on a machine learning model, monitoring the driver's weight distribution and sitting posture in real time, and then using the central processing unit to run the machine learning SVM classifier model to perform machine learning to obtain a segmentation value; judging the driver's sitting posture based on the segmentation value and the sample entropy value generated by the historical driving habit data of each driver in the sample library in the database.

[0111] In step S22, the plurality of temperature sensors distributed on different parts of the seat surface are distributed according to the corner points of the rectangle and the intersection points of the diagonal lines, and all data affecting the temperature changes of various parts of the seat are collected in real time.

[0112] The specific contents of step S32 include:

[0113] The temperature data collected by the temperature sensor of the preset time length is used as the input of the neural network model and as the training sample of the neural network model. By collecting multiple sets of training samples, a sample set X is produced. p , where subscript p = 1, 2, 3, ..., Y, using the sample set X pThe neural network model is trained to analyze the current temperature status of various parts of the seat; all the collected data that affect the temperature changes of various parts of the seat are input into the trained neural network model, and the neural network model is used to calculate the difference between the current temperature of various parts of the seat and the driver's perceived comfort temperature. The central processing unit outputs the temperature adjustment value based on the current temperature of various parts of the seat.

[0114] The factors that affect the temperature change of each part of the seat include: the temperature of each part of the seat T 01 、T 02 、T 03 、T 04 、T 05 , Temperature of each part of the seat back and lumbar support T 11 、T 12 、T 13 、T 14 、T 15 And the current intensity of the heating wire of the seat electric heating module I hc1 ,I hc2 , Seat semiconductor cooling chip current intensity I cc1 ,I cc2 ,I cc3 ,I cc4 , backrest and lumbar support electric heating module heating wire current intensity I hb1 ,I hb2 、Seat semiconductor cooling chip current I cb1 ,I cb2 ,I cb3 ,I cb4 ,I cb5 ,I cb6 Therefore, in the neural network model, the number of input layer nodes is 24, the number of hidden layer nodes is 1, and the number of hidden layer nodes is 3, which are respectively the heat transfer Q to the driver's body caused by the temperature difference between the inside and outside of the seat a , Heat transfer to the driver's body caused by the temperature difference between the backrest and the lumbar support b , Temperature change caused by driver's body heat Q c , the number of output layer nodes is 1, which is the comfort of the driver's seat.

[0115] The use of sample set X p Training a neural network model involves:

[0116] 1) Forward learning;

[0117] ③Start from the input layer of the neural network model to calculate the hidden layer:

[0118] The total input of the jth node in the hidden layer from the input layer is Where i represents the node number of the input layer, xi The specific values ​​of the 24 input parameters corresponding to the input layer are Wij, the weight between the i-th node of the input layer and the j-th node of the hidden layer, and the output of the j-th node of the hidden layer is Vj = f(hj), where the function is the sigmoid function, j represents the hidden layer node number;

[0119] ④ Calculate from the hidden layer to the output layer of the neural network model;

[0120] The input of the output layer from the hidden layer is: Substituting Vj into the above formula, the input of the output layer is: Among them, Wik represents the weight between the hidden layer and the output layer, k is the number of nodes in the output layer, that is, k = 1;

[0121] The output of the output layer is:

[0122] 2) Error reverse calculation;

[0123] ①First, define the model error as: δ is the square of the difference between the neural network output and the output layer sample signal, T i is the comfort of the driver's seat in the given learning sample. The error formula can be customized according to the specific characteristics of the model. Yk represents the output of the output layer;

[0124] ② Define the reverse learning efficiency β of the model as β1, β2 and β3, and β3>β2>β1, define three threshold values ​​of the model error δ as A, B, C, and A>B>C, where C≤γ is defined, and γ is the minimum model error allowed after the model training is completed;

[0125] ③ Determine whether δ≥A is true. If so, define the learning efficiency as β1, use β1 to adjust the initial weights Wjk and Wij, and restart the forward learning process. If not, determine whether δ≥B is true. If so, define the learning efficiency as β2, use β2 to adjust the initial weights Wjk and Wij, and restart the forward learning process. If not, determine whether δ≥C is true. If so, define the learning efficiency as β3, use β3 to adjust the initial weights Wjk and Wij, and restart the forward learning process. If not, it is considered that the error meets the requirements, update the weights, stop training, and the model training is completed;

[0126] Among them, the gradient descent method is used to adjust the initial weights Wjk and Wij.

[0127] Represents the negative partial derivative of the error function δ with respect to Wjk;

[0128] Represents the negative partial derivative of the error function δ with respect to Wij;

[0129] 3) Weight update calculation:

[0130] Wlq n =Wjk n-1 +ΔWjk, where Wjk n Indicates the result of this operation of Wjk, Wjk n-1 represents the result of the last operation of Wjk, ΔWjk represents the negative partial derivative of the error function δ with respect to Wjk;

[0131] Wij n =Wij n-1 +ΔWij, where Wij n Indicates the result of Wij's current operation. n-1 represents the result of one operation of Wij, ΔWij represents the negative partial derivative of the error function δ with respect to Wij;

[0132] According to the above logic, the model adjusts the values ​​of weight coefficients Wjk and Wij each time it learns.

[0133] The step S23 comprises:

[0134] Intercepting the electrical signal of the heart rate and blood oxygen sensor at the driver's steering wheel grip position for a preset time length, and obtaining Z signal sampling points from multiple electrical signal sampling points of the heart rate and blood oxygen sensor at the driver's steering wheel grip position for the preset time length;

[0135] The Z signal sampling points are u(1), u(2), ..., u(Z); based on the Z signal sampling points, m sampling points are sequentially intercepted with u(1), u(2), ..., u(N-m+1) as starting points to construct Z-m+1 m-dimensional vectors; for each of the Z-m+1 m-dimensional vectors, the average value of the number of vectors whose distance between the m-dimensional vector and each other vector is less than r is calculated, and the average value of the obtained Z-m+1 average values ​​is calculated to obtain a first average value. value; based on the Z signal sampling points, respectively, taking u(1), u(2), ..., u(Zm) as starting points, sequentially intercepting m+1 sampling points to construct Zm m+1-dimensional vectors; for each of the Zm m+1-dimensional vectors, calculating the average value of the number of vectors whose distance between the m+1-dimensional vector and each other vector is less than r, and calculating the average value of the Zm average values ​​to obtain a second average value; calculating the value of sample entropy based on the ratio of the first average value to the second average value.

[0136] The specific contents of step S33 include:

[0137] The driver's heart rate and blood oxygen saturation data are processed based on a machine learning model, the driver's heart rate value and blood oxygen saturation are monitored in real time, and then the central processing unit is used to run the machine learning SVM classifier model to obtain a segmentation value through machine learning; according to the segmentation value and the sample entropy value generated by the historical driving habit data of each driver in the sample library in the database, according to the fatigue characteristics of the fast and slow heart rate and the high and low changes in blood oxygen saturation, the driver's fatigue state is judged, and the driver is promptly reminded whether he is driving fatigued to ensure safe driving.

[0138] The fatigue characteristics of the heart rate and the changes in blood oxygen saturation include:

[0139] When the heart rate sensor monitors the driver's hand heart rate, the heart rate value in the fatigue state is lower than the heart rate value when awake;

[0140] When the heart rate sensor monitors the blood oxygen saturation of the driver's hand, the blood oxygen saturation in the fatigue state is lower than the blood oxygen saturation in the awake state;

[0141] The heart rate and blood oxygen sensor is used to collect electrical signals of heart rate and blood oxygen saturation from the back of the steering wheel where both hands usually grip the steering wheel; and is used to obtain the driver's fatigue status from the electrical signals of the heart rate and blood oxygen sensor on the back of the steering wheel.

[0142] The specific contents of step S4 include:

[0143] S41. Control and adjust the seat and steering wheel position:

[0144] According to the analysis results of the driver's sitting posture, the central processing unit of the car's intelligent cockpit domain controller sends control instructions to control the electric motor and transmission mechanism to automatically adjust the front and back, height, tilt angle of the seat, and the telescopic and tilt angle of the steering wheel;

[0145] S42. Temperature control:

[0146] S421. According to the optimal seat temperature analysis results, if the temperature distribution characteristics match the appropriate temperature range of each part of the seat in the database, then maintain the status quo, if not, proceed to step S422;

[0147] S422. For the seat parts that do not match the temperature range, the semiconductor cooling sheet is used for cooling, or the electric heating module is used for heating, so as to adjust the temperature of each part of the driver's seat;

[0148] In different seasons, based on the normal human body surface temperature of 33-35 degrees Celsius, the initial set temperature of each part of the seat is 32 degrees Celsius. After reaching the initial set temperature through cooling by the semiconductor refrigeration sheet or heating by the electric heating module, the initial set temperature of each part in different seasons is adjusted through neural network learning according to the driver's feedback on the need for cooling or heating of each part during daily use of the car, so as to achieve the most comfortable temperature for each part under different ambient temperatures.

[0149] S43. Fatigue warning:

[0150] S431. According to the fatigue assessment result, if it is fatigue, proceed to step S432, if not, no prompt;

[0151] S432. Preliminary reminder: slight vibration of the seat or steering wheel;

[0152] S433. Determine again whether you are still tired, if yes, proceed to step S434, if no, no prompt;

[0153] S434. Issue a voice warning through the car intercom, suggest taking a break and automatically plan a route to the nearest service area.

[0154] The step S5. Feedback and continuous optimization includes:

[0155] S51. User feedback collection: Collect user feedback on seat comfort and fatigue warning accuracy through vehicle system or APP; continuously adjust seat position and warning accuracy through machine learning

[0156] S52. Data analysis and optimization: Based on user feedback and system operation data, continuously optimize algorithms and parameter settings to improve system performance.

[0157] In the present invention, the intelligent car seat system includes a plurality of sensors, a car intelligent cockpit domain controller, an execution and adjustment mechanism, a user interface and a feedback system, wherein:

[0158] The multiple sensors include:

[0159] The seat pressure sensor is a high-precision pressure sensor placed at the bottom of the seat. It is used to accurately measure changes in the driver's weight and transmit the data to the car's intelligent cockpit domain controller via a data cable.

[0160] The steering wheel fingerprint sensor uses a capacitive or optical fingerprint sensor and is integrated into the commonly used holding position of the steering wheel to ensure that the driver can naturally contact and complete fingerprint collection when holding the steering wheel. The steering wheel fingerprint sensor transmits data to the car's smart cockpit domain controller via a data cable.

[0161] The seat back light sensor is an infrared sensor installed on the seat back to detect the position and posture of the driver's back and transmit the data to the car's intelligent cockpit domain controller via a data cable.

[0162] The heart rate and blood oxygen sensor uses a non-invasive bioelectric sensor installed on the steering wheel to monitor the driver's heart rate and blood oxygen saturation in real time, and transmit the data to the car's intelligent cockpit domain controller via a data cable;

[0163] Temperature sensors distributed throughout the seat surface are used to monitor the temperature of various parts of the driver in real time and transmit the data to the car's intelligent cockpit domain controller via data cables.

[0164] The automobile intelligent cockpit domain controller comprises:

[0165] Central Processing Unit: Responsible for receiving and processing data from various sensors, running algorithms to make decisions, and controlling the execution and adjustment mechanisms to make corresponding adjustments.

[0166] Data storage unit: used to store the driver's personal information, historical driving habits data and historical sensor readings for personalized settings and historical data analysis.

[0167] Communication module: realizes data interaction with other vehicle systems, such as receiving service area information from the navigation system to provide navigation guidance when a rest is needed.

[0168] The execution and adjustment mechanism includes:

[0169] Seat and steering wheel adjustment mechanism: uses electric motor and transmission mechanism to automatically adjust the front and rear position, height, tilt angle of the seat and the telescopic and tilt angle of the steering wheel according to control instructions.

[0170] Semiconductor Refrigeration Chip: The seat has multiple semiconductor refrigeration chips built in. The seat temperature is adjusted by adjusting the working state of the refrigeration chip according to the temperature data fed back by the temperature sensor.

[0171] Electric heating module: The seat has multiple electric heaters built in. The seat temperature is adjusted by adjusting the working state of the electric heating module based on the temperature data fed back by the temperature sensor.

[0172] The user interface and feedback system includes:

[0173] Display screen or voice prompt module: used to display system status, reminder information and operation suggestions to the driver, such as fatigue driving reminder, temperature adjustment reminder, etc.

[0174] Feedback module: Provides immediate feedback to the driver through vibration, sound or visual prompts to ensure that the driver pays attention to the system's suggestions and reminders.

Claims

1. A control method for an intelligent automobile seat system based on multi-sensor fusion, characterized in that: The method comprises the following steps: S1. System initialization and identity identification; S2. Data collection, including: S21. The seat pressure sensor installed at the bottom of the seat collects the weight change information on the seat, and the seat back light sensor installed on the seat back detects the driver's sitting posture information. S22. Multiple temperature sensors distributed on different parts of the seat surface monitor the temperature information of various parts of the driver in real time; S23. The heart rate and blood oxygen sensor installed on the steering wheel monitors the driver's heart rate information and blood oxygen saturation information in real time; S3. Data fusion, processing and analysis, including: S31. The intelligent cockpit domain controller of the car fuses and processes the electrical signals of the seat back light sensor and the seat pressure sensor, and performs driver posture analysis; S32. The intelligent cockpit domain controller of the automobile fuses and processes the electrical signals collected by multiple temperature sensors distributed on different parts of the seat surface, and performs optimal seat temperature analysis; S33. The intelligent cockpit domain controller of the car fuses and processes the electrical signal data collected by the heart rate and blood oxygen sensor, and performs fatigue assessment and analysis; S4. The car's intelligent cockpit domain controller controls the adjustment of seat and steering wheel positions, temperature control, or fatigue warning based on the analysis results.

2. The intelligent automobile seat system control method based on multi-sensor fusion according to claim 1 is characterized in that: The step S1. system initialization and identity recognition includes: S11. The vehicle is ignited, the intelligent car seat system is started and initialized, the driver's weight is detected through the seat pressure sensor, and the fingerprint data is collected through the steering wheel fingerprint sensor; S12. After receiving the weight data and fingerprint data, the car intelligent cockpit domain controller determines whether the weight data, fingerprint data and pre-stored data in the database match; if yes, proceed to step S13, if not, proceed to step S14; S13. Load the corresponding driver's personalized seat and steering wheel parameters, and proceed to step S15; S14. Display that the driver information cannot be matched, and wait for the fingerprint to be re-entered or the default driver data to be selected, and then proceed to step S15; S15. The vehicle intelligent cockpit domain controller controls the execution and adjustment system to adjust the front and rear position, pitch position and steering wheel position of the seat respectively.

3. The intelligent automobile seat system control method based on multi-sensor fusion according to claim 1 is characterized in that: The step S21 comprises: intercepting electrical signals of the seat back light sensor and the seat pressure sensor for a preset time length, and obtaining N signal sampling points from a plurality of electrical signal sampling points of the seat back light sensor and one electrical signal sampling point of the seat pressure sensor; The N signal sampling points are u(1), u(2), ..., u(N); based on the N signal sampling points, m sampling points are sequentially intercepted with u(1), u(2), ..., u(N-m+1) as starting points to construct N-m+1 m-dimensional vectors; for each of the N-m+1 m-dimensional vectors, the average value of the number of vectors whose distance between the m-dimensional vector and each other vector is less than r is calculated, and the average value of the obtained N-m+1 average values ​​is calculated to obtain a first average value. value; based on the N signal sampling points, respectively, taking u(1), u(2), ..., u(Nm) as starting points, sequentially intercepting m+1 sampling points to construct Nm m+1-dimensional vectors; for each of the Nm m+1-dimensional vectors, calculating the average value of the number of vectors whose distance between the m+1-dimensional vector and each other vector is less than r, and calculating the average value of the Nm average values ​​to obtain a second average value; calculating the value of sample entropy based on the ratio of the first average value to the second average value.

4. The intelligent automobile seat system control method based on multi-sensor fusion according to claim 3 is characterized in that: The specific contents of step S31 include: obtaining the driver's sitting posture and weight distribution data based on a machine learning model, monitoring the driver's weight distribution and sitting posture in real time, and then using the central processing unit to run the machine learning SVM classifier model to perform machine learning to obtain a segmentation value; judging the driver's sitting posture based on the segmentation value and the sample entropy value generated by the historical driving habit data of each driver in the sample library in the database.

5. The intelligent automobile seat system control method based on multi-sensor fusion according to claim 1 is characterized in that: In step S22, the plurality of temperature sensors distributed on different parts of the seat surface are distributed according to the corner points of the rectangle and the intersection points of the diagonal lines, and all data affecting the temperature change of each part of the seat are collected in real time; The specific contents of step S32 include: The temperature data collected by the temperature sensor of the preset time length is used as the input of the neural network model and as the training sample of the neural network model. By collecting multiple sets of training samples, a sample set X is produced. p , where subscript p = 1, 2, 3, ..., Y, using the sample set X p The neural network model is trained to analyze the current temperature status of various parts of the seat; all the collected data that affect the temperature changes of various parts of the seat are input into the trained neural network model, and the neural network model is used to calculate the difference between the current temperature of various parts of the seat and the driver's perceived comfort temperature. The central processing unit outputs the temperature adjustment value based on the current temperature of various parts of the seat.

6. The intelligent automobile seat system control method based on multi-sensor fusion according to claim 5 is characterized in that: In the neural network model, the number of input layer nodes is 24, the number of hidden layer nodes is 1, and the number of hidden layer nodes is 3, which are respectively the heat transfer Q to the driver's body caused by the temperature difference between the inside and outside of the seat a , Heat transfer to the driver's body caused by the temperature difference between the backrest and the lumbar support b , Temperature change caused by driver's body heat Q c , the number of output layer nodes is 1, which is the comfort of the driver's seat.

7. The intelligent automobile seat system control method based on multi-sensor fusion according to claim 6 is characterized in that: The use of sample set X p Training a neural network model involves: 1) Forward learning; ①Start from the input layer of the neural network model to the hidden layer: The total input of the jth node in the hidden layer from the input layer is Where i represents the node number of the input layer, x i The specific values ​​of the 24 input parameters corresponding to the input layer in turn, Wij is the weight between the i-th node of the input layer and the j-th node of the hidden layer, and the output of the j-th node of the hidden layer is Vj=f(hj), where the function is the sigmoid function, j represents the hidden layer node number; ② Calculate from the hidden layer to the output layer of the neural network model; The input of the output layer from the hidden layer is: Substituting Vj into the above formula, the input of the output layer is: Among them, Wik represents the weight between the hidden layer and the output layer, k is the number of nodes in the output layer, that is, k = 1; The output of the output layer is: 2) Error reverse calculation; ①First, define the model error as: δ is the square of the difference between the neural network output and the output layer sample signal, T i is the comfort of the driver's seat in the given learning sample. The error formula can be customized according to the specific characteristics of the model. Yk represents the output of the output layer; ② Define the reverse learning efficiency β of the model as β1, β2 and β3, and β3>β2>β1, define three threshold values ​​of the model error δ as A, B, C, and A>B>C, where C≤γ is defined, and γ is the minimum model error allowed after the model training is completed; ③ Determine whether δ≥A is true. If so, define the learning efficiency as β1, use β1 to adjust the initial weights Wjk and Wij, and restart the forward learning process. If not, determine whether δ≥B is true. If so, define the learning efficiency as β2, use β2 to adjust the initial weights Wjk and Wij, and restart the forward learning process. If not, determine whether δ≥C is true. If so, define the learning efficiency as β3, use β3 to adjust the initial weights Wjk and Wij, and restart the forward learning process. If not, it is considered that the error meets the requirements, update the weights, stop training, and the model training is completed; Among them, the gradient descent method is used to adjust the initial weights Wjk and Wij. Represents the negative partial derivative of the error function δ with respect to Wjk; Represents the negative partial derivative of the error function δ with respect to Wij; 3) Weight update calculation: Wlq n =Wjk n-1 +ΔWjk, where Wjk n Indicates the result of this operation of Wjk, Wjk n-1 represents the result of the last operation of Wjk, ΔWjk represents the negative partial derivative of the error function δ with respect to Wjk; Wij n =Wij n-1 +ΔWij, where Wij n Indicates the result of Wij's current operation. n-1 represents the result of one operation of Wij, ΔWij represents the negative partial derivative of the error function δ with respect to Wij; According to the above logic, the model adjusts the values ​​of weight coefficients Wjk and Wij each time it learns.

8. The intelligent automobile seat system control method based on multi-sensor fusion according to claim 1 is characterized in that: The step S23 comprises: Intercepting the electrical signal of the heart rate and blood oxygen sensor at the driver's steering wheel grip position for a preset time length, and obtaining Z signal sampling points from multiple electrical signal sampling points of the heart rate and blood oxygen sensor at the driver's steering wheel grip position for the preset time length; The Z signal sampling points are u(1), u(2), ..., u(Z); based on the Z signal sampling points, m sampling points are sequentially intercepted with u(1), u(2), ..., u(N-m+1) as starting points to construct Z-m+1 m-dimensional vectors; for each of the Z-m+1 m-dimensional vectors, the average value of the number of vectors whose distance between the m-dimensional vector and each other vector is less than r is calculated, and the average value of the obtained Z-m+1 average values ​​is calculated to obtain a first average value. value; based on the Z signal sampling points, respectively, taking u(1), u(2), ..., u(Zm) as starting points, sequentially intercepting m+1 sampling points to construct Zm m+1-dimensional vectors; for each of the Zm m+1-dimensional vectors, calculating the average value of the number of vectors whose distance between the m+1-dimensional vector and each other vector is less than r, and calculating the average value of the Zm average values ​​to obtain a second average value; calculating the value of sample entropy based on the ratio of the first average value to the second average value.

9. The intelligent automobile seat system control method based on multi-sensor fusion according to claim 8 is characterized in that: The specific contents of step S33 include: The driver's heart rate and blood oxygen saturation data are processed by a machine learning model, the driver's heart rate value and blood oxygen saturation are monitored in real time, and then the central processing unit is used to run the machine learning SVM classifier model to obtain a segmentation value through machine learning; according to the segmentation value and the sample entropy value generated by the historical driving habit data of each driver in the sample library in the database, according to the fast and slow heart rate and the high and low changes in blood oxygen saturation fatigue characteristics, the driver's fatigue state is judged, and the driver is reminded in time whether he is driving fatigued to ensure safe driving; The fatigue characteristics of the heart rate and the changes in blood oxygen saturation include: When the heart rate sensor monitors the driver's hand heart rate, the heart rate value in the fatigue state is lower than the heart rate value when awake; When the heart rate sensor monitors the blood oxygen saturation of the driver's hand, the blood oxygen saturation in the fatigue state is lower than the blood oxygen saturation in the awake state; The heart rate and blood oxygen sensor is used to collect electrical signals of heart rate and blood oxygen saturation from the back of the steering wheel where both hands usually grip the steering wheel; and is used to obtain the driver's fatigue status from the electrical signals of the heart rate and blood oxygen sensor on the back of the steering wheel.

10. The intelligent automobile seat system control method based on multi-sensor fusion according to claim 9 is characterized in that: The specific contents of step S4 include: S41. Control and adjust the seat and steering wheel position: According to the analysis results of the driver's sitting posture, the central processing unit of the car's intelligent cockpit domain controller sends control instructions to control the electric motor and transmission mechanism to automatically adjust the front and back, height, tilt angle of the seat, and the telescopic and tilt angle of the steering wheel; S42. Temperature control: S421. According to the optimal seat temperature analysis results, if the temperature distribution characteristics match the appropriate temperature range of each part of the seat in the database, then maintain the status quo, if not, proceed to step S422; S422. Cool or heat the seat parts that do not match the temperature range; S43. Fatigue warning: S431. According to the fatigue assessment result, if it is fatigue, proceed to step S432, if not, no prompt; S432. Preliminary reminder: slight vibration of the seat or steering wheel; S433. Determine again whether you are still tired, if yes, proceed to step S434, if no, no prompt; S434. Issue a voice warning through the car intercom, suggest taking a break and automatically plan a route to the nearest service area.

Images