Expressway safety control method and system based on control unit

Abstract

The invention provides an expressway safety control method and system based on a control unit, and an efficient data processing and risk assessment system is constructed by carefully selecting driving road condition data and risk factor knowledge blocks from the control unit of a sample expressway. Risk description label prediction is performed on sample data by using a reference neural network, and the accuracy and reliability of a risk prediction model are ensured through network training convergence verification. According to the method, the corresponding risk description label can be quickly and accurately predicted according to the input actual driving road condition data of the expressway target management and control unit, and particularly, the matching of a plurality of local risk description labels can be refined, so that a more refined risk assessment result is provided. The management and control suggestions generated based on the risk description label prediction results provide a scientific basis for the safety management of the expressway, are helpful to timely identify and deal with potential risks, and significantly improve the safety management and control level and emergency response capability of the expressway.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a highway safety control method and system based on a control unit. Background Art

[0002] As an important part of the modern transportation network, highway safety management is of great significance to ensuring driving safety and improving traffic efficiency. However, the traffic conditions on highways are complex and changeable, involving multiple factors such as vehicle speed, traffic volume, weather conditions, road conditions, etc., which makes safety management face huge challenges. Traditional highway safety management methods often rely on manual monitoring and experience judgment, which makes it difficult to fully and accurately identify potential risks and make timely and effective responses. Summary of the invention

[0003] In view of the above-mentioned problems, in combination with the first aspect of the present invention, an embodiment of the present invention provides a highway safety control method based on a control unit, the method comprising:

[0004] Obtain x sample traffic condition data and Y risk factor knowledge blocks from the sample traffic condition data sequence of each sample control unit of the sample highway; the sample traffic condition data sequence includes X sample traffic condition data and the Y risk factor knowledge blocks, the Y risk factor knowledge blocks correspond one-to-one to the Y local risk description labels of the target risk description label, and the risk factor knowledge blocks are used to represent the corresponding local risk description labels with knowledge vectors; x, X and Y are positive integers, and x is less than X;

[0005] The x sample driving road condition data and the Y risk factor knowledge blocks are loaded into a reference neural network for risk description label prediction, and a risk description label prediction result for each sample driving road condition data is generated; the reference neural network is generated by fine-tuning parameters based on an initialized neural network, and the initialized neural network is generated by optimizing network parameters based on a plurality of training data combinations, and the training data combination is a combination of training driving road condition data and training risk factor knowledge blocks;

[0006] If it is determined based on the risk description label prediction result that the network training convergence requirement is met, obtaining a risk description label prediction network from the reference neural network;

[0007] The target driving road condition data of the target control unit of the input highway is obtained, and the risk description label prediction network is used to predict the risk description label of the target driving road condition data to generate the risk description label prediction result of the target driving road condition data, and the control suggestion of the target control unit is generated according to the risk description label prediction result; the risk description label prediction result is used to reflect the target local risk description label that matches the target driving road condition data among the multiple local risk description labels included in the target risk description label.

[0008] In a possible implementation of the first aspect, the step of loading the x sample driving road condition data and the Y risk factor knowledge blocks into a reference neural network for risk description label prediction and generating a risk description label prediction result for each sample driving road condition data includes:

[0009] Loading the x sample driving road condition data and the Y risk factor knowledge blocks into a reference neural network, using the reference neural network to encode and represent the x sample driving road condition data and the Y risk factor knowledge blocks, and generating a target driving road condition feature vector of the x sample driving road condition data and a target risk factor knowledge vector of the Y risk factor knowledge blocks;

[0010] The reference neural network is used to predict risk description labels based on the target driving road condition feature vector and the target risk factor knowledge vector, thereby generating risk description label prediction results for each of the sample driving road condition data.

[0011] In a possible implementation of the first aspect, the reference neural network includes a driving condition knowledge representation unit and a risk factor knowledge representation unit, and the x sample driving condition data and the Y risk factor knowledge blocks are loaded into the reference neural network, and the x sample driving condition data and the Y risk factor knowledge blocks are encoded and represented by using the reference neural network to generate a target driving condition feature vector of the x sample driving condition data and a target risk factor knowledge vector of the Y risk factor knowledge blocks, including:

[0012] Loading the x sample driving road condition data into the driving road condition knowledge representation unit, encoding and representing the x sample driving road condition data using the driving road condition knowledge representation unit, and generating target driving road condition feature vectors of the x sample driving road condition data;

[0013] The Y risk factor knowledge blocks are loaded into the risk factor knowledge representation unit, and the risk factor knowledge representation unit is used to encode and represent the Y risk factor knowledge blocks to generate target risk factor knowledge vectors of the Y risk factor knowledge blocks.

[0014] In a possible implementation of the first aspect, the initialized neural network includes a driving road condition static extraction layer, and the driving road condition knowledge representation unit includes the driving road condition static extraction layer and the driving road condition dynamic extraction layer;

[0015] The step of encoding and representing the x sample driving road condition data using the driving road condition knowledge representation unit to generate a target driving road condition feature vector of the x sample driving road condition data includes:

[0016] Using the traffic condition static extraction layer to extract features from the x sample traffic condition data to generate a first traffic condition feature vector;

[0017] Using the traffic condition dynamic extraction layer to extract features from the x sample traffic condition data to generate a second traffic condition feature vector;

[0018] The first driving road condition feature vector and the second driving road condition feature vector are merged to generate a target driving road condition feature vector of the x sample driving road condition data.

[0019] In a possible implementation of the first aspect, the extracting features from the x sample driving condition data includes:

[0020] Decomposing each of the sample driving road condition data to generate K sample driving road condition segments for each of the sample driving road condition data; K is a positive integer;

[0021] Performing a fusion calculation on each of the sample driving road condition segments and the set conversion vector to generate a segment vector of each of the sample driving road condition segments;

[0022] Based on each of the segment vectors, a corresponding driving road condition feature vector is generated.

[0023] In a possible implementation manner of the first aspect, encoding and representing the Y risk factor knowledge blocks using the risk factor knowledge representation unit to generate target risk factor knowledge vectors of the Y risk factor knowledge blocks includes:

[0024] Using natural language processing algorithms to perform semantic analysis on each risk factor knowledge block to extract corresponding key semantic information, including risk type, risk level, risk occurrence conditions, risk impact scope, and historical risk event records;

[0025] Classifying the extracted key semantic information, and assigning corresponding numerical codes to the key semantic information according to the classification results and preset coding rules, and generating a dictionary containing all the key semantic information and corresponding numerical codes;

[0026] For each type of key semantic information, construct a corresponding feature mapping rule, wherein the feature mapping rule defines a rule for converting the key semantic information into a numerical feature;

[0027] According to the constructed feature mapping rules, the eigenvalues ​​corresponding to each key semantic information are calculated, and the eigenvalues ​​are normalized to generate a risk factor feature vector matrix containing all eigenvalues ​​and corresponding normalized results;

[0028] Using a covariance analysis algorithm or an association rule mining algorithm to identify the association relationship between different features in the risk factor feature vector matrix, and based on the identified association relationship, constructing an associated risk factor feature matrix, wherein the associated features in the associated risk factor feature matrix are used to reflect the interaction and influence between different risk factors;

[0029] Using a recursive feature elimination algorithm or expert knowledge to evaluate the importance of each associated feature in the associated risk factor feature matrix, and assigning a corresponding feature weight to each associated feature according to the evaluation result, wherein the size of the feature weight reflects the contribution of the associated feature in risk prediction;

[0030] After weighting the associated risk factor features using the feature weights to generate a weighted risk factor feature vector, the weighted risk factor feature vector is fused according to the selected fusion strategy to generate a fused risk factor vector, and a corresponding standardization method is matched according to the distribution pattern of the fused risk factor vector, wherein the standardization method includes normalization, standardization or regularization;

[0031] The fused risk factor vector is standardized using a selected standardization method to generate a target risk factor knowledge vector of the Y risk factor knowledge blocks.

[0032] In a possible implementation of the first aspect, using the reference neural network to perform risk description label prediction according to the target driving road condition feature vector and the target risk factor knowledge vector to generate a risk description label prediction result for each sample driving road condition data includes:

[0033] Connecting the target driving road condition feature vector and the target risk factor knowledge vector to generate a connection feature vector;

[0034] Performing mutual correlation training based on the connection feature vector to generate mutual correlation training features;

[0035] For each of the sample traffic condition data, obtain the target inter-correlation training features of the sample traffic condition data for each of the local risk description labels from the inter-correlation training features, and calculate the similarity score between each of the target inter-correlation training features and the set label feature to generate a similarity score result;

[0036] Performing a standardized conversion on each of the similarity score results to generate a standardized conversion result, and determining the standardized conversion result as the confidence level of the sample driving road condition data matching the corresponding local risk description label;

[0037] Based on the confidence levels corresponding to the sample driving road condition data, a risk description label prediction result of the sample driving road condition data is generated.

[0038] In a possible implementation manner of the first aspect, the step of performing mutual correlation training based on the connection feature vector to generate mutual correlation training features includes:

[0039] According to the dimension information of the connection feature vector, the format information of the weight matrix required for the mutual correlation training is determined, wherein if the dimension of the connection feature vector is n-dimensional, a weight matrix with a shape of [n, m] is initialized, wherein m is a preset value related to the dimension of the desired mutual correlation training feature or the network structure, and at the same time, a bias vector matching the weight matrix is ​​initialized, the dimension of the bias vector is m-dimensional, and then the number of rounds of mutual correlation training and the learning rate are set, and the learning rate will be used to adjust the update step size of the weight matrix and the bias vector during the training process;

[0040] Taking the connection feature vector as the input data structure of the input layer, determining whether it is necessary to construct a hidden layer of the mutual correlation training infrastructure according to the format information of the weight matrix, and if it is necessary to construct a hidden layer, determining the number of neurons in the hidden layer, wherein the number of neurons is determined according to the dimension of the connection feature vector and the complexity of the desired output feature, and defining the activation function of the hidden layer of the mutual correlation training infrastructure and constructing an output layer, wherein the number of neurons in the output layer matches the dimension of the final desired mutual correlation training feature;

[0041] The connection feature vector is input into the input layer of the constructed mutual correlation training infrastructure to directly pass the connection feature vector to the next layer through the input layer, wherein, if there is a hidden layer, for each neuron in the hidden layer, the input connection feature vector is weighted summed with the corresponding weight in the weight matrix and the corresponding bias value in the bias vector is added to generate an intermediate result, and the intermediate result is nonlinearly transformed through an activation function to obtain the output of the hidden layer, and the output of the hidden layer is used as the input of the next layer to repeat the above weighted summation, bias value addition and activation function operations until it reaches the output layer, and the weighted summation and bias value addition operations are performed in the output layer to obtain a temporary output result, and the temporary output result is a preliminary result of the mutual correlation training calculated based on the current weight matrix and bias vector;

[0042] Input the temporary output result into a mean square error loss function, calculate the average of the squares of the difference between each element of the temporary output result and a preset target cross-correlation training feature, and output the corresponding loss function value;

[0043] According to the calculated loss function value, starting from the output layer, reversely calculate the contribution of each weight and bias to the loss function value, wherein, for the output layer, use the derivation rule to calculate the derivative of the loss function value with respect to the weight and bias of the output layer, the derivative reflects the influence of the change of the weight and bias on the loss function value, and then multiply the derivative by a preset learning rate to obtain the updated amount of the output layer weight and bias, and then, for the hidden layer, multiply the error information transmitted by the output layer by the derivative of the activation function of the hidden layer to obtain the updated amount of the hidden layer weight and bias, thereby obtaining the updated amount of the weight and bias of each layer;

[0044] Using the calculated weight and bias update amounts, the current weight matrix and bias vector are updated, wherein for the weight matrix, the corresponding weight update amount is subtracted from each element; for the bias vector, the corresponding bias update amount is subtracted from each element, thereby obtaining the updated weight matrix and bias vector;

[0045] Check whether the number of completed training rounds has reached the preset number of training rounds. If it has reached the preset number of training rounds, it means that the mutual correlation training is completed; otherwise, it means that the training is not completed and the next round of training needs to be continued;

[0046] After the mutual correlation training is completed, the connection feature vector is input again into the trained mutual correlation training basic structure to obtain the mutual correlation training feature, which reflects the mutual correlation relationship between the internal elements of the connection feature vector.

[0047] In a possible implementation of the first aspect, the reference neural network further includes a feature fusion unit; and obtaining a risk description label prediction network from the reference neural network includes:

[0048] The current target risk factor knowledge vector is associated with the current driving road condition knowledge representation unit and the current feature fusion unit to generate a risk description label prediction network.

[0049] On the other hand, an embodiment of the present invention also provides a highway safety control system based on a control unit, including a processor and a machine-readable storage medium, wherein the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0050] Based on the above aspects, the embodiment of the present application constructs an efficient data processing and risk assessment system by selecting driving road condition data and risk factor knowledge blocks from the control units of the sample highway. The reference neural network is used to predict the risk description labels of the sample data, and the network training convergence verification is performed to ensure the accuracy and reliability of the risk prediction model. This method can quickly and accurately predict the corresponding risk description labels for the actual driving road condition data of the input highway target control unit, and in particular can be refined to the matching of multiple local risk description labels, thereby providing a more refined risk assessment result. The control recommendations generated based on these risk description label prediction results provide a scientific basis for the safety management of highways, help to timely identify and respond to potential risks, and significantly improve the safety management level and emergency response capabilities of highways. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a schematic diagram of the execution flow of the highway safety control method based on the control unit provided in an embodiment of the present invention.

[0052] Figure 2 It is a schematic diagram of the hardware architecture of a highway safety management and control system based on a management and control unit provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 It is a flow chart of a highway safety control method based on a control unit provided by an embodiment of the present invention. The highway safety control method based on the control unit is introduced in detail below.

[0054] Step S110, obtaining x sample traffic condition data and Y risk factor knowledge blocks from the sample traffic condition data sequence of each sample control unit of the sample highway. The sample traffic condition data sequence includes X sample traffic condition data and the Y risk factor knowledge blocks, the Y risk factor knowledge blocks correspond one-to-one to the Y local risk description labels of the target risk description label, and the risk factor knowledge blocks are used to represent the corresponding local risk description labels with knowledge vectors. x, X and Y are positive integers, and x is less than X.

[0055] In this embodiment, the sample highway is divided into multiple sample control units, and each sample control unit has a corresponding sample driving road condition data sequence. For example, the sample driving road condition data sequence of sample control unit 1 contains the driving road condition information every hour in the past period of time (such as the past year), and the driving road condition information here includes multiple factors such as vehicle speed, traffic volume, weather conditions (such as whether it is raining, snowing, foggy, etc.), road conditions (such as whether there is construction, road damage, etc.) and traffic accidents. Set X to 1000, that is, this data sequence contains 1000 sample driving road condition data. At the same time, the target risk description label contains multiple local risk description labels, such as "driving risk in bad weather", "driving risk on construction sections", "driving risk during high traffic flow periods", etc. Assume that Y is 5 here, corresponding to 5 local risk description labels, there are also 5 risk factor knowledge blocks. Set x to 100, and the server obtains 100 sample driving road condition data and the five risk factor knowledge blocks from the sample driving road condition data sequence of the sample control unit 1. The five risk factor knowledge blocks are pre-constructed. For example, the risk factor knowledge block corresponding to the local risk description label "driving risk in bad weather" contains the historical driving accident rate in bad weather (such as heavy rain, heavy snow, dense fog, etc.), the difficulty of maneuvering different types of vehicles (such as cars, trucks, buses, etc.) in bad weather, and the relationship between road visibility and vehicle speed in bad weather. These contents exist in a specific knowledge vector representation so that the subsequent neural network can process them.

[0056] Step S120, loading the x sample driving road condition data and the Y risk factor knowledge blocks into the reference neural network for risk description label prediction, and generating risk description label prediction results for each sample driving road condition data. The reference neural network is generated by fine-tuning parameters based on the initialization neural network, and the initialization neural network is generated by optimizing network parameters based on a plurality of training data combinations, and the training data combination is a combination of training driving road condition data and training risk factor knowledge blocks.

[0057] In this embodiment, the acquired 100 sample traffic condition data and 5 risk factor knowledge blocks can be loaded into the reference neural network for risk description label prediction. This reference neural network is generated through a series of complex processes. First, the initialization neural network is generated by optimizing the network parameters based on a large number of training data combinations. These training data combinations are composed of training traffic condition data and training risk factor knowledge blocks. For example, the training traffic condition data may come from highway driving records in multiple different regions and different time periods, and the training risk factor knowledge blocks are similar to the previously mentioned knowledge content constructed for different local risk description labels. Then the reference neural network is obtained by fine-tuning the parameters of the initialization neural network.

[0058] After the data is loaded into the reference neural network, the traffic condition knowledge representation unit and the risk factor knowledge representation unit in the reference neural network start working. Taking the traffic condition knowledge representation unit as an example, it is assumed that the traffic condition static extraction layer extracts the features of the vehicle speed information in the sample traffic condition data. If the vehicle speed data is a continuous numerical sequence, it may extract some static features, such as average speed, maximum speed, minimum speed, etc., to generate the first traffic condition feature vector. The traffic condition dynamic extraction layer will consider the change of vehicle speed over time, such as a sharp change in vehicle speed within a certain period of time, which may indicate an emergency (such as a traffic accident or temporary road control). Through this dynamic analysis, the second traffic condition feature vector is generated. Finally, the two feature vectors are fused to obtain the target traffic condition feature vector of 100 sample traffic condition data.

[0059] For the risk factor knowledge representation unit, take the risk factor knowledge block corresponding to "driving risk on construction road section" as an example. Use natural language processing algorithm to perform semantic analysis on this risk factor knowledge block and extract key semantic information, such as the risk type is driving risk caused by construction, the risk level may be divided into high, medium and low levels according to the construction scale and the importance of the road section, the risk occurrence condition is during the construction period and no reasonable warning signs are set or the road narrows, the risk impact range is the construction road section and a certain distance before and after it, and the historical risk event records include the number and type of traffic accidents that occurred in similar construction roads in the past. These key semantic information are classified and assigned corresponding numerical codes according to the preset coding rules. For example, the risk type may be coded as 1, and the high risk level is coded as 3, etc., and a dictionary containing all key semantic information and corresponding numerical codes is generated. Then, for each type of key semantic information, a feature mapping rule is constructed. For example, for the risk level, high, medium and low levels are mapped to a specific numerical range. According to this rule, the eigenvalue corresponding to each key semantic information is calculated and normalized to generate a risk factor feature vector matrix containing all eigenvalues ​​and corresponding normalized results. Then, the covariance analysis algorithm is used to identify the correlation between different features in the matrix and construct the correlation risk factor feature matrix. For example, it is found that there is a certain correlation between the risk level and the risk impact range. If the risk level is high, the risk impact range may be larger. The recursive feature elimination algorithm is then used to evaluate the importance of each correlation feature, and a feature weight is assigned to each correlation feature based on the evaluation results. For example, the feature weight of the risk level is 0.4, and the feature weight of the risk impact range is 0.3. These feature weights are used to weight the correlation risk factor features to generate a weighted risk factor feature vector. The weighted risk factor feature vector is then fused according to a fusion strategy (such as weighted average) to generate a fused risk factor vector. Finally, the fused risk factor vector is standardized according to the distribution pattern matching standardization method (such as normalization) to obtain the target risk factor knowledge vector of the five risk factor knowledge blocks.

[0060] After that, the target driving condition feature vector and the target risk factor knowledge vector are connected with reference to the neural network to generate a connection feature vector. Inter-correlation training is performed based on the connection feature vector. For example, based on the dimension information of the connection feature vector (assuming it is n-dimensional), the format information of the weight matrix required for the inter-correlation training is determined to be [n, m] (m is determined according to the dimension of the expected inter-correlation training feature, such as 100), and an m-dimensional bias vector is initialized at the same time, and the number of rounds of inter-correlation training is set to 100 rounds, and the learning rate is 0.01. The connection feature vector is used as the input data structure of the input layer. If a hidden layer needs to be constructed (determined according to the dimension of the connection feature vector and the complexity of the expected output feature), assuming that the number of neurons in the hidden layer is 50, the activation function of the hidden layer (such as the ReLU function) is defined and the output layer is constructed (the number of neurons matches the dimension of the final expected inter-correlation training feature, assuming it is 30). The connection feature vector is input into the input layer of the constructed inter-correlation training infrastructure and passed to the next layer through the input layer. If there is a hidden layer, for each neuron in the hidden layer, perform a weighted sum operation on the input connection feature vector and the corresponding weight in the weight matrix and add the corresponding bias value in the bias vector to generate an intermediate result, and then perform a nonlinear transformation through the activation function to obtain the output of the hidden layer. This output is used as the input of the next layer and the above operation is repeated until it reaches the output layer. Perform weighted summation and bias value addition operations in the output layer to obtain a temporary output result, input the temporary output result into the mean square error loss function, calculate the average of the square of the difference between each element and the preset target cross-correlation training feature, and output the loss function value. According to the loss function value, reversely calculate the contribution of each weight and bias to the loss function value starting from the output layer, and update the weight matrix and bias vector. Check whether the number of training rounds reaches 100 rounds. If so, the cross-correlation training is completed and the cross-correlation training feature is obtained. For each sample driving road condition data, the target inter-correlation training features of the sample driving road condition data for each local risk description label are obtained from the inter-correlation training features, and the similarity score between each target inter-correlation training feature and the set label feature is calculated, for example, using the cosine similarity calculation method to generate a similarity score result. Each similarity score result is standardized and converted to the confidence level of the sample driving road condition data matching the corresponding local risk description label, and the risk description label prediction result of each sample driving road condition data is generated based on these confidence levels.

[0061] Step S130: If it is determined based on the risk description label prediction result that the network training convergence requirement is met, a risk description label prediction network is obtained from the reference neural network.

[0062] Assume that in the previous risk description label prediction results, for each sample driving road condition data, the confidence of matching each local risk description label gradually tends to be stable, and the overall error rate (such as mean square error) has been reduced to below a pre-set threshold, which meets the network training convergence requirements. For example, the mean square error threshold is set to 0.05, and after multiple iterations of training, the calculated mean square error is 0.03. At this time, the server obtains the risk description label prediction network from the reference neural network. In this process, the current target risk factor knowledge vector is associated with the current driving road condition knowledge representation unit and the current feature fusion unit to generate a risk description label prediction network. This risk description label prediction network has the ability to predict risk description labels for new driving road condition data.

[0063] Step S140, obtaining the target traffic condition data of the target control unit of the input highway, predicting the risk description label of the target traffic condition data through the risk description label prediction network, generating the risk description label prediction result of the target traffic condition data, and generating the control suggestion of the target control unit according to the risk description label prediction result. The risk description label prediction result is used to reflect the target local risk description label that the target traffic condition data matches among the multiple local risk description labels included in the target risk description label.

[0064] When there is input target traffic condition data of the target control unit of the highway, for example, the real-time traffic condition data of a specific section (target control unit) of this highway, including the current vehicle speed, traffic volume, weather conditions (it is raining lightly, the road is a bit slippery) and road construction conditions (there is a small-scale construction) and other information. The server predicts the risk description labels of these target traffic condition data through the risk description label prediction network that has been generated. First, the target traffic condition data is processed in the risk description label prediction network similar to the previous steps to obtain the risk description label prediction result of the target traffic condition data. Assume that this risk description label prediction result shows that the target traffic condition data has a higher matching confidence (such as 0.8) under the local risk description label of "driving risk in bad weather" in the target risk description label, and the confidence under other local risk description labels is lower. The control suggestions of the target control unit are generated according to this risk description label prediction result. For situations where the confidence level of "driving risk in bad weather" is high, management and control suggestions may include adding warning signs on the road section to remind drivers to slow down and adjust the speed limit appropriately according to the traffic flow. This completes the entire process from obtaining driving road condition data to predicting risk description labels and then generating management and control suggestions, which helps improve driving safety and management efficiency on highways.

[0065] Based on the above steps, the embodiment of the present application constructs an efficient data processing and risk assessment system by selecting driving road condition data and risk factor knowledge blocks from the control units of the sample highway. The reference neural network is used to predict the risk description labels of the sample data, and the network training convergence verification is performed to ensure the accuracy and reliability of the risk prediction model. This method can quickly and accurately predict the corresponding risk description labels for the actual driving road condition data of the input highway target control unit, and in particular can be refined to the matching of multiple local risk description labels, thereby providing a more refined risk assessment result. The control recommendations generated based on these risk description label prediction results provide a scientific basis for the safety management of highways, help to timely identify and respond to potential risks, and significantly improve the safety management level and emergency response capabilities of highways.

[0066] In a possible implementation, step S120 includes:

[0067] Step S121, load the x sample driving road condition data and the Y risk factor knowledge blocks into a reference neural network, use the reference neural network to encode the x sample driving road condition data and the Y risk factor knowledge blocks, and generate a target driving road condition feature vector of the x sample driving road condition data and a target risk factor knowledge vector of the Y risk factor knowledge blocks.

[0068] Step S122, using the reference neural network, predicting risk description labels based on the target driving road condition feature vector and the target risk factor knowledge vector, and generating risk description label prediction results for each of the sample driving road condition data.

[0069] In a possible implementation, the reference neural network includes a driving condition knowledge representation unit and a risk factor knowledge representation unit, and step S121 includes:

[0070] Step S1211: load the x sample driving road condition data into the driving road condition knowledge representation unit, use the driving road condition knowledge representation unit to encode and represent the x sample driving road condition data, and generate target driving road condition feature vectors of the x sample driving road condition data.

[0071] Step S1212: load the Y risk factor knowledge blocks into the risk factor knowledge representation unit, use the risk factor knowledge representation unit to encode and represent the Y risk factor knowledge blocks, and generate target risk factor knowledge vectors of the Y risk factor knowledge blocks.

[0072] In this embodiment, first, the server loads x sample driving road condition data and Y risk factor knowledge blocks into the reference neural network, and then uses the reference neural network to encode and represent these data, thereby generating target driving road condition feature vectors of x sample driving road condition data and target risk factor knowledge vectors of Y risk factor knowledge blocks. In this process, the reference neural network includes a driving road condition knowledge representation unit and a risk factor knowledge representation unit.

[0073] For the sample traffic condition data, the server loads x sample traffic condition data into the traffic condition knowledge representation unit, and uses the traffic condition knowledge representation unit to encode and represent these sample traffic condition data to generate the target traffic condition feature vector. For example, taking the sample traffic condition data in the sample control unit mentioned above as an example, it contains various information such as vehicle speed, traffic flow, weather conditions, road conditions, and traffic accident occurrence. The traffic condition static extraction layer in the traffic condition knowledge representation unit starts to work. For the information of vehicle speed, it will comprehensively analyze the vehicle speed data in all sample traffic condition data, calculate statistical features such as average vehicle speed, maximum vehicle speed, minimum vehicle speed, and standard deviation of vehicle speed, and these features are combined to initially form a feature representation. At the same time, the traffic condition dynamic extraction layer will pay attention to the change trend of vehicle speed over time, such as how the vehicle speed suddenly decreases from normal speed in several consecutive data points. This dynamic change feature is combined with the static feature. In addition, for traffic flow, it will analyze the peaks and troughs of traffic flow in different time periods, as well as the fluctuation amplitude of traffic flow. In terms of weather conditions, if it is raining, it will record relevant characteristics such as the intensity of rain and the duration of rain. Road conditions focus on characteristics such as the location and size of road damage areas. Traffic accidents will count the frequency, severity and location of accidents. By integrating, quantifying and encoding these many features, the target traffic condition feature vector for these x sample traffic condition data is finally generated.

[0074] For the risk factor knowledge block, the server loads Y risk factor knowledge blocks into the risk factor knowledge representation unit, and uses the risk factor knowledge representation unit to encode and represent these knowledge blocks to generate the target risk factor knowledge vector. Taking the risk factor knowledge block "driving risk in bad weather" mentioned above as an example, the risk factor knowledge representation unit uses the natural language processing algorithm to perform semantic analysis on it. This risk factor knowledge block contains many knowledge contents such as the driving accident rate in bad weather (rainstorm, snowstorm, dense fog, etc.) in history, the difficulty of driving different types of vehicles (sedans, trucks, buses, etc.) in bad weather, and the relationship between road visibility and vehicle speed in bad weather. During the semantic analysis process, it will identify the risk type of driving risk in bad weather. The risk level may be divided into high, medium, and low levels according to factors such as historical accident rates. The risk occurs when bad weather occurs. The risk impact range may be the entire road section affected by bad weather and its surrounding areas. The historical risk event record contains key semantic information such as the specific number of traffic accidents and accident types (such as rear-end collisions, skidding, etc.) that occurred in such bad weather in the past. Then, these key semantic information are classified and assigned corresponding numerical codes according to the preset coding rules. For example, the risk type of driving risk in bad weather may be coded as 1, and the high risk level is coded as 3, etc., so as to generate a dictionary containing all key semantic information and corresponding numerical codes. Then, for each type of key semantic information, a feature mapping rule is constructed. For example, for the risk level, high, medium and low levels are mapped to a specific numerical interval. According to this rule, the eigenvalue corresponding to each key semantic information is calculated and normalized to generate a risk factor feature vector matrix containing all eigenvalues ​​and corresponding normalized results. Then, the covariance analysis algorithm is used to identify the association between different features in the matrix. For example, it is found that there is an association between the risk level and the risk impact range. If the risk level is high, the risk impact range may be larger, so as to construct the associated risk factor feature matrix. Then, the recursive feature elimination algorithm is used to evaluate the importance of each associated feature, and a feature weight is assigned to each associated feature according to the evaluation result. For example, the feature weight of the risk level is 0.4, and the feature weight of the risk impact range is 0.3. These feature weights are used to weight the associated risk factor features to generate a weighted risk factor feature vector, which is then fused according to a fusion strategy (such as weighted average) to generate a fused risk factor vector. Finally, the fused risk factor vector is standardized according to a distribution pattern matching standardization method (such as normalization) to obtain the target risk factor knowledge vector of Y risk factor knowledge blocks.

[0075] After obtaining the target traffic condition feature vectors of x sample traffic condition data and the target risk factor knowledge vectors of Y risk factor knowledge blocks, the server uses the reference neural network to predict the risk description label based on these two vectors, and generates the risk description label prediction result for each sample traffic condition data. The server first connects the target traffic condition feature vector and the target risk factor knowledge vector to generate a connection feature vector. For example, if the target traffic condition feature vector is a vector with a dimension of n1, and the target risk factor knowledge vector is a vector with a dimension of n2, then the dimension of the connection feature vector is n1+n2. Then, mutual correlation training is performed based on the connection feature vector, and this mutual correlation training process is relatively complicated. According to the dimension information of the connection feature vector (assuming it is m-dimensional), the format information of the weight matrix required for mutual correlation training is determined to be [m, p] (p is determined according to the dimension of the expected mutual correlation training feature, such as 100), and a p-dimensional bias vector is initialized at the same time, and the number of rounds of mutual correlation training is set to 100 rounds, and the learning rate is 0.01. Take the connection feature vector as the input data structure of the input layer. If a hidden layer needs to be constructed (determined by the dimension of the connection feature vector and the complexity of the expected output feature), assume that the number of neurons in the hidden layer is 50, define the activation function of the hidden layer (such as the ReLU function) and construct the output layer (the number of neurons matches the dimension of the final expected cross-correlation training feature, assuming it is 30). Input the connection feature vector to the input layer of the constructed cross-correlation training infrastructure and pass it to the next layer through the input layer. If there is a hidden layer, for each neuron in the hidden layer, perform a weighted sum operation on the input connection feature vector and the corresponding weight in the weight matrix and add the corresponding bias value in the bias vector to generate an intermediate result, and then perform a nonlinear transformation through the activation function to obtain the output of the hidden layer. This output is used as the input of the next layer and the above operation is repeated until it reaches the output layer. Perform weighted summation and add bias value operations in the output layer to obtain a temporary output result, input the temporary output result into the mean square error loss function, calculate the average of the square of the difference between each element and the preset target cross-correlation training feature, and output the loss function value. According to the loss function value, reversely calculate the contribution of each weight and bias to the loss function value from the output layer, and update the weight and bias vectors. Check whether the number of training rounds reaches 100 rounds. If so, the mutual correlation training is completed and the mutual correlation training features are obtained. For each sample driving road condition data, the target mutual correlation training features of the sample driving road condition data for each local risk description label are obtained from the mutual correlation training features, and the similarity score between each target mutual correlation training feature and the set label feature is calculated, for example, using the cosine similarity calculation method to generate a similarity score result.Each similarity score result is standardized and converted to the confidence level of the sample driving condition data matching the corresponding local risk description label, and the risk description label prediction results of each sample driving condition data are generated based on these confidence levels. For example, if the confidence level of a sample driving condition data under the local risk description label "Risk of driving in bad weather" is 0.8, and the confidence levels under other local risk description labels are lower, then it can be considered that this sample driving condition data is more likely to match the local risk description label "Risk of driving in bad weather".

[0076] In a possible implementation, the initialized neural network includes a driving road condition static extraction layer, and the driving road condition knowledge representation unit includes the driving road condition static extraction layer and the driving road condition dynamic extraction layer.

[0077] Step S1211 includes:

[0078] The driving road condition static extraction layer is used to perform feature extraction on the x sample driving road condition data to generate a first driving road condition feature vector.

[0079] The driving road condition dynamic extraction layer is used to perform feature extraction on the x sample driving road condition data to generate a second driving road condition feature vector.

[0080] The first driving road condition feature vector and the second driving road condition feature vector are merged to generate a target driving road condition feature vector of the x sample driving road condition data.

[0081] In a possible implementation manner, extracting features from the x sample driving condition data includes:

[0082] Each of the sample traffic condition data is decomposed to generate K sample traffic condition segments of each of the sample traffic condition data, where K is a positive integer.

[0083] Each of the sample driving road condition segments is fused with a set conversion vector to generate a segment vector for each of the sample driving road condition segments.

[0084] Based on each of the segment vectors, a corresponding driving road condition feature vector is generated.

[0085] In this embodiment, taking the sample traffic condition data in the sample control unit mentioned above as an example, these data contain a variety of information such as vehicle speed, traffic flow, weather conditions, road conditions, and traffic accident occurrence. For the information of vehicle speed, the static extraction layer of traffic condition will conduct a comprehensive analysis of the vehicle speed data in the x sample traffic condition data. It may count statistical features such as the average vehicle speed, by adding all the vehicle speed data and dividing it by the number of data to get the average value; the highest vehicle speed, that is, finding the maximum value from all the vehicle speed data; the lowest vehicle speed, finding the minimum value; the standard deviation of the vehicle speed, calculating the discrete degree of the vehicle speed data through a specific mathematical formula. For traffic flow, statistics such as the average vehicle flow, the calculation method is similar to the average vehicle speed, and the peak value of the traffic flow, that is, the maximum value in the traffic flow data, will be counted. In terms of weather conditions, if it is rainy weather, the proportion of rainy weather in these x sample traffic condition data will be counted, the road condition will count features such as the average area of ​​the road damage area, and the traffic accident occurrence will count features such as the average frequency of accidents. By combining the features extracted from aspects such as vehicle speed, traffic volume, weather conditions, road conditions and traffic accident occurrence, the first driving road condition feature vector is preliminarily formed.

[0086] Next, taking these sample traffic condition data as an example, for vehicle speed, the dynamic traffic condition extraction layer will focus on the change of vehicle speed over time. Each sample traffic condition data can be decomposed, assuming that it is decomposed into K sample traffic condition segments, where K is a positive integer. For example, for a certain sample traffic condition data, if every 10 minutes is a segment in chronological order, if this data covers 100 minutes, then K is 10. For each sample traffic condition segment, it is fused with the set conversion vector to generate a segment vector for each sample traffic condition segment. The set conversion vector may be a vector predefined according to prior knowledge or a model, for example, this vector contains some weight information related to the change of vehicle speed. During the fusion calculation, the speed data in the sample traffic condition segment may be weighted summed with the corresponding weight in the set conversion vector. Then, the corresponding traffic condition feature vector is generated based on each segment vector. For the dynamic changes of vehicle speed, it may calculate the rate of change of vehicle speed between adjacent segments, which is obtained by subtracting the average speed of the previous segment from the average speed of the next segment and dividing it by the average speed of the previous segment; or the fluctuation amplitude of vehicle speed within a certain time window. For traffic flow, attention will be paid to the increase and decrease trend of traffic flow between different segments, such as calculating the difference in traffic flow between adjacent segments, and the fluctuation frequency of traffic flow in a specific time period. In terms of weather conditions, if the weather changes in different segments, such as from sunny to rainy, the frequency of such weather changes will be recorded. Road conditions will focus on features such as the expansion or reduction of road damage areas in different segments. Traffic accidents will count the distribution patterns of accidents in different segments. By extracting and quantifying these features, a second driving road condition feature vector is formed.

[0087] Finally, in the fusion process, multiple methods may be used, such as weighted fusion. Assuming that some features in the first traffic condition feature vector are considered to have a more important position in the overall risk assessment, they will be assigned higher weights, and the features in the second traffic condition feature vector will also be assigned corresponding weights according to their importance. Taking the vehicle speed feature as an example, if the average vehicle speed is considered to have an importance coefficient of 0.4 in the static feature, and the rate of change of the vehicle speed is considered to have an importance coefficient of 0.6 in the dynamic feature, then the two features will be weighted summed according to this weight when fused. Similar weighted fusion operations are also performed for features such as traffic flow, weather conditions, road conditions, and traffic accident occurrence. Through this comprehensive fusion method, all features in the first traffic condition feature vector and the second traffic condition feature vector are integrated together, and finally a target traffic condition feature vector of x sample traffic condition data is generated. This target traffic condition feature vector contains comprehensive feature information of the sample traffic condition data, which can provide accurate input information for subsequent risk description label prediction based on the reference neural network.

[0088] In a possible implementation, step S1212 includes:

[0089] Step S1212-1, using a natural language processing algorithm to perform semantic analysis on each risk factor knowledge block to extract corresponding key semantic information, wherein the key semantic information includes risk type, risk level, risk occurrence conditions, risk impact scope, and historical risk event records.

[0090] In this embodiment, taking the risk factor knowledge block "Risk of driving in bad weather" as an example, the natural language processing algorithm can accurately identify the risk type as the risk of driving in bad weather from the text description in the knowledge block. For the risk level, if the knowledge block mentions that the risk level is classified as high in this case based on historical accident data and weather severity, then the key semantic information that the risk level is high will be extracted. In terms of risk occurrence conditions, if the description points out that the risk occurs when bad weather conditions such as heavy rain, heavy snow, and dense fog occur and the visibility is lower than a certain value, then this is the corresponding risk occurrence condition. If the risk impact range is described as the entire road section affected by bad weather and the roads within a certain distance before and after it, this information will also be extracted. The historical risk event record is to sort out event information such as the number of accidents, accident types (such as rear-end collisions, skidding, etc.) that occurred in this bad weather in the past from the knowledge block. Such semantic parsing operations are performed on all Y risk factor knowledge blocks to obtain the key semantic information corresponding to each knowledge block.

[0091] Step S1212-2, classify the extracted key semantic information, and assign corresponding numerical codes to the key semantic information according to the classification results and preset coding rules, and generate a dictionary containing all key semantic information and corresponding numerical codes.

[0092] For example, for the risk type category, if there are multiple risk types, such as driving risks in bad weather and driving risks on construction sections, according to the preset coding rules, driving risks in bad weather may be coded as 1, and driving risks on construction sections may be coded as 2, etc. For risk levels, if they are divided into three levels: high, medium, and low, high risk levels may be coded as 3, medium risk levels as 2, and low risk levels as 1. If the risk occurrence conditions are classified according to factors such as whether they are easy to trigger risks, easy triggering conditions may be coded as 1, and difficult triggering conditions as 2, etc. The risk impact range is classified according to the size of the impact area. Large-scale impact may be coded as 3, medium-scale as 2, and small-scale as 1. Historical risk event records are classified according to the number of accidents. A large number of accidents is coded as 3, a medium number is 2, and a small number is 1. These encoded key semantic information are organized into dictionaries to facilitate subsequent operations.

[0093] Step S1212 - 3 : for each type of key semantic information, construct a corresponding feature mapping rule, wherein the feature mapping rule defines a rule for converting the key semantic information into a numerical feature.

[0094] Step S1212-4, according to the constructed feature mapping rules, calculate the eigenvalues ​​corresponding to each key semantic information, and normalize the eigenvalues ​​to generate a risk factor feature vector matrix including all eigenvalues ​​and corresponding normalized results.

[0095] Taking the risk level as an example, the constructed feature mapping rule may be to map the high risk level to the numerical interval [0.7, 1], the medium risk level to [0.4, 0.6], and the low risk level to [0, 0.3]. For the risk impact range, if the road length is used as the measurement standard, the large range impact may be mapped to [0.6, 1], the medium range to [0.3, 0.5], and the small range to [0, 0.2], etc. According to such feature mapping rules, the server calculates the eigenvalue corresponding to each key semantic information. For example, for a risk factor knowledge block with a high risk level, the eigenvalue calculated according to the mapping rule may be 0.8. After calculating the eigenvalues ​​for all key semantic information, these eigenvalues ​​are normalized to unify their values ​​into a specific interval, such as [0, 1], thereby generating a risk factor feature vector matrix containing all eigenvalues ​​and corresponding normalized results.

[0096] Step S1212-5, using a covariance analysis algorithm or an association rule mining algorithm to identify the association relationship between different features in the risk factor feature vector matrix, and based on the identified association relationship, construct an associated risk factor feature matrix, wherein the associated features in the associated risk factor feature matrix are used to reflect the interaction and influence between different risk factors.

[0097] In the risk factor feature vector matrix, there is a potential relationship between the feature values ​​corresponding to different key semantic information. For example, in the matrix corresponding to the risk factor knowledge block of "driving risk in bad weather", there may be a correlation between the risk level and the risk impact range. If the risk level is high, the risk impact range is often large. This correlation relationship can be quantified through the covariance analysis algorithm. The association rule mining algorithm may find that if there are many accidents in the historical risk event records, the risk level is often high. The correlation risk factor feature matrix is ​​constructed based on these identified correlation relationships. The correlation features in this matrix can reflect the interaction and influence between different risk factors. For example, the mutual influence relationship between the risk level and the risk impact range will be reflected accordingly in the matrix.

[0098] Step S1212-6, using a recursive feature elimination algorithm or expert knowledge to evaluate the importance of each associated feature in the associated risk factor feature matrix, and assigning a corresponding feature weight to each associated feature based on the evaluation result, wherein the size of the feature weight reflects the contribution of the associated feature in risk prediction.

[0099] For example, for the associated risk factor feature matrix of "driving risk in bad weather", if the recursive feature elimination algorithm analysis finds that the risk level is more important in risk prediction, it may be assigned a feature weight of 0.4, while the risk impact range is relatively less important and may be assigned a feature weight of 0.3. If the expert knowledge base is used for evaluation, experts may assign a higher weight to historical risk event records based on their experience that they are important for risk prediction in some cases. Based on these evaluation results, a reasonable feature weight is assigned to each associated feature.

[0100] Step S1212-7, using the feature weights to perform weighted processing on the associated risk factor features, after generating a weighted risk factor feature vector, the weighted risk factor feature vector is fused according to the selected fusion strategy to generate a fused risk factor vector, and a corresponding standardization method is matched according to the distribution pattern of the fused risk factor vector, and the standardization method includes normalization, standardization or regularization.

[0101] Next, the server uses the feature weights to weight the associated risk factor features to generate a weighted risk factor feature vector. For each associated feature in the associated risk factor feature matrix, its eigenvalue is multiplied by the corresponding feature weight, and then these weighted eigenvalues ​​are combined in a certain order to form a weighted risk factor feature vector. After that, according to the selected fusion strategy, the weighted risk factor feature vector is fused to generate a fused risk factor vector. If the selected fusion strategy is weighted average, then each element in the weighted risk factor feature vector is averaged according to the weight to obtain a fused risk factor vector. According to the distribution pattern of the fused risk factor vector, the corresponding standardization method is matched, and the standardization method includes normalization, standardization or regularization. If the distribution of the fused risk factor vector is relatively dispersed, the normalization method may be selected.

[0102] Step S1212-8: Standardize the fused risk factor vector using the selected standardization method to generate the target risk factor knowledge vector of the Y risk factor knowledge blocks.

[0103] For example, if the normalization method is selected, each element in the fused risk factor vector is calculated according to the normalization formula so that the elements in the vector meet specific requirements, such as the sum of all elements is 1 or the value range of the elements is within a specific interval, etc., and finally a target risk factor knowledge vector of Y risk factor knowledge blocks that meet the requirements is obtained. This target risk factor knowledge vector can accurately represent the characteristics of the risk factor knowledge blocks and provide effective input for subsequent risk description label prediction.

[0104] In a possible implementation, step S122 includes:

[0105] Step S1221, connecting the target driving road condition feature vector and the target risk factor knowledge vector to generate a connection feature vector.

[0106] Step S1222: performing mutual correlation training based on the connection feature vector to generate mutual correlation training features.

[0107] Step S1223, for each of the sample driving road condition data, obtain the target inter-correlation training features of the sample driving road condition data for each of the local risk description labels from the inter-correlation training features, and calculate the similarity score between each of the target inter-correlation training features and the set label features to generate a similarity score result.

[0108] Step S1224, performing a standardized transformation on each of the similarity score results to generate a standardized transformation result, and determining the standardized transformation result as the confidence level of the sample driving road condition data matching the corresponding local risk description label.

[0109] Step S1225: generating a risk description label prediction result of the sample driving road condition data based on the confidence levels corresponding to the sample driving road condition data.

[0110] In a possible implementation, step S1222 includes:

[0111] Step S1222-1, according to the dimension information of the connection feature vector, determine the format information of the weight matrix required for the mutual correlation training, wherein, if the dimension of the connection feature vector is n, initialize a weight matrix with a shape of [n, m], wherein m is a preset value related to the dimension of the desired mutual correlation training feature or the network structure, and at the same time, initialize a bias vector matching the weight matrix, wherein the dimension of the bias vector is m, and then set the number of rounds of mutual correlation training and the learning rate, wherein the learning rate will be used to adjust the update step size of the weight matrix and the bias vector during the training process.

[0112] Step S1222-2, using the connection feature vector as the input data structure of the input layer, determine whether it is necessary to construct a hidden layer of the mutual correlation training infrastructure according to the format information of the weight matrix, if it is necessary to construct a hidden layer, determine the number of neurons in the hidden layer, the number of neurons is determined according to the dimension of the connection feature vector and the complexity of the desired output feature, and define the activation function of the hidden layer of the mutual correlation training infrastructure and construct an output layer, the number of neurons in the output layer matches the dimension of the final desired mutual correlation training feature.

[0113] Step S1222-3, input the connection feature vector into the input layer of the constructed mutual correlation training infrastructure, so as to directly pass the connection feature vector to the next layer through the input layer, wherein, if there is a hidden layer, for each neuron in the hidden layer, perform a weighted sum operation on the input connection feature vector and the corresponding weight in the weight matrix and add the corresponding bias value in the bias vector to generate an intermediate result, and perform a nonlinear transformation on the intermediate result through an activation function to obtain the output of the hidden layer, and the output of the hidden layer is used as the input of the next layer to repeat the above-mentioned weighted summation, bias value addition and activation function operations until it reaches the output layer, and perform weighted summation and bias value addition operations in the output layer to obtain a temporary output result, and the temporary output result is a preliminary result of the mutual correlation training calculated based on the current weight matrix and bias vector.

[0114] Step S1222-4, input the temporary output result into the mean square error loss function, calculate the average of the square of the difference between each element of the temporary output result and the preset target cross-correlation training feature, and output the corresponding loss function value.

[0115] Step S1222-5, based on the calculated loss function value, starting from the output layer, reversely calculate the contribution of each weight and bias to the loss function value, wherein, for the output layer, use the derivation rule to calculate the derivative of the loss function value with respect to the output layer weight and bias, the derivative reflects the degree of influence of the change of weight and bias on the loss function value, and then multiply the derivative by a preset learning rate to obtain the updated amount of the output layer weight and bias, then, for the hidden layer, multiply the error information transmitted by the output layer by the derivative of the activation function of the hidden layer to obtain the updated amount of the hidden layer weight and bias, thereby obtaining the updated amount of the weight and bias of each layer.

[0116] Step S1222-6, using the calculated weight and bias update amounts, the current weight matrix and bias vector are updated, wherein for the weight matrix, each element is subtracted by the corresponding weight update amount. For the bias vector, each element is subtracted by the corresponding bias update amount, thereby obtaining the updated weight matrix and bias vector.

[0117] Step S1222-7, check whether the number of completed training rounds has reached the preset number of training rounds, if it has reached the preset number of training rounds, it means that the mutual correlation training is completed. Otherwise, it means that the training is not completed and the next round of training needs to be continued.

[0118] Step S1222-8, after the mutual correlation training is completed, the connection feature vector is input again into the trained mutual correlation training infrastructure to obtain the mutual correlation training feature, which reflects the mutual correlation relationship between the internal elements of the connection feature vector.

[0119] In this embodiment, first, the server connects the target road condition feature vector and the target risk factor knowledge vector to generate a connection feature vector. For example, if the target road condition feature vector is a vector with a dimension of p, which contains the characteristic values ​​of information such as vehicle speed, traffic flow, weather conditions, road conditions, and traffic accident occurrence, and the target risk factor knowledge vector is a vector with a dimension of q, which contains the characteristic values ​​of risk factors such as risk type, risk level, risk occurrence conditions, risk impact scope, and historical risk event records, then the dimension of the connection feature vector obtained by connecting these two vectors is p+q. This connection feature vector integrates information on both road conditions and risk factors, and provides comprehensive input for subsequent cross-correlation training.

[0120] Next, the server performs mutual correlation training based on the connection feature vector to generate mutual correlation training features. This process is relatively complex and involves multiple sub-steps. According to the dimension information of the connection feature vector, the format information of the weight matrix required for mutual correlation training is determined. Assuming that the dimension of the connection feature vector is n-dimensional, a weight matrix with a shape of [n, m] is initialized, where m is a pre-set value, which is related to the dimension of the desired mutual correlation training feature or the network structure. At the same time, a bias vector matching the weight matrix is ​​initialized, and its dimension is m-dimensional. For example, if n=100, m=50 is determined according to the network structure and the feature complexity of the desired output, then a weight matrix of [100, 50] and a 50-dimensional bias vector are initialized. Then set the number of rounds of mutual correlation training to T rounds (such as T=100) and the learning rate α (such as α=0.01). The learning rate will be used to adjust the update step size of the weight matrix and the bias vector during the training process.

[0121] After that, the connection feature vector is used as the input data structure of the input layer, and the format information of the weight matrix is ​​used to determine whether it is necessary to construct a hidden layer of the cross-correlation training infrastructure. If it is necessary to construct a hidden layer, the number of neurons in the hidden layer is determined according to the dimension of the connection feature vector and the complexity of the expected output features. For example, if the dimension of the connection feature vector is high and the expected output is more complex, the number of neurons in the hidden layer may be determined to be 30. At the same time, the activation function (such as the ReLU function) of the hidden layer of the cross-correlation training infrastructure is defined and the output layer is constructed. The number of neurons in the output layer matches the dimension of the final expected cross-correlation training feature (assuming it is 20).

[0122] Subsequently, the connection feature vector is input into the input layer of the constructed cross-correlation training infrastructure to directly pass the connection feature vector to the next layer through the input layer. If there is a hidden layer, for each neuron in the hidden layer, the input connection feature vector is weighted summed with the corresponding weight in the weight matrix and the corresponding bias value in the bias vector is added to generate an intermediate result. For example, for the i-th neuron in the hidden layer, let the connection feature vector be x, the corresponding weight in the weight matrix be w_i, and the bias value in the bias vector be b_i, then the intermediate result z_i=∑(x_j*w_i_j)+b_i (j represents the dimension index of the connection feature vector). Then the intermediate result is nonlinearly transformed through the activation function to obtain the output of the hidden layer. The output of the hidden layer is used as the input of the next layer to repeat the above weighted summation, bias value addition and activation function operations until it reaches the output layer. The weighted summation and bias value addition operations are performed in the output layer to obtain a temporary output result. This temporary output result is a preliminary result of the cross-correlation training calculated based on the current weight matrix and bias vector.

[0123] Then input the temporary output result into the mean square error loss function, calculate the average of the square of the difference between each element of the temporary output result and the preset target mutual correlation training feature, and output the corresponding loss function value. Assume that the temporary output result is y_pred, the target mutual correlation training feature is y_true, and the mean square error loss function is MSE = (1 / N) * ∑ ((y_pred_i-y_true_i)2) (N is the number of elements, i is the element index).

[0124] According to the calculated loss function value, starting from the output layer, reversely calculate the contribution of each weight and bias to the loss function value. For the output layer, the derivative of the loss function value to the output layer weight and bias is calculated using the derivation rule. The derivative reflects the degree of influence of the change of weight and bias on the loss function value. Then the derivative is multiplied by the preset learning rate to obtain the update amount of the output layer weight and bias. Next, for the hidden layer, the error information transmitted by the output layer is multiplied by the derivative of the activation function of the hidden layer to obtain the update amount of the hidden layer weight and bias, thereby obtaining the update amount of the weight and bias of each layer. For example, for the output layer weight w_out, let its derivative be dw_out, then the update amount Δw_out = α*dw_out, for the bias b_out, let its derivative be db_out, and the update amount Δb_out = α*db_out. For the hidden layer weight w_hidden and bias b_hidden, similar calculations are performed based on the error information transmitted from the output layer to obtain the update amount.

[0125] Use the calculated weight and bias update amounts to update the current weight matrix and bias vector. For the weight matrix, subtract the corresponding weight update amount from each element, and for the bias vector, subtract the corresponding bias update amount from each element, thereby obtaining the updated weight matrix and bias vector. Then check whether the number of completed training rounds has reached the pre-set number of training rounds. If the set number of training rounds (such as 100 rounds) has been reached, it means that the cross-correlation training is completed. Otherwise, it means that the training is not yet completed and the next round of training needs to be continued.

[0126] After the cross-correlation training is completed, the connection feature vector is input into the trained cross-correlation training infrastructure again to obtain the cross-correlation training feature. This cross-correlation training feature reflects the cross-correlation relationship between the internal elements of the connection feature vector. For example, for the relationship between the speed and the risk level in the connection feature vector, if it is found during the training process that the risk level tends to be higher when the speed is higher, then this relationship will be reflected in the cross-correlation training feature.

[0127] Next, for each sample driving road condition data, the target cross-correlation training features for each local risk description label of the sample driving road condition data are obtained from the cross-correlation training features, and the similarity score between each target cross-correlation training feature and the set label feature is calculated to generate a similarity score result. Taking a certain data in the sample driving road condition data as an example, it is assumed that the local risk description labels include "driving risk in bad weather", "driving risk in construction section", "driving risk in high traffic flow period", etc. For the local risk description label "driving risk in bad weather", the target cross-correlation training features related to the label are obtained from the cross-correlation training features, and then the similarity score between it and the set "driving risk in bad weather" label feature is calculated. The similarity score may be calculated using the cosine similarity calculation method. Let the target cross-correlation training feature be x and the set label feature be y, then the similarity score s = (x·y) / (||x||*||y||), where · represents the vector dot product, and ||x|| and ||y|| represent the modulus of vectors x and y respectively. This calculation is performed for each local risk description label to generate the corresponding similarity score result.

[0128] After that, each similarity score result is standardized and converted to generate a standardized conversion result, and the standardized conversion result is determined as the confidence of the sample driving road condition data matching the corresponding local risk description label. For example, the similarity score result is mapped to the interval [0, 1] using a normalization method. Assuming that the similarity score result of a sample driving road condition data to the label "Risk of driving in bad weather" is 0.6, and the standardized conversion result obtained after normalization is 0.6, this 0.6 is determined as the confidence of the sample driving road condition data matching the local risk description label "Risk of driving in bad weather".

[0129] Finally, based on the confidence levels corresponding to the sample driving road condition data, the risk description label prediction results of the sample driving road condition data are generated. For example, if the confidence level of a sample driving road condition data for "driving risk in bad weather" is 0.8, the confidence level of "driving risk on construction road sections" is 0.1, and the confidence level of "driving risk during high traffic flow periods" is 0.1, then it can be considered that the risk description label prediction results of this sample driving road condition data are more inclined to "driving risk in bad weather".

[0130] In a possible implementation, the reference neural network further includes a feature fusion unit.

[0131] Step S130 may include: associating the current target risk factor knowledge vector with the current driving road condition knowledge representation unit and the current feature fusion unit to generate a risk description label prediction network.

[0132] For example, the target risk factor knowledge vector contains information about various risk factors, the road condition knowledge representation unit contains feature information about road conditions, and the feature fusion unit has the ability to fuse different information. By associating them, the risk description label prediction network can make comprehensive use of this information to make accurate risk description label predictions. Specifically, it may be possible to fuse and associate information such as the risk level and risk impact range in the target risk factor knowledge vector with information such as vehicle speed and traffic volume in the road condition knowledge representation unit through the feature fusion unit according to certain rules, thereby constructing a risk description label prediction network, which can play a role in the subsequent risk description label prediction of new road condition data.

[0133] Figure 2 The hardware structure of the highway safety control system 100 based on the control unit provided in the embodiment of the present invention for implementing the above-mentioned highway safety control method based on the control unit is shown. Figure 2As shown, the highway safety management system 100 based on the management unit may include a processor 110 , a machine-readable storage medium 120 , a bus 130 , and a communication unit 140 .

[0134] The machine-readable storage medium 120 may store data and / or instructions. In some embodiments, the machine-readable storage medium 120 may store data obtained from an external terminal. In some embodiments, the machine-readable storage medium 120 may store data and / or instructions that the highway safety management system 100 based on the management unit uses to execute or use to complete the exemplary method described in the present invention.

[0135] During the specific implementation process, one or more processors 110 execute computer executable instructions stored in the machine-readable storage medium 120, so that the processor 110 can execute the highway safety control method based on the control unit in the above method embodiment. The processor 110, the machine-readable storage medium 120 and the communication unit 140 are connected through the bus 130, and the processor 110 can be used to control the sending and receiving actions of the communication unit 140.

[0136] The specific implementation process of the processor 110 can refer to the various method embodiments executed by the above-mentioned highway safety management and control system 100 based on the management and control unit. The implementation principles and technical effects are similar, and this embodiment will not be repeated here.

[0137] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer executable instructions are preset. When a processor executes the computer executable instructions, the above-mentioned highway safety control method based on the control unit is implemented.

[0138] It should be noted that in order to simplify the description of the present invention and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, various features are sometimes combined into one embodiment, drawing or description thereof.

Claims

1. A highway safety control method based on a control unit, characterized in that: The method comprises: Obtain x sample traffic condition data and Y risk factor knowledge blocks from the sample traffic condition data sequence of each sample control unit of the sample highway; the sample traffic condition data sequence includes X sample traffic condition data and the Y risk factor knowledge blocks, the Y risk factor knowledge blocks correspond one-to-one to the Y local risk description labels of the target risk description label, and the risk factor knowledge blocks are used to represent the corresponding local risk description labels with knowledge vectors; x, X and Y are positive integers, and x is less than X; The x sample driving road condition data and the Y risk factor knowledge blocks are loaded into a reference neural network for risk description label prediction, and a risk description label prediction result for each sample driving road condition data is generated; the reference neural network is generated by fine-tuning parameters based on an initialized neural network, and the initialized neural network is generated by optimizing network parameters based on a plurality of training data combinations, and the training data combination is a combination of training driving road condition data and training risk factor knowledge blocks; If it is determined based on the risk description label prediction result that the network training convergence requirement is met, obtaining a risk description label prediction network from the reference neural network; The target driving road condition data of the target control unit of the input highway is obtained, and the risk description label prediction network is used to predict the risk description label of the target driving road condition data to generate the risk description label prediction result of the target driving road condition data, and the control suggestion of the target control unit is generated according to the risk description label prediction result; the risk description label prediction result is used to reflect the target local risk description label that matches the target driving road condition data among the multiple local risk description labels included in the target risk description label.

2. The highway safety control method based on the control unit according to claim 1 is characterized in that: The step of loading the x sample traffic condition data and the Y risk factor knowledge blocks into a reference neural network for risk description label prediction, and generating a risk description label prediction result for each sample traffic condition data, includes: Loading the x sample driving road condition data and the Y risk factor knowledge blocks into a reference neural network, using the reference neural network to encode and represent the x sample driving road condition data and the Y risk factor knowledge blocks, and generating a target driving road condition feature vector of the x sample driving road condition data and a target risk factor knowledge vector of the Y risk factor knowledge blocks; The reference neural network is used to predict risk description labels based on the target driving road condition feature vector and the target risk factor knowledge vector, thereby generating risk description label prediction results for each of the sample driving road condition data.

3. The highway safety control method based on the control unit according to claim 2 is characterized in that: The reference neural network includes a traffic condition knowledge representation unit and a risk factor knowledge representation unit. The x sample traffic condition data and the Y risk factor knowledge blocks are loaded into the reference neural network, and the x sample traffic condition data and the Y risk factor knowledge blocks are encoded and represented by the reference neural network to generate a target traffic condition feature vector of the x sample traffic condition data and a target risk factor knowledge vector of the Y risk factor knowledge blocks, including: Loading the x sample driving road condition data into the driving road condition knowledge representation unit, encoding and representing the x sample driving road condition data using the driving road condition knowledge representation unit, and generating target driving road condition feature vectors of the x sample driving road condition data; The Y risk factor knowledge blocks are loaded into the risk factor knowledge representation unit, and the risk factor knowledge representation unit is used to encode and represent the Y risk factor knowledge blocks to generate target risk factor knowledge vectors of the Y risk factor knowledge blocks.

4. The highway safety control method based on the control unit according to claim 3 is characterized in that: The initialization neural network includes a static road condition extraction layer, and the road condition knowledge representation unit includes the static road condition extraction layer and the dynamic road condition extraction layer; The step of encoding and representing the x sample driving road condition data using the driving road condition knowledge representation unit to generate a target driving road condition feature vector of the x sample driving road condition data includes: Using the traffic condition static extraction layer to extract features from the x sample traffic condition data to generate a first traffic condition feature vector; Using the traffic condition dynamic extraction layer to extract features from the x sample traffic condition data to generate a second traffic condition feature vector; The first driving road condition feature vector and the second driving road condition feature vector are merged to generate a target driving road condition feature vector of the x sample driving road condition data.

5. The highway safety control method based on the control unit according to claim 4 is characterized in that: The extracting features of the x sample driving condition data includes: Decomposing each of the sample driving road condition data to generate K sample driving road condition segments for each of the sample driving road condition data; K is a positive integer; Performing a fusion calculation on each of the sample driving road condition segments and the set conversion vector to generate a segment vector of each of the sample driving road condition segments; Based on each of the segment vectors, a corresponding driving road condition feature vector is generated.

6. The highway safety control method based on the control unit according to claim 3 is characterized in that: The step of encoding and representing the Y risk factor knowledge blocks by using the risk factor knowledge representation unit to generate target risk factor knowledge vectors of the Y risk factor knowledge blocks includes: Using natural language processing algorithms to perform semantic analysis on each risk factor knowledge block to extract corresponding key semantic information, including risk type, risk level, risk occurrence conditions, risk impact scope, and historical risk event records; Classifying the extracted key semantic information, and assigning corresponding numerical codes to the key semantic information according to the classification results and preset coding rules, and generating a dictionary containing all the key semantic information and corresponding numerical codes; For each type of key semantic information, construct a corresponding feature mapping rule, wherein the feature mapping rule defines a rule for converting the key semantic information into a numerical feature; According to the constructed feature mapping rules, the eigenvalues ​​corresponding to each key semantic information are calculated, and the eigenvalues ​​are normalized to generate a risk factor feature vector matrix containing all eigenvalues ​​and corresponding normalized results; Using a covariance analysis algorithm or an association rule mining algorithm to identify the association relationship between different features in the risk factor feature vector matrix, and based on the identified association relationship, constructing an associated risk factor feature matrix, wherein the associated features in the associated risk factor feature matrix are used to reflect the interaction and influence between different risk factors; Using a recursive feature elimination algorithm or expert knowledge to evaluate the importance of each associated feature in the associated risk factor feature matrix, and assigning a corresponding feature weight to each associated feature according to the evaluation result, wherein the size of the feature weight reflects the contribution of the associated feature in risk prediction; After weighting the associated risk factor features using the feature weights to generate a weighted risk factor feature vector, the weighted risk factor feature vector is fused according to the selected fusion strategy to generate a fused risk factor vector, and a corresponding standardization method is matched according to the distribution pattern of the fused risk factor vector, wherein the standardization method includes normalization, standardization or regularization; The fused risk factor vector is standardized using a selected standardization method to generate a target risk factor knowledge vector of the Y risk factor knowledge blocks.

7. The highway safety control method based on the control unit according to claim 2 is characterized in that: The using of the reference neural network to predict the risk description label according to the target driving road condition feature vector and the target risk factor knowledge vector to generate the risk description label prediction result of each of the sample driving road condition data includes: Connecting the target driving road condition feature vector and the target risk factor knowledge vector to generate a connection feature vector; Performing mutual correlation training based on the connection feature vector to generate mutual correlation training features; For each of the sample traffic condition data, obtain the target inter-correlation training features of the sample traffic condition data for each of the local risk description labels from the inter-correlation training features, and calculate the similarity score between each of the target inter-correlation training features and the set label feature to generate a similarity score result; Performing a standardized conversion on each of the similarity score results to generate a standardized conversion result, and determining the standardized conversion result as the confidence level of the sample driving road condition data matching the corresponding local risk description label; Based on the confidence levels corresponding to the sample driving road condition data, a risk description label prediction result of the sample driving road condition data is generated.

8. The highway safety control method based on the control unit according to claim 7 is characterized in that: The step of performing mutual correlation training based on the connection feature vector to generate mutual correlation training features includes: According to the dimension information of the connection feature vector, the format information of the weight matrix required for the mutual correlation training is determined, wherein if the dimension of the connection feature vector is n-dimensional, a weight matrix with a shape of [n, m] is initialized, wherein m is a preset value related to the dimension of the desired mutual correlation training feature or the network structure, and at the same time, a bias vector matching the weight matrix is ​​initialized, the dimension of the bias vector is m-dimensional, and then the number of rounds of mutual correlation training and the learning rate are set, and the learning rate will be used to adjust the update step size of the weight matrix and the bias vector during the training process; Taking the connection feature vector as the input data structure of the input layer, determining whether it is necessary to construct a hidden layer of the mutual correlation training infrastructure according to the format information of the weight matrix, and if it is necessary to construct a hidden layer, determining the number of neurons in the hidden layer, wherein the number of neurons is determined according to the dimension of the connection feature vector and the complexity of the desired output feature, and defining the activation function of the hidden layer of the mutual correlation training infrastructure and constructing an output layer, wherein the number of neurons in the output layer matches the dimension of the final desired mutual correlation training feature; The connection feature vector is input into the input layer of the constructed mutual correlation training infrastructure to directly pass the connection feature vector to the next layer through the input layer, wherein, if there is a hidden layer, for each neuron in the hidden layer, the input connection feature vector is weighted summed with the corresponding weight in the weight matrix and the corresponding bias value in the bias vector is added to generate an intermediate result, and the intermediate result is nonlinearly transformed through an activation function to obtain the output of the hidden layer, and the output of the hidden layer is used as the input of the next layer to repeat the above weighted summation, bias value addition and activation function operations until it reaches the output layer, and the weighted summation and bias value addition operations are performed in the output layer to obtain a temporary output result, and the temporary output result is a preliminary result of the mutual correlation training calculated based on the current weight matrix and bias vector; Input the temporary output result into a mean square error loss function, calculate the average of the squares of the difference between each element of the temporary output result and a preset target cross-correlation training feature, and output the corresponding loss function value; According to the calculated loss function value, starting from the output layer, reversely calculate the contribution of each weight and bias to the loss function value, wherein, for the output layer, use the derivation rule to calculate the derivative of the loss function value with respect to the weight and bias of the output layer, the derivative reflects the influence of the change of the weight and bias on the loss function value, and then multiply the derivative by a preset learning rate to obtain the updated amount of the output layer weight and bias, and then, for the hidden layer, multiply the error information transmitted by the output layer by the derivative of the activation function of the hidden layer to obtain the updated amount of the hidden layer weight and bias, thereby obtaining the updated amount of the weight and bias of each layer; Using the calculated weight and bias update amounts, the current weight matrix and bias vector are updated, wherein for the weight matrix, the corresponding weight update amount is subtracted from each element; for the bias vector, the corresponding bias update amount is subtracted from each element, thereby obtaining the updated weight matrix and bias vector; Check whether the number of completed training rounds has reached the preset number of training rounds. If it has reached the preset number of training rounds, it means that the mutual correlation training is completed; otherwise, it means that the training is not completed and the next round of training needs to be continued; After the mutual correlation training is completed, the connection feature vector is input again into the trained mutual correlation training basic structure to obtain the mutual correlation training feature, which reflects the mutual correlation relationship between the internal elements of the connection feature vector.

9. The highway safety control method based on the control unit according to claim 3 is characterized in that: The reference neural network also includes a feature fusion unit; The step of obtaining a risk description label prediction network from the reference neural network includes: The current target risk factor knowledge vector is associated with the current driving road condition knowledge representation unit and the current feature fusion unit to generate a risk description label prediction network.

10. A highway safety control system based on a control unit, characterized in that: The highway safety control system based on the control unit includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the highway safety control method based on the control unit as described in any one of claims 1 to 9 above.

Images