Vehicle speed deviation analysis method based on brain-like cognitive calculation model

Abstract

The invention discloses a vehicle speed deviation analysis method based on a brain-like cognitive calculation model, and belongs to the technical field of vehicle speed deviation analysis, and the method comprises the following steps: S1, building a low-grade road simulation environment to carry out a driving simulation experiment; s2, establishing a visual road environment model; s3, preprocessing the electroencephalogram data of the driver, performing spectral analysis, and establishing a relationship between the environment and the electroencephalogram signal intensity on the basis; and S4, based on a QN-MHP type brain cognitive calculation model, establishing an LMM model by using the signal intensity relationship of each brain region, and analyzing the deviation between the expected vehicle speed of the driver and the actual vehicle speed according to the LMM model. According to the vehicle speed deviation analysis method based on the brain-like cognitive calculation model, the perception condition of a driver on the road environment condition is quantified, and the method plays an important role in analyzing the perception and cognitive characteristics when the expected vehicle speed of the driver is determined and the influence on the difference between the expected vehicle speed and the actual running vehicle speed; the method is of great significance in improving road traffic safety.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle speed deviation analysis, and in particular to a vehicle speed deviation analysis method based on a brain-like cognitive computing model. Background Art

[0002] Low-grade highway transportation plays a decisive role in transportation in mountainous and rural areas. However, low-grade traffic accidents occur frequently in my country, with a high proportion of serious and major accidents. The number and severity of accidents are still on the rise. Due to the low design index requirements and complex driving environment, when driving on low-grade highways, there is often a large deviation between the driver's expected speed and the actual speed, which seriously affects driving safety. In this process, the visual road environment perceived by the driver will affect the perception of the expected speed, thereby affecting the decision of the actual speed.

[0003] Vehicle speed is a key factor affecting the occurrence and consequences of road traffic accidents, and the driver's expected speed determines his speed decision-making behavior. The expected speed is the running speed selected by the driver in a free flow state without linear constraints; or the highest "safe" speed that the driver hopes to achieve when the vehicle is not or is basically not constrained by other vehicles. When the driver misjudges the safe running speed allowed by the road environment, he will subjectively generate an "optimal running speed" that does not meet the road conditions, that is, control the speed with an inappropriate expected speed. This speed is often higher than the design speed of the road, which is prone to safety hazards and even traffic accidents.

[0004] Due to various factors such as the road environment, the driver's expected speed in a certain road environment will deviate from the actual operating speed, and this deviation will affect driving safety to a certain extent. From the perspective of brain-like technology, we can deeply analyze the mechanism of driver behavior and analyze the process of "perceiving the road environment, making speed decisions, and taking speed control" from a micro perspective. At present, there are sufficient studies to calibrate and predict the driver's expected speed, and quantitative models based on visual road environment diagrams are also used to evaluate road safety and other aspects. However, existing research has not quantified the driver's perception and cognition of the road environment based on brain-like cognitive computing models, or analyzed the deviation between the driver's expected speed and the actual operating speed based on this. Summary of the invention

[0005] The purpose of the present invention is to provide a vehicle speed deviation analysis method based on a brain-like cognitive computing model, which takes the deviation between the driver's expected speed and the actual speed as an indicator, quantifies the driver's perception of the road environment conditions, and plays an important role in analyzing the driver's perception and cognitive characteristics when determining the expected speed, and its impact on the difference between the expected speed and the actual operating speed, and is of great significance to improving road traffic safety.

[0006] To achieve the above object, the present invention provides a vehicle speed deviation analysis method based on a brain-like cognitive computing model, which is characterized by comprising the following steps:

[0007] S1. Build a low-grade highway simulation environment to conduct a driving simulation experiment, and collect the driver's expected speed, actual speed, and EEG data during the experiment;

[0008] S2. Based on the schema theory, the driver's visual driving environment is digitally analyzed to establish a visual road environment model;

[0009] S3. Preprocess the driver's EEG data and perform spectrum analysis, and on this basis establish the relationship between the environment and the strength of the EEG signal;

[0010] S4. Based on the QN-MHP brain-like cognitive computing model, an LMM model is established by utilizing the signal intensity relationship of each brain region to analyze the deviation between the driver's expected speed and the actual speed.

[0011] Preferably, in step S1, a traffic scene simulation software is used to build a low-grade highway simulation environment, including three different scenes: towns, plains, and mountainous areas;

[0012] The EEG data were obtained by recording the EEG signals through BrainVision Recorder software. During the acquisition, the sampling rate was set to 1000 Hz, the maximum value of the electrode resistance was 40 μΩ, and the serial number of each electrode position was based on the international standard system;

[0013] The vehicle speed deviation ΔV is obtained by subtracting the actual vehicle speed V from the expected vehicle speed Vp.

[0014] Preferably, the visual road environment model in step S2 is divided into five different layers, including a visual road linear layer, a dynamic visual layer, a visual semantic layer, a visual depth layer and a visual sensitivity layer;

[0015] The boundaries of the visual road linearity in the visual road linearity layer are fitted by Catmull-Rom spline curves; the dynamic visual layer represents the dynamic changes of the driver's field of view with speed. The driver's field of view is negatively correlated with the speed, and his visual attention environment will be compressed to the center of the road as the speed increases;

[0016] The visual semantic layer uses different colors to reflect different parts of the road environment perceived by the driver. The visual semantic layer includes the following different components, including roads, forests, sky, desert, traffic signs, protective measures, and cliffs. The area ratio of the road environment perceived by the driver is used as a quantitative parameter.

[0017] The visual depth layer reflects the driver's perception of the depth information of the road environment. The average value of the driver's field of view is extracted from the visual depth layer and the dynamic visual layer as a quantitative parameter;

[0018] The visual sensitivity layer reflects the distribution of sensitive areas in each road environment area under the driver's visual perception. The calculation results are as follows:

[0019]

[0020] Among them, L Grade-CAM represents the heat map formed based on the gradient weighted class activation mapping algorithm; M i represents the i-th feature map; α i Represents the weight corresponding to the i-th feature map; ReLU is the rectified linear unit; c1, c2 are the sizes of the feature map; Z = c1*c2; Represents the pixel value of the jth row and kth column in the i-th feature map.

[0021] Preferably, the preprocessing method in step S3 is as follows:

[0022] Filtering, extracting useful information and removing unnecessary noise and interference bandpass filtering;

[0023] Resampling, in order to speed up the batch data processing, the sampling rate of the EEG signal was resampled from the original frequency of 1000Hz to 500Hz;

[0024] Independent component analysis, performed on EEG signals to separate the components of the mixed signal in order to analyze the activities of different neural sources;

[0025] Remove artifacts, remove electrooculogram, electromyography, electrode noise and power supply interference, and improve the purity and accuracy of EEG signals to better analyze and interpret brain activity;

[0026] Remove extreme values ​​and remove extreme values ​​in EEG signals to ensure the accuracy and reliability of the data;

[0027] Re-referencing changes the reference of an EEG signal to improve interpretation and analysis of the signal.

[0028] Preferably, the steps of spectrum analysis are as follows:

[0029] S301, time domain signal conversion, converting the EEG signal from the time domain to the frequency domain;

[0030] S302, calculating the power spectrum of the converted EEG signal, and determining the energy distribution at each frequency by calculating the power spectrum of the EEG signal, in decibels;

[0031] S303, using data window and averaging technology to process the EEG signal to improve the accuracy and reliability of spectrum estimation, wherein the EEG signal of the driver during the entire driving process is located at the selected points, and a 1-second time period is selected at each point to perform spectral band analysis;

[0032] S304, spectral band analysis, analyze the power spectra of different frequency bands, study the changes and characteristics of EEG signals in different frequency bands, select theta waves, alpha waves, beta waves, and gamma waves from the common frequency band range, and average the power spectrum intensity values ​​of these frequency bands to obtain the driver's brain activity.

[0033] Preferably, in step S4, a linear mixed effects regression model LMM is used for data analysis, and its expression is as follows:

[0034] Y = Xβ + Zγ + ε;

[0035] Where Y is the vector of dependent variables; X is the fixed effect score design matrix; β is the parameter vector of fixed effects; Z is the design matrix of random effects; γ is the parameter vector of random effects; ε is the vector of random error terms; Xβ expresses the effects of fixed effects; and Zγ expresses the effects of random effects.

[0036] Preferably, in the LMM model, the construction of the linear mixed model is realized by means of the lme4 package in the R language. In the modeling process, the random error term is set to the subject number participating in the simulation experiment and the different visual road environment scene points, which respectively represent the random effects between different subjects and different road scenes;

[0037] Due to the differences in the size and distribution of each data, the maximum and minimum normalization method is used to scale the data to the range of [0,1]. The processing method is as follows:

[0038]

[0039] Where x′ is the processed data point; x is the data point in the original set; min(x) is the minimum value in the data set; max(x) is the maximum value in the data set;

[0040] For the vehicle speed deviation ΔV, since it can be positive or negative, the Z-Score standardization method is used to convert the data into a form with a mean of 0 and a standard deviation of 1. The Z-Score standardization method is as follows:

[0041]

[0042] Among them, x′ is the processed data point; is the mean of the original set; σ is the standard deviation of the original set.

[0043] Preferably, the expected vehicle speed Vp, the actual vehicle speed V, and the vehicle speed deviation ΔV are predicted respectively according to the fitting results of the linear mixed model.

[0044] Therefore, the present invention adopts the above-mentioned vehicle speed deviation analysis method based on the brain-like cognitive computing model to conduct an in-depth analysis of the driver's visual perception information presented by the visual road environment model, explore the impact of different road environment elements on the driver's perception of the road environment, and thus analyze its definition of the expected vehicle speed.

[0045] Based on the QN-MHP model, a linear mixed effects model is built to model the driver's visual perception in different road environments to reflect the perception-cognition-response process. In this way, the relationship between the various parts of the brain-like cognitive computing model is analyzed based on the influence of various visual elements on the driver's cognitive behavior.

[0046] The relationship between the deviation between the driver's expected speed and the actual operating speed and the influencing factors is interpreted from a visual perspective, thereby contributing to the optimization of road geometry and road environment.

[0047] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a schematic diagram of an overall method of an embodiment of a vehicle speed deviation analysis method based on a brain-like cognitive computing model of the present invention;

[0049] Figure 2 It is a visual road environment model diagram of an embodiment of a vehicle speed deviation analysis method based on a brain-like cognitive computing model of the present invention;

[0050] Figure 3 It is a construction diagram of a visual road linear layer of an embodiment of a vehicle speed deviation analysis method based on a brain-like cognitive computing model of the present invention;

[0051] Figure 4 It is a visual semantic layer and a visual depth layer of an embodiment of a vehicle speed deviation analysis method based on a brain-like cognitive computing model of the present invention;

[0052] Figure 5 It is a visual sensitivity layer of an embodiment of a vehicle speed deviation analysis method based on a brain-like cognitive computing model of the present invention;

[0053] Figure 6 It is a visual processing architecture of a QN-MHP model of an embodiment of a vehicle speed deviation analysis method based on a brain-like cognitive computing model of the present invention;

[0054] Figure 7It is a prediction result diagram of the expected vehicle speed Vp of an embodiment of a vehicle speed deviation analysis method based on a brain-like cognitive computing model of the present invention; Figure 7 (a) is a scatter plot of the prediction results of the server F for the expected vehicle speed Vp; Figure 7 (b) is a scatter plot of the prediction results of the visual road environment model for the expected vehicle speed Vp; Figure 7 (c) is a histogram of the prediction results of the server F for the expected vehicle speed Vp; Figure 7 (d) is the histogram of the prediction results of the visual road environment model for the expected vehicle speed Vp;

[0055] Figure 8 It is a prediction result diagram of the actual vehicle speed V of an embodiment of a vehicle speed deviation analysis method based on a brain-like cognitive computing model of the present invention; Figure 8 (a) is a scatter plot of the prediction results of server Z for the actual vehicle speed V; Figure 8 (b) is a scatter plot of the prediction results of the visual road environment model for the actual vehicle speed V; Figure 8 (c) is a histogram of the prediction results of the server Z for the actual vehicle speed V; Figure 8 (d) is the histogram of the prediction results of the visual road environment model for the actual vehicle speed V;

[0056] Fig. 9 It is a prediction result diagram of a vehicle speed deviation ΔV of an embodiment of a vehicle speed deviation analysis method based on a brain-like cognitive computing model of the present invention; Fig. 9 (a) is a scatter plot of the prediction results of the vehicle speed deviation ΔV by servers F and Z; Fig. 9 (b) is a scatter plot of the prediction results of the vehicle speed deviation ΔV by server Z; Fig. 9 (c) is a scatter plot of the prediction results of the visual road environment model for the vehicle speed deviation ΔV; Fig. 9 (d) is the histogram of the prediction results of the vehicle speed deviation ΔV for servers F and Z; Fig. 9 (e) is the histogram of the prediction results of the server Z for the vehicle speed deviation ΔV; Fig. 9 (f) in the figure is a histogram of the prediction results of the vehicle speed deviation ΔV by the visual road environment model. DETAILED DESCRIPTION

[0057] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0058] Unless otherwise defined, technical or scientific terms used in the present invention shall have the common meanings understood by one having ordinary skills in the field to which the present invention belongs.

[0059] Embodiment 1

[0060] like Figure 1As shown, the present invention provides a vehicle speed deviation analysis method based on a brain-like cognitive computing model, which is characterized by comprising the following steps:

[0061] S1. Build a low-grade highway simulation environment to conduct a driving simulation experiment, and collect the driver's expected speed, actual speed, and EEG data during the experiment;

[0062] In step S1, traffic scene simulation software is used to build low-level highway driving scenes, including three different scenes: towns, plains, and mountainous areas;

[0063] The EEG data were obtained by recording the EEG signals through BrainVision Recorder software. During the acquisition, the sampling rate was set to 1000 Hz, the maximum value of the electrode resistance was 40 μΩ, and the serial number of each electrode position was based on the international standard system;

[0064] The speed data includes the driver's actual vehicle speed V, the expected vehicle speed Vp, and the speed error ΔV obtained by subtracting the two.

[0065] S2, based on the schema theory, the driver's visual driving environment is digitally analyzed to establish a visual road environment model; the visual road environment model in step S2 is divided into five different layers, including a visual road linear layer, a dynamic visual layer, a visual semantic layer, a visual depth layer, and a visual sensitivity layer; Figure 2 shown.

[0066] The boundaries of the visual road alignment in the visual road alignment layer are fitted by Catmull-Rom spline curves. The advantage of this method is that it can pass through all control points, is efficient and accurate, and the local change of one control point does not affect other parts, which has local controllability. Figure 3 As shown in the figure, the establishment of the visual road linear layer first requires the construction of a coordinate system. This coordinate system takes the lower left corner of the driver's field of view as the origin, and the left and right spline curves each have four control points, namely (P 1L ,P 2L ,P 3L ,P 4L )、(P 1R ,P 2R ,P 3R ,,P 4R ), these control points are used to fit the road line shape and divide the road into three areas, namely "near view", "mid view" and "distant view". These three areas can quantify the driver's visual perception from the perspective of different visual features.

[0067] The dynamic visual layer represents the dynamic changes of the driver's field of view with speed. The driver's field of view is negatively correlated with the speed of the vehicle, and his visual attention environment will be compressed to the center area of ​​the road as the speed increases.

[0068] The visual semantic layer uses different colors to reflect different parts of the road environment perceived by the driver. The parts of the road environment will be grouped by semantic segmentation technology and represented by the same color, such as Figure 4 As shown, the visual semantic layer includes the following different components, including roads, forests, sky, desert, traffic signs, protective measures, and cliffs, with the area ratio of their appearance in the visual road environment as a quantitative parameter;

[0069] The visual depth layer reflects the driver’s perception of the depth information of the road environment, such as Figure 4 As shown, the depth value of each part of the image is estimated through depth estimation technology, and then the road environment components with different depth values ​​are represented in the visual depth layer with different colors. The average value of the driver's field of view is extracted from the visual depth layer and the dynamic visual layer as a quantitative parameter;

[0070] The visual sensitivity layer reflects the distribution of sensitive areas in each road environment area under the driver's visual perception. The visual sensitivity layer is based on visual sensitivity and divides the driver's visual sensitive area based on the gradient weighted class activation mapping algorithm (Grad-CAM). The sensitivity layer is established, such as Figure 5 As shown, the calculation results are as follows:

[0071]

[0072] Among them, L Grade-CAM represents the heat map formed based on the gradient weighted class activation mapping algorithm; M i represents the i-th feature map; α i Represents the weight corresponding to the i-th feature map; ReLU is the rectified linear unit; c1, c2 are the sizes of the feature map; Z = c1*c2; Represents the pixel value of the jth row and kth column in the i-th feature map.

[0073] S3, preprocessing the driver's EEG data and performing spectrum analysis, on this basis, establishing the relationship between the environment and the strength of the EEG signal; the preprocessing method in step S3 is as follows:

[0074] Filtering, extracting useful information and removing unnecessary noise and interference bandpass filtering;

[0075] Resampling, in order to speed up the batch data processing, the sampling rate of the EEG signal was resampled from the original frequency of 1000Hz to 500Hz;

[0076] Independent component analysis, performed on EEG signals to separate the components of the mixed signal in order to analyze the activities of different neural sources;

[0077] Remove artifacts, remove electrooculogram, electromyography, electrode noise and power supply interference, and improve the purity and accuracy of EEG signals to better analyze and interpret brain activity;

[0078] Remove extreme values ​​and remove extreme values ​​in EEG signals to ensure the accuracy and reliability of the data;

[0079] Re-referencing changes the reference of an EEG signal to improve interpretation and analysis of the signal.

[0080] Spectral analysis describes the frequency characteristics of a signal by calculating the power spectrum of the signal in a field. The power spectrum shows the power or energy distribution of the signal at different frequencies. The present invention uses spectral analysis to process EEG signals, aiming to calculate the power spectrum density of the EEG signal of the subject at a specific point in a short period of time, which represents the power or energy density of the EEG signal in different frequency ranges. The steps of spectral analysis are as follows:

[0081] S301, time domain signal conversion, converting the EEG signal from the time domain to the frequency domain;

[0082] S302, calculating the power spectrum of the converted EEG signal, and determining the energy distribution at each frequency by calculating the power spectrum of the EEG signal, in decibels;

[0083] S303, using data window and averaging technology to process the EEG signal to improve the accuracy and reliability of spectrum estimation, wherein the EEG signal of the driver during the entire driving process is located at the selected points, and a 1-second time period is selected at each point to perform spectral band analysis;

[0084] S304, spectral band analysis, analyze the power spectra of different frequency bands, study the changes and characteristics of EEG signals in different frequency bands, select theta waves, alpha waves, beta waves, and gamma waves from the common frequency band range, and average the power spectrum intensity values ​​of these frequency bands to obtain the driver's brain activity.

[0085] S4. Based on the QN-MHP brain-like cognitive computing model, an LMM model is established using the signal intensity relationship of each brain region to analyze the deviation between the driver's expected speed and the actual speed. The visual processing architecture of the QN-MHP model is as follows: Figure 6 As shown, the servers involved are: visual servers 1, 2, 3, 4, cognitive servers A, C, F, H, and motion servers W, Y, Z.

[0086] The linear mixed effects regression model LMM was used for data analysis, and its expression is as follows:

[0087] Y = Xβ + Zγ + ε;

[0088] Where Y is the vector of dependent variables; X is the fixed effect score design matrix; β is the parameter vector of fixed effects; Z is the design matrix of random effects; γ is the parameter vector of random effects; ε is the vector of random error terms; Xβ expresses the effects of fixed effects; and Zγ expresses the effects of random effects.

[0089] In the LMM model, the linear mixed model was constructed with the help of the lme4 package in the R language. During the modeling process, the random error term was set as the subject number participating in the simulation experiment and the different visual road environment scene points, which respectively represented the random effects between different subjects and different road scenes;

[0090] Due to the differences in the size and distribution of each data, the maximum and minimum normalization method is used to scale the data to the range of [0,1]. The processing method is as follows:

[0091]

[0092] Where x′ is the processed data point; x is the data point in the original set; min(x) is the minimum value in the data set; max(x) is the maximum value in the data set;

[0093] For the vehicle speed deviation ΔV, since it can be positive or negative, the Z-Score standardization method is used to convert the data into a form with a mean of 0 and a standard deviation of 1. The Z-Score standardization method is as follows:

[0094]

[0095] Among them, x′ is the processed data point; is the mean of the original set; σ is the standard deviation of the original set.

[0096] The results obtained after normalizing the power spectrum density and speed index of the electrical signals of each server are shown in the following table.

[0097] Table 1 Model data normalization results

[0098] server Minimum Maximum average value Standard Deviation 1 0.01 0.83 0.38 0.18 2 0.01 0.85 0.40 0.20 3 0.01 0.80 0.33 0.16 4 0.01 0.82 0.36 0.18 A 0.01 0.81 0.35 0.16 C 0.01 0.85 0.38 0.20 F 0.01 0.83 0.41 0.21 H 0.01 0.81 0.35 0.16 W 0.01 0.79 0.32 0.15 Y 0.00 0.80 0.36 0.17 Z 0.00 0.88 0.40 0.21

[0099] Table 2 Normalization results of speed index

[0100]

[0101]

[0102] According to the fitting results of the linear mixed model, the expected vehicle speed Vp, actual vehicle speed V and vehicle speed deviation ΔV are predicted respectively.

[0103] Prediction of expected vehicle speed Vp

[0104] Based on the linear mixed effect model, the relationship between server F and expected vehicle speed Vp is established with different points as random effects; the relationship between the significant elements of the visual road environment model and the expected vehicle speed Vp is established with different drivers as random effects. After cross-validation by the bootstrap method, its distribution characteristics and prediction accuracy indicators are shown in Tables 3 and 4, and the scatter plot and prediction error histogram are shown in Figure 7 shown.

[0105] Table 3 Distribution characteristic indicators (unit: km / h)

[0106] Prediction Methods Mean median Standard Deviation Maximum Minimum Server F <![CDATA[-2e -14 ]]> 0.67 10.76 52.56 -41.99 Visual road environment model <![CDATA[-6e -14 ]]> 0.88 11.12 42.60 -51.67

[0107] Table 4 Prediction accuracy indicators

[0108]

[0109] From the data in Table 3 and Table 4, we can see that compared with the visual road environment model, the root mean square error (RMSE) of predicting the driver's speed through server F is significantly smaller, and the mean absolute error (MAE), mean square error (MSE) and mean absolute percentage error (MAPE) are smaller. The mean and median of the driver's speed predicted by server F are small, indicating that the overall prediction performance of the model is good; however, the maximum and minimum values ​​of the prediction error are relatively extreme, which are -41.99 and 52.56 respectively, and the standard deviation is 10.76, indicating that the model still has great limitations in predicting extreme values.

[0110] Depend on Figure 7 The scatter plot in shows that the distribution of the scatter points around the perfect prediction line is relatively ideal when predicting the driver's speed through server F, but there are still many scatter points with extreme values ​​that deviate from the perfect prediction line; the histogram shows that the prediction error is mainly concentrated in the interval [-15,15], indicating that the overall prediction error is small and relatively ideal. It can be seen that the overall prediction performance of the model is good, but the prediction accuracy for extreme data needs to be improved.

[0111] Prediction of actual vehicle speed V

[0112] Based on the linear mixed effect model, the relationship between server Z and actual vehicle speed V is established with different drivers as random effects; the relationship between the significant elements of the visual road environment model and actual vehicle speed V is established with different drivers as random effects. Its distribution characteristics and prediction accuracy indicators are shown in Tables 5 and 6, and the scatter plot and prediction error histogram are shown in Figure 8 shown.

[0113] Table 5 Distribution characteristic indicators (unit: km / h)

[0114] Prediction Methods Mean median Standard Deviation Maximum Minimum Server Z -2e-13 0.43 13.66 40.46 -50.60 Visual road environment model -9e-14 0.53 13.46 39.66 -49.26

[0115] Table 6 Prediction accuracy indicators

[0116]

[0117] From the data in Table 5 and Table 6, we can see that compared with the visual road environment model, the root mean square error (RMSE) of predicting the driver's speed through server Z is significantly smaller, and the mean absolute error (MAE), mean square error (MSE) and mean absolute percentage error (MAPE) are similar. The mean and median of the driver's speed predicted by server Z are small, indicating that the overall prediction performance of the model is good; however, the maximum and minimum values ​​of the prediction error are relatively extreme, which are -50.60 and 40.46 respectively, and the standard deviation is 13.66, indicating that the model still has great limitations in predicting extreme values.

[0118] Depend on Figure 8 The scatter plot in the figure shows that the distribution of the scatter points around the perfect prediction line is relatively ideal when predicting the driver's speed through server Z, but there are still many scatter points with extreme values ​​that deviate from the perfect prediction line; the histogram shows that the prediction error is mainly concentrated in the interval [-15,15], indicating that the overall prediction error is small and relatively ideal. It can be seen that the overall prediction performance of the model is good, but the prediction accuracy for extreme data needs to be improved.

[0119] Prediction of vehicle speed deviation ΔV

[0120] Based on the linear mixed effect model, with different drivers as random effects, the relationship between server Z and vehicle speed deviation ΔV, and the relationship between server F, Z and vehicle speed deviation ΔV are established; with different drivers as random effects, the relationship between the significant elements of the visual road environment model and vehicle speed deviation ΔV is established. Its distribution characteristics and prediction accuracy indicators are shown in Tables 7 and 8, and the scatter plot and prediction error histogram are shown in Fig. 9 shown.

[0121] Table 7 Distribution characteristic indicators (unit: km / h)

[0122] Prediction Methods Mean median Standard Deviation Maximum Minimum Server F, Z <![CDATA[-4e -14 ]]> 0.43 16.20 70.85 -57.33 Server Z <![CDATA[-2e -14 ]]> 0.42 16.37 70.15 -58.02 Visual road environment model <![CDATA[3e -14 ]]> -0.04 13.39 60.73 -50.89

[0123] Table 8 Prediction accuracy indicators

[0124]

[0125]

[0126] From the data in Table 7 and Table 8, we can see that compared with the visual road environment model, the root mean square error (RMSE) of predicting the driver's speed through server Z is significantly smaller, and the mean absolute error (MAE) and mean square error (MSE) are slightly larger. Among them, since MAPE is very sensitive to small observations, its numerical value has little reference significance without comparison. The mean and median of the driver's speed predicted by server Z are small, indicating that the overall prediction performance of the model is good; but the maximum and minimum values ​​of the prediction error are relatively extreme, which are -58.02 and 70.15 respectively, indicating that the model still has great limitations in predicting extreme values.

[0127] Depend on Fig. 9 The scatter plot in the figure shows that the distribution of the scatter points around the perfect prediction line is relatively ideal when predicting the driver's speed through server Z, but there are still many scatter points with extreme values ​​that deviate from the perfect prediction line; the histogram shows that the prediction error is mainly concentrated in the interval [-15,15], indicating that the overall prediction error is small and relatively ideal. It can be seen that the overall prediction performance of the model is good, but the prediction accuracy for extreme data needs to be improved.

[0128] From the above results, it can be concluded that using the linear mixed effects model, taking into account the heterogeneity of drivers or driving space, the speed index is predicted by the server of the brain-like cognitive computing model. Compared with the prediction of the speed index by the visual road environment model, it has relatively better stability and accuracy, but the prediction accuracy of extreme values ​​still needs to be improved.

[0129] Therefore, the present invention adopts the above-mentioned vehicle speed deviation analysis method based on the brain-like cognitive computing model to quantify the driver's perception of the road environment conditions, which plays an important role in analyzing the driver's perception and cognitive characteristics when determining the expected speed, and its impact on the difference between the expected speed and the actual operating speed, and is of great significance to improving road traffic safety.

[0130] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A vehicle speed deviation analysis method based on a brain-like cognitive computing model, characterized in that: The following steps are involved: S1. Build a low-grade highway simulation environment to conduct a driving simulation experiment, and collect the driver's expected speed, actual speed, speed deviation and EEG data during the experiment; S2. Based on the schema theory, the driver's visual driving environment is digitally analyzed to establish a visual road environment model; S3, preprocessing the driver's EEG data and performing spectrum analysis, and establishing the relationship between the environment and the EEG signal intensity based on the spectrum analysis results; S4. Based on the QN-MHP brain-like cognitive computing model, an LMM model is established using the relationship between the environment and the intensity of EEG signals to analyze the deviation between the driver's expected speed and the actual speed.

2. The vehicle speed deviation analysis method based on the brain-like cognitive computing model according to claim 1 is characterized by: In step S1, traffic scene simulation software is used to build a low-grade highway simulation environment, including three different scenes: township, plain, and mountainous area; The EEG data were obtained by recording the EEG signals using the Brain Vision Recorder software. During the acquisition, the sampling rate was set to 1000 Hz, the maximum electrode resistance was set to 40 μΩ, and the position number of each electrode was based on the international standard system; The vehicle speed deviation ΔV is obtained by subtracting the actual vehicle speed V from the expected vehicle speed Vp.

3. The vehicle speed deviation analysis method based on the brain-like cognitive computing model according to claim 2 is characterized by: In step S2, the visual road environment model is divided into five different layers, including a visual road linear layer, a dynamic visual layer, a visual semantic layer, a visual depth layer, and a visual sensitivity layer; The boundaries of the visual road linearity in the visual road linearity layer are fitted by Catmull-Rom spline curves; the dynamic visual layer represents the dynamic changes of the driver's field of view with speed. The driver's field of view is negatively correlated with the speed, and his visual attention environment will be compressed to the center of the road as the speed increases; The visual semantic layer uses different colors to reflect different parts of the road environment perceived by the driver. The visual semantic layer includes the following different components, including roads, forests, sky, desert, traffic signs, protective measures, and cliffs. The area ratio of the road environment perceived by the driver is used as a quantitative parameter. The visual depth layer reflects the driver's perception of the depth information of the road environment. The average value of the driver's field of view is extracted from the visual depth layer and the dynamic visual layer as a quantitative parameter; The visual sensitivity layer reflects the distribution of sensitive areas in each road environment area under the driver's visual perception. The calculation results are as follows: Among them, L Grade-CAM represents the heat map formed based on the gradient weighted class activation mapping algorithm; M i represents the i-th feature map; α i Represents the weight corresponding to the i-th feature map; ReLU is the rectified linear unit; c1, c2 are the sizes of the feature map; Z = c1*c2; Represents the pixel value of the jth row and kth column in the i-th feature map.

4. The vehicle speed deviation analysis method based on the brain-like cognitive computing model according to claim 3 is characterized in that: The preprocessing method in step S3 is as follows: Filtering, resampling, independent component analysis, artifact removal, extreme value removal, re-referencing.

5. The vehicle speed deviation analysis method based on the brain-like cognitive computing model according to claim 4 is characterized in that: In step S3, the steps of spectrum analysis are as follows: S301, time domain signal conversion, converting the EEG signal from the time domain to the frequency domain; S302, calculating the power spectrum of the converted EEG signal, and determining the energy distribution at each frequency by calculating the power spectrum of the EEG signal; S303, using data window and averaging technology to process the EEG signal, wherein the EEG signal of the driver during the entire driving process is located at the selected points, and a 1-second time period is selected at each point to perform spectral band analysis; S304, spectral band analysis, specifically, analyzing the power spectra of different frequency bands, studying the changes and characteristics of EEG signals in different frequency bands, selecting theta waves, alpha waves, beta waves, and gamma waves from the common frequency band range, and averaging the power spectrum intensity values ​​of these frequency bands to obtain the driver's brain activity.

6. The vehicle speed deviation analysis method based on the brain-like cognitive computing model according to claim 5 is characterized in that: In step S4, a linear mixed effects regression model LMM is used for data analysis, and its expression is as follows: Y = Xβ + Zγ + ε; Where Y is the vector of dependent variables; X is the fixed effect score design matrix; β is the parameter vector of fixed effects; Z is the design matrix of random effects; γ is the parameter vector of random effects; ε is the vector of random error terms; Xβ expresses the effects of fixed effects; and Zγ expresses the effects of random effects.

7. The vehicle speed deviation analysis method based on the brain-like cognitive computing model according to claim 6 is characterized in that: In the LMM model, the linear mixed model was constructed with the help of the lme4 package in the R language. During the modeling process, the random error term was set as the subject number participating in the simulation experiment and the different visual road environment scene points, which respectively represented the random effects between different subjects and different road scenes; Due to the differences in the size and distribution of each data, the maximum and minimum normalization method is used to scale the data to the range of [0,1]. The processing method is as follows: Where x′ is the processed data point; x is the data point in the original set; min(x) is the minimum value in the data set; max(x) is the maximum value in the data set; For the vehicle speed deviation ΔV, since it can be positive or negative, the Z-Score standardization method is used to convert the data into a form with a mean of 0 and a standard deviation of 1. The Z-Score standardization method is as follows: Among them, x′ is the processed data point; is the mean of the original set; σ is the standard deviation of the original set.

Images