Highway road condition environment safety early warning and monitoring system

Abstract

The invention relates to the field of road condition environment safety early warning and monitoring, in particular to an expressway road condition environment safety early warning and monitoring system which comprises a data fusion module, a self-adaptive risk assessment module, a path optimization module and platform software. The data fusion module fuses RTK positioning, pavement temperature, humidity, friction coefficient and other multi-dimensional information by improving a weighted Kalman filtering algorithm, and dynamically adjusts the weight according to the sensor confidence coefficient; the adaptive risk assessment module constructs a dynamic risk assessment model by using a random forest algorithm, calculates a risk value of a road section and grades the risk value; the path optimization module generates a real-time optimal patrol path based on a deep network algorithm and a training algorithm, preferentially covers a high-risk road section, and improves the monitoring efficiency. According to the invention, accurate monitoring, risk assessment and patrol path optimization of the highway road condition environment are realized, and the driving safety of the highway is more effectively guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of road condition and environmental safety early warning monitoring systems, in particular to a highway road condition and environmental safety early warning monitoring system. Background Art

[0002] Currently, highway road condition and environmental monitoring faces numerous challenges, becoming a bottleneck restricting improvements in traffic safety and management efficiency. Existing risk assessment methods are outdated and inaccurate, relying heavily on simple rules or a small number of parameters, failing to fully tap into the value of historical accident data and lacking the ability to integrate real-time environmental parameters. Faced with complex and changing road conditions, such as varying weather and traffic flow, it is difficult to accurately assess road section risk levels in real time, making it impossible to issue effective warnings in a timely manner, increasing the risk of accidents.

[0003] Patrol car route planning lacks intelligence and is mostly based on fixed routes or experience, without considering factors such as the real-time risk level of the road section, traffic flow and speed limit; high-risk sections may not be patrolled enough, while low-risk sections are patrolled repeatedly, wasting resources and reducing monitoring efficiency, making it difficult to detect and deal with safety hazards in a timely manner. Therefore, the invention of a highway road condition environment safety early warning monitoring system is needed to solve the above problems. Summary of the Invention

[0004] In order to solve the technical problems raised by the above background technology, the present invention provides a highway road condition environment safety early warning monitoring system.

[0005] The purpose of the present invention can be achieved through the following technical solutions: The present invention is a highway road condition environment safety early warning monitoring system, including a data fusion module, an adaptive risk assessment module, a path optimization module and platform software.

[0006] The data fusion module uses an improved weighted Kalman filter algorithm to integrate RTK positioning data, road surface temperature, humidity, and friction coefficient information, dynamically adjust the weight of sensor confidence, and effectively reduce noise interference. The specific process is as follows:

[0007] The data fusion module is equipped with the patrol car's front-end acquisition equipment. The various sensors of the front-end acquisition equipment collect real-time data on the highway's RTK positioning, road surface temperature, humidity, and friction coefficient. These data are marked as (x, y), T, H, μ, and S, respectively, where S represents the timestamp. This is used to construct a 6-dimensional state vector fusion model Zf = [xy TH μ s] T ;

[0008] Set the sensor dynamic weight w i ∈[0,1], its calculation logic is: Among them C i The initial calibration value of the sensor, σ is expressed as the data fluctuation coefficient, Ei is the environmental adaptability coefficient, i represents the sensor number, and n represents the total number of sensors

[0009] Establish the state transfer equation, and its calculation logic is: Z f | f-1 =P f Z f-1 |f -1 +C f +E f , where Z f | f-1 is the state prediction value at time f based on the state estimation at time f-1, P f is the state transfer matrix, C f is the control input matrix, E f is the process noise; linearize it using first-order Taylor expansion: Where β represents the sensitivity coefficient of humidity to temperature change, which is calibrated by historical data; process noise E f According to the dynamic adjustment of the patrol car's motion state, the calculation logic is: E f =E base +η·α 2 , where α is the vehicle acceleration, E base is the baseline value of process noise, η is the noise amplification factor;

[0010] The Kalman gain is calculated based on the observation equation and predicted state of each sensor to obtain the fused state estimate. The calculation logic is as follows: Among them, θ f|f-1 is the prediction covariance, is the Kalman gain of the i-th sensor at time f, BP i,f The observation matrix of the i-th sensor, is the observation noise of i sensors; the prediction covariance θ f|f-1 The calculation logic is: θ f-1|f-1 Expressed as the covariance matrix of the optimal state estimate at time f-1; the observation matrix BP i,f It maps the 6-dimensional state vector to the observation space, expressed as: Temperature and humidity are directly observed, and the friction coefficient is indirectly estimated through other parameters; noise is observed is based on the sensor confidence w i Dynamic weighting.

[0011] The adaptive risk assessment module builds a dynamic risk assessment model based on the random forest algorithm, integrates historical accident data with real-time environmental parameters, and calculates the risk level of highway sections in real time. Specifically:

[0012] Obtain historical accident data from the platform software, extract historical accident characteristics and road condition characteristics from the historical accident data and real-time parameters to form a feature set. Historical accident characteristics include the frequency, severity, and type of historical accidents on the road section, and road condition characteristics include real-time friction coefficient, water film height, road ice content, road temperature, and relative humidity. Clean the collected data to remove outliers and missing values, and use normalization to convert the data to the same dimension.

[0013] Construct a random forest model. The random forest algorithm consists of multiple decision trees. The random forest model constructed in this module contains 100 decision trees. During the training process, each decision tree randomly selects some features from the feature set. The number of features selected is the square root of the total number of features. Sub-datasets are generated by random sampling with replacement from the training dataset.

[0014] For each decision tree, based on the selected sub-dataset and feature set, the Gini index is used as the splitting criterion to build a decision tree model. The Gini index is used to measure the impurity of the data. Its calculation logic is: Where m represents the total number of risk level categories, j represents the number of risk level categories, δ j is the probability value of the sample belonging to the jth class; in the process of splitting the nodes of the decision tree, the feature and split point that causes the Gini index to drop the most are selected until the depth of the tree reaches the preset threshold of the platform software;

[0015] When a new batch of real-time environmental parameters and historical accident feature sets are input, each decision tree will make a prediction and output the corresponding risk level prediction result. The random forest model uses a voting mechanism to count the prediction results of all decision trees and use the result of the majority vote as the final risk level prediction. The frequency of each risk level in the prediction results of all decision trees is then calculated to obtain the risk probability value. The 100 decision trees are divided into r1 low-risk ones, r2 medium-risk ones, and r3 high-risk ones. The risk probability calculation logic is: and Obtain low risk probability value δ1, medium risk probability value δ2 and high risk probability value δ3;

[0016] Based on the calculated risk probability value, the road section risk level is divided into three levels: low, medium and high. The specific division rules are as follows: when δ1≥0.6, it is judged as a low risk level; when 0.3<δ2 is less than 0.6, it is judged as a medium risk level; when δ3≥0.3, it is judged as a high risk level; the corresponding road section risk level is sent to the software platform and path optimization module.

[0017] The path optimization module generates real-time optimal patrol routes for patrol vehicles based on the deep Q-network algorithm, giving priority to covering high-risk sections and improving monitoring efficiency. The specific process is as follows:

[0018] The highway network is abstracted into a directed graph G = (K, L), where K is a node set representing each monitoring point on the highway, and L is an edge set representing the road sections connecting the nodes. The state e is set to consist of the real-time risk level of the current patrol car's location, the traffic flow of the road section, and the speed limit. The action d is set to indicate that the patrol car selects the next adjacent road section to travel from the current location. If the current patrol car is at node k, l , then the set of adjacent nodes is N(k l ), then the action space D={d1,d2......d g}, where g is the number of adjacent nodes, and each action d corresponds to l Drive to N(k l ) a node in

[0019] A reward function M is introduced into the deep Q-network algorithm to prioritize high-risk sections. When high-risk sections > medium-risk sections > low-risk sections > 0, patrol cars are encouraged to prioritize high-risk sections. When patrol cars increase their patrol time, they receive a negative reward. When patrol cars repeat a patrolled section, they also receive a negative reward.

[0020] The deep Q network algorithm is trained. The patrol car continuously performs actions in the environment, obtains rewards according to the reward function, and converts the state e, action d, reward M and next state e into * The data is stored in the experience replay pool. Every certain driving distance, a batch of samples are randomly drawn from the experience replay pool for training. The target Q value is calculated. The calculation logic is as follows: in is the discount factor, ψ - is the mean square error between the target network parameters and the current Q network; the stochastic gradient descent algorithm U is used to update the Q network parameters and continuously optimize the decision-making strategy of the intelligent agent. Its calculation logic is:

[0021] After the algorithm training is completed, the patrol car selects the action d that maximizes the Q(e,d) value in the action space based on the current state e, determines the next driving section, and so on. The real-time optimal patrol path is generated and sent to the software platform. The calculation logic is as follows:

[0022] Compared with existing technologies, the present invention offers the following advantages: The path optimization module, based on a deep network algorithm, abstracts the highway network into a directed graph and rationally sets state, action, and reward functions. Through a training algorithm, patrol vehicles can generate optimal patrol routes in real time based on real-time risk levels, traffic flow, speed limits, and other information, prioritizing high-risk sections. This optimization strategy avoids blind patrols, reduces patrol time and duplicate patrols, significantly improves monitoring efficiency, and enables more rational allocation of patrol resources, enabling more timely detection and resolution of safety hazards.

[0023] The adaptive risk assessment module uses a random forest algorithm to deeply integrate historical accident data with real-time environmental parameters. It extracts features from multiple aspects to construct a feature set, and accurately calculates the risk level of highway sections in real time based on the model, dividing the risk level into three levels. Compared with traditional risk assessment methods, it not only considers more comprehensive factors, but also can update the assessment results in a timely manner as the road conditions change, providing accurate risk information to traffic management departments, facilitating the implementation of targeted measures in advance, and effectively reducing the probability of accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. The following drawings are not intentionally scaled to the actual size, and the focus is on illustrating the main purpose of the present invention.

[0025] Figure 1 This is a principle block diagram of the present invention.

[0026] Figure 2 This is a block diagram of the path optimization module of the present invention. DETAILED DESCRIPTION

[0027] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts also fall within the scope of protection of the present invention.

[0028] Please refer to Figure 1 As shown, the present invention is a highway road condition environment safety early warning monitoring system, which includes a data fusion module, an adaptive risk assessment module, a path optimization module and platform software.

[0029] The data fusion module uses an improved weighted Kalman filter algorithm to integrate RTK positioning data, road surface temperature, humidity, friction coefficient and other multi-dimensional information, dynamically adjust the weight of sensor confidence, and effectively reduce noise interference. The specific process is as follows:

[0030] The data fusion module is equipped with the patrol car's front-end acquisition equipment. The various sensors of the front-end acquisition equipment collect real-time data such as RTK positioning, road surface temperature, humidity, and friction coefficient of the highway. These data are marked as (x, y), T, H, μ, and S, where S represents the timestamp. This is used to construct the 6-dimensional state vector fusion model Zf = [xy TH μ s] T ;

[0031] Set the sensor dynamic weight w i ∈[0,1], its calculation logic is: Among them C i The initial calibration value of the sensor, σ, is expressed as the data fluctuation coefficient, which is obtained by dividing the standard deviation of the data within 30 minutes by the mileage range of the patrol car, E i is the environmental adaptability coefficient, i represents the sensor number, and n represents the total number of sensors. It should be noted that the purpose of setting the dynamic weight is to dynamically adjust the weight in the fusion according to the real-time performance of the sensor to improve robustness.

[0032] To address the vibrations generated by patrol cars during normal driving, interference from vehicle exhaust, and the influence of road contamination on the friction coefficient sensor without affecting measurement accuracy, dynamic weighting of sensor confidence, unified modeling of heterogeneous data, and dynamic adjustment of covariance are added to the traditional Kalman filter to reduce noise and improve accuracy.

[0033] Establish the state transfer equation, and its calculation logic is: Z f | f-1 =P f Z f-1 | f-1 +C f +E f , where Z f | f-1 is the state prediction value at time f based on the state estimation at time f-1, P f is the state transfer matrix, which describes the transition relationship of the system state from time f-1 to time f, C f is the control input matrix, E f is process noise; the state transfer matrix needs to consider the freezing process of temperature and humidity, and is linearized using first-order Taylor expansion: Where β represents the sensitivity coefficient of humidity to temperature change, which is calibrated by historical data; process noise E f According to the dynamic adjustment of the patrol car's motion state, such as acceleration and deceleration, the calculation logic is: E f =E base +η·α 2 , where α is the vehicle acceleration, E base is the baseline value of process noise, η is the noise amplification factor;

[0034] The Kalman gain is calculated based on the observation equation and predicted state of each sensor to obtain the fused state estimate. The calculation logic is as follows: Among them, θ f|f-1 is the prediction covariance, is the Kalman gain of the i-th sensor at time f, BP i,f The observation matrix of the i-th sensor, is the observation noise of i sensors; the prediction covariance θ f|f-1 The calculation logic is: θ f-1|f-1 Expressed as the covariance matrix of the optimal state estimate at time f-1; the observation matrix BP i,f It maps the 6-dimensional state vector to the observation space, expressed as: Temperature and humidity are directly observed, and the friction coefficient is indirectly estimated through other parameters; noise is observed is based on the sensor confidence w i Dynamic weighting.

[0035] The adaptive risk assessment module builds a dynamic risk assessment model based on the random forest algorithm, integrates historical accident data with real-time environmental parameters, and calculates the risk level of highway sections in real time. Specifically:

[0036] Historical accident data is obtained from the platform software. Historical accident characteristics and road condition characteristics are extracted from the historical accident data and real-time parameters to form a feature set. Historical accident characteristics include the frequency, severity, and type of historical accidents on the road section, which can be used as a reference for assessing current risks. Road condition characteristics include real-time friction coefficient, water film height, road ice content, road temperature, and relative humidity. The collected data is cleaned to remove outliers and missing values. Normalization is used to transform the data so that data with different characteristics are in the same dimension, which facilitates subsequent algorithm processing.

[0037] Construct a random forest model. The random forest algorithm consists of multiple decision trees. The random forest model constructed in this module contains 100 decision trees. During the training process, each decision tree randomly selects some features from the feature set. The number of features selected is the square root of the total number of features. Sub-datasets are generated by random sampling with replacement from the training dataset.

[0038] For each decision tree, based on the selected sub-dataset and feature set, the Gini index is used as the splitting criterion to build a decision tree model. The Gini index is used to measure the impurity of the data. Its calculation logic is: Where m represents the total number of risk level categories, j represents the number of risk level categories, δ jis the probability value of the sample belonging to the jth class; in the process of splitting the nodes of the decision tree, the feature and split point that causes the Gini index to drop the most are selected until the depth of the tree reaches the preset threshold of the platform software;

[0039] When a new batch of real-time environmental parameters and historical accident feature sets are input, each decision tree will make a prediction and output the corresponding risk level prediction result. The random forest model uses a voting mechanism to count the prediction results of all decision trees and use the result of the majority vote as the final risk level prediction. The frequency of each risk level in the prediction results of all decision trees is then calculated to obtain the risk probability value. The 100 decision trees are divided into r1 low-risk ones, r2 medium-risk ones, and r3 high-risk ones. The risk probability calculation logic is: and Obtain low risk probability value δ1, medium risk probability value δ2 and high risk probability value δ3;

[0040] Based on the calculated risk probability value, the road section risk level is divided into three levels: low, medium and high. The specific division rules are as follows: when δ1≥0.6, it is judged as a low risk level; when 0.3<δ2 is less than 0.6, it is judged as a medium risk level; when δ3≥0.3, it is judged as a high risk level; the corresponding road section risk level is sent to the software platform and path optimization module.

[0041] The path optimization module generates real-time optimal patrol routes for patrol vehicles based on the deep Q-network algorithm, giving priority to covering high-risk sections and improving monitoring efficiency. The specific process is as follows:

[0042] The highway network is abstracted into a directed graph G = (K, L), where K is a node set representing each monitoring point on the highway, and L is an edge set representing the road sections connecting the nodes. The state e is set to consist of the real-time risk level of the current patrol car's location, the traffic flow of the road section, and the speed limit. The action d is set to indicate that the patrol car selects the next adjacent road section to travel from the current location. If the current patrol car is at node k, l , then the set of adjacent nodes is N(k l ), then the action space D={d1,d2......d g}, where g is the number of adjacent nodes, and each action d corresponds to l Drive to N(k l ) a node in

[0043] A reward function M is introduced into the deep Q-network algorithm, which prioritizes high-risk sections, reduces patrol time, and avoids repeated patrols. When high-risk sections > medium-risk sections > low-risk sections > 0, patrol cars are encouraged to prioritize high-risk sections. When patrol cars increase their patrol time, they receive negative rewards, prompting them to complete their patrol missions as quickly as possible. When patrol cars repeat patrolled sections, they receive negative rewards, thereby avoiding repeated patrol routes and optimizing patrol paths.

[0044] The deep Q network algorithm is trained. The patrol car continuously performs actions in the environment, obtains rewards according to the reward function, and converts the state e, action d, reward M and next state e into * The data is stored in the experience replay pool. Every certain driving distance, a batch of samples are randomly drawn from the experience replay pool for training. The target Q value is calculated. The calculation logic is as follows: in is a discount factor used to balance current rewards and future rewards, ψ - is the mean square error between the target network parameters and the current Q network; the stochastic gradient descent algorithm U is used to update the Q network parameters and continuously optimize the decision-making strategy of the intelligent agent. Its calculation logic is:

[0045] After the algorithm training is completed, the patrol car selects the action d that maximizes the Q(e,d) value in the action space based on the current state e, determines the next driving section, and so on. The real-time optimal patrol path is generated and sent to the software platform. The calculation logic is as follows:

[0046] The above is an illustration of the present invention and should not be considered as limiting thereof. Although several exemplary embodiments of the present invention have been described, it will be readily understood by those skilled in the art that many modifications may be made to the exemplary embodiments without departing from the novel teachings and advantages of the present invention. Therefore, all such modifications are intended to be included within the scope of the present invention as defined by the claims. It should be understood that the above is an illustration of the present invention and should not be considered as being limited to the specific embodiments disclosed, and modifications to the disclosed embodiments and other embodiments are intended to be included within the scope of the appended claims. The present invention is defined by the claims and their equivalents.

Claims

1. The highway road condition and environmental safety early warning monitoring system includes a data fusion module, an adaptive risk assessment module, a path optimization module, and platform software, and is characterized by: The deep Q network algorithm is used to generate a real-time optimal patrol path for the patrol car, giving priority to covering high-risk sections. The specific process is as follows: the deep Q network algorithm is trained, the patrol car continuously performs actions in the environment, obtains rewards according to the reward function, and calculates the state e, action d, reward M and the next state e. * The data is stored in the experience replay pool. Every time a certain driving distance is traveled, a batch of samples are randomly drawn from the experience replay pool for training. The target Q value is calculated. The calculation logic is as follows: in is the discount factor, ψ - is the mean square error between the target network parameters and the current Q network; the stochastic gradient descent algorithm U is used to update the Q network parameters and continuously optimize the decision-making strategy of the intelligent agent. Its calculation logic is: After the algorithm training is completed, the patrol car selects the action d that maximizes the Q(e,d) value in the action space based on the current state e, determines the next driving section, and so on. The real-time optimal patrol path is generated and sent to the software platform. The calculation logic is as follows:

2. The highway road condition and environmental safety early warning monitoring system according to claim 1 is characterized in that: The path optimization module sets the state, action and reward, which are specifically as follows: The highway network is abstracted into a directed graph G = (K, L), where K is the node set and L is the edge set; the state e is set to consist of the real-time risk level, traffic flow and speed limit of the current patrol car's location; the action d is set to indicate that the patrol car selects the next adjacent road section to travel from the current location. If the current patrol car is at node k l , then the set of adjacent nodes is N(k l ), then the action space D={d1,d2......d g }, where g is the number of adjacent nodes, and each action d corresponds to l Drive to N(k l ) a node in A reward function M is introduced into the deep Q-network algorithm to prioritize covering high-risk sections. When high-risk sections > medium-risk sections > low-risk sections > 0, patrol cars are encouraged to prioritize high-risk sections. When a patrol car increases its patrol time, it receives a negative reward; when a patrol car repeatedly passes through a patrolled section, it receives a negative reward.

3. The highway road condition and environmental safety early warning monitoring system according to claim 1 is characterized in that: The adaptive risk assessment module extracts the feature set and processes the data, specifically: Historical accident data is obtained from the platform software. Historical accident characteristics and road surface condition characteristics are extracted from the historical accident data and real-time parameters to form a feature set. Historical accident characteristics include the frequency, severity and type of historical accidents on the road section. Road surface condition characteristics include the real-time friction coefficient, water film height, road ice content, road surface temperature and relative humidity. The collected data is cleaned to remove outliers and missing values. Normalization is used to convert the data to the same dimension.

4. The highway road condition and environmental safety early warning monitoring system according to claim 3 is characterized in that: The adaptive risk assessment module builds a dynamic risk assessment model based on the random forest algorithm to calculate the risk level of highway sections in real time. Specifically: Construct a random forest model. The random forest algorithm consists of multiple decision trees. The random forest model constructed in this module contains 100 decision trees. During the training process, each decision tree randomly selects some features from the feature set. The number of features selected is the square root of the total number of features. Sub-datasets are generated by random sampling with replacement from the training dataset. For each decision tree, based on the selected sub-dataset and feature set, the Gini index is used as the splitting criterion to build a decision tree model. The Gini index is used to measure the impurity of the data. Its calculation logic is: Where m represents the total number of risk level categories, j represents the number of risk level categories, δ j is the probability value of the sample belonging to the jth class; in the process of splitting the nodes of the decision tree, the feature and split point that causes the Gini index to drop the most are selected until the depth of the tree reaches the preset threshold of the platform software; When a new batch of real-time environmental parameters and historical accident feature sets are input, each decision tree will make a prediction and output the corresponding risk level prediction result. The random forest model uses a voting mechanism to count the prediction results of all decision trees and use the result of the majority vote as the final risk level prediction; Then calculate the frequency of each risk level in all decision tree prediction results and obtain the risk probability value; The 100 decision trees are divided into r1 and r2, respectively, with low risk, medium risk, and high risk, respectively. The risk probability calculation logic is: and Obtain low risk probability value δ1, medium risk probability value δ2 and high risk probability value δ3; According to the calculated risk probability value, the road section risk level is divided into three levels: low, medium and high. The specific classification rules are as follows: when δ1 ≥ 0.6, it is judged as a low risk level; when 0.3 < δ2 is less than 0.6, it is judged as a medium risk level; when δ3 ≥ 0.3, it is judged as a high risk level; The corresponding road section risk level is sent to the software platform and path optimization module.

5. The highway road condition and environmental safety early warning monitoring system according to claim 1 is characterized in that: The data fusion module integrates RTK positioning data, road surface temperature, humidity and friction coefficient multi-dimensional information, specifically: The data fusion module is equipped with the patrol car's front-end acquisition equipment. The various sensors of the front-end acquisition equipment collect real-time data on the highway's RTK positioning, road surface temperature, humidity, and friction coefficient. These data are marked as (x, y), T, H, μ, and S, respectively, where S represents the timestamp. This is used to construct a 6-dimensional state vector fusion model Zf = [xy TH μ s] T .

6. The highway road condition and environmental safety early warning monitoring system according to claim 5 is characterized in that: The data fusion module dynamically adjusts the weight based on the sensor confidence. The specific process is as follows: Set the sensor dynamic weight w i ∈[0,1], its calculation logic is: Among them C i The initial calibration value of the sensor, σ is expressed as the data fluctuation coefficient, E i is the environmental adaptability coefficient, i represents the sensor number, and n represents the total number of sensors.

7. The highway road condition and environmental safety early warning monitoring system according to claim 6 is characterized in that: The data fusion module is based on an improved weighted Kalman filter algorithm, specifically: Establish the state transfer equation, and its calculation logic is: Z f | f-1 =P f Z f-1 | f-1 +C f +E f , where Z f | f-1 is the state prediction value at time f based on the state estimation at time f-1, P f is the state transfer matrix, C f is the control input matrix, E f is the process noise; linearize it using first-order Taylor expansion: Where β represents the sensitivity coefficient of humidity to temperature change, which is calibrated by historical data; process noise E f According to the dynamic adjustment of the patrol car's motion state, the calculation logic is: E f =E base +η·α 2 , where α is the vehicle acceleration, E base is the baseline value of process noise, η is the noise amplification factor; The Kalman gain is calculated based on the observation equation and predicted state of each sensor to obtain the fused state estimate. The calculation logic is as follows: Among them, θ f|f-1 is the prediction covariance, is the Kalman gain of the i-th sensor at time f, BP i,f The observation matrix of the i-th sensor, is the observation noise of i sensors; the prediction covariance θ f|f-1 The calculation logic is: θ f-1|f-1 Expressed as the covariance matrix of the optimal state estimate at time f-1; the observation matrix BP i,f It maps the 6-dimensional state vector to the observation space, expressed as: Temperature and humidity are directly observed, and the friction coefficient is indirectly estimated through other parameters; noise is observed is based on the sensor confidence w i Dynamic weighting.

Images