Traffic accident early warning method and system based on neural network algorithm
Through the diffusion process and reverse process extraction and combining environmental impact, a neural network risk assessment model is constructed, which solves the problem of insufficient generalization ability of traffic accident warning models in the existing technology, and achieves more accurate risk assessment and timely warning. Improve traffic safety and management efficiency.
Patent Information
- Application Number
- CN202411226526.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-03
- Publication Date
- 2024-12-03
AI Technical Summary
The existing neural network-based traffic accident risk prediction method does not extract the potential characteristics of the original data when preparing the training data, resulting in insufficient generalization capabilities of the model and failing to effectively solve the potential risk identification problem in traffic accident warning.
The potential feature representations in traffic monitoring data are extracted through the diffusion process and reverse process, and revised in combination with environmental impact, a risk assessment model based on neural network is constructed, and the modified feature representation is used for training and optimization, and a traffic accident risk assessment is generated in real time and set the warning threshold to generate an early warning signal.
It improves the prediction accuracy of the model, provides a more reliable traffic accident risk assessment, can quickly respond to current traffic conditions, identify potential accident risks in advance, reduce the frequency and severity of accidents, and improve traffic safety and management efficiency.
Smart Images

Figure CN119068677A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of traffic accident prediction, and in particular to a traffic accident early warning method and system based on a neural network algorithm. Background Art
[0002] Traffic accidents are unexpected events on the road that usually result in property damage, injury or death. These accidents can be caused by a variety of factors, including driver error, vehicle failure, road conditions, weather conditions or illegal behavior. The main purpose of traffic accident warning is to reduce the frequency and severity of traffic accidents by identifying and predicting potential dangerous situations in advance.
[0003] Neural networks are a type of mathematical model inspired by the structure of the human brain and are good at processing complex nonlinear relationships and large-scale data sets. Using neural network models to analyze and predict traffic data can identify potential traffic accident risks in advance and issue early warning signals, thereby improving traffic safety and reducing economic losses.
[0004] For example, Chinese patent CN113222218B discloses a traffic accident risk prediction method based on a convolutional long short-term memory neural network, which models traffic accidents from two aspects: continuity and periodicity. The neural network model is used to extract the temporal and spatial features of traffic accident data at the same time, and further integrates external environmental features to improve the accuracy of the prediction results. However, this method still has the following shortcomings: when preparing the training data for the neural network model, the potential features of the original data are not extracted, so the neural network model faces the problem of insufficient generalization ability when predicting the risk of traffic accidents.
[0005] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention
[0006] In view of the problems in the related art, the present invention proposes a traffic accident warning method and system based on a neural network algorithm to overcome the above-mentioned technical problems existing in the existing related art.
[0007] To this end, the specific technical solution adopted by the present invention is as follows:
[0008] According to one aspect of the present invention, a traffic accident early warning method based on a neural network algorithm is provided, and the traffic accident early warning method based on a neural network algorithm comprises:
[0009] S1. Acquire and preprocess traffic monitoring data including vehicle behavior data and traffic flow data, and configure corresponding traffic accident labels.
[0010] S2. Use the diffusion process and the reverse process mechanism to extract the potential feature representation in the traffic monitoring data, and modify the potential feature representation in combination with environmental influences.
[0011] S3. Build a risk assessment model based on neural network, and use the modified latent feature representation and corresponding traffic accident labels to train and optimize the risk assessment model.
[0012] S4. Acquire real-time traffic monitoring data and generate a revised real-time potential feature representation, input the revised real-time potential feature representation into the trained risk assessment model to obtain a traffic accident risk assessment value.
[0013] S5. Compare the traffic accident risk assessment value with the warning threshold, and when the risk assessment value exceeds the preset threshold, generate a traffic accident warning signal.
[0014] Furthermore, traffic monitoring data including vehicle behavior data and traffic flow data are obtained and preprocessed, and corresponding traffic accident labels are configured, including:
[0015] S11, obtaining vehicle behavior data and traffic flow data to form traffic monitoring data, and performing data cleaning and normalization processing on the vehicle behavior data and traffic flow data;
[0016] S12. Associating traffic monitoring data and traffic accident labels according to timestamps and locations.
[0017] Furthermore, the potential feature representation in the traffic monitoring data is extracted by using the diffusion process and the reverse process mechanism, and the potential feature representation is modified in combination with the environmental impact, including:
[0018] S21, select a diffusion model, and set diffusion steps and noise parameters;
[0019] S22, gradually adding noise to the traffic monitoring data and recording the feature representation of each step;
[0020] S23, using the opposite method of the diffusion process to gradually remove the noise in the traffic monitoring data, and in each step of denoising, modifying the feature representation according to the environmental impact coefficient;
[0021] S24. Obtain the revised latent feature representation from the last step of the reverse process.
[0022] Furthermore, at each step of denoising, the feature representation is modified according to the environmental impact coefficient, including:
[0023] S231. Analyze the impact of different environmental factors on vehicle behavior and traffic flow based on traffic monitoring data and environmental data to obtain candidate environmental factors;
[0024] S232. Establishing a regression model between the candidate environmental factors and the vehicle behavior and traffic flow indicators through a regression analysis algorithm, and calculating the environmental impact coefficient of each candidate environmental factor;
[0025] S233, calculating a correction value according to the environmental impact coefficient and the current environmental conditions;
[0026] S234. Apply the correction value to the current feature representation to obtain the current corrected feature representation.
[0027] Furthermore, based on traffic monitoring data and environmental data, the impact of different environmental factors on vehicle behavior and traffic flow is analyzed, and the candidate environmental factors include:
[0028] S2311. Use statistical charts to obtain the relationship between environmental factors and vehicle behavior and traffic flow;
[0029] S2312. Calculate the Pearson correlation coefficient between environmental factors and vehicle behavior and traffic flow;
[0030] S2313. Determine the degree of influence of different environmental factors on vehicle behavior and traffic flow based on the absolute value of the Pearson correlation coefficient, and screen out candidate environmental factors.
[0031] Furthermore, through the regression analysis algorithm, a regression model between environmental factors and vehicle behavior and traffic flow indicators is established, and the environmental impact coefficient of each candidate environmental factor is calculated, including:
[0032] S2321. Construct a regression model by taking candidate environmental factors as independent variables and vehicle behavior or traffic flow as dependent variables;
[0033] S2322. Fit the regression model to obtain the regression coefficient of each candidate environmental factor, and use the regression coefficient as the environmental impact coefficient.
[0034] Furthermore, a risk assessment model based on a neural network is constructed, and the risk assessment model is trained and optimized using the modified latent feature representation and the corresponding traffic accident labels, including:
[0035] S31, dividing the modified latent feature representation and traffic accident labels into a training set, a validation set and a test set;
[0036] S32. Construct a risk assessment model based on a neural network, and compile the model and set hyperparameters for the risk assessment model;
[0037] S33, using the training set to train the risk assessment model, and using the validation set to validate the risk assessment model during the training process;
[0038] S34. Use the test set to evaluate the performance of the neural network model and adjust the architecture and hyperparameters of the risk assessment model based on the evaluation results.
[0039] Furthermore, real-time traffic monitoring data is obtained and a modified real-time potential feature representation is generated. The modified real-time potential feature representation is input into the trained risk assessment model to obtain a traffic accident risk assessment value including:
[0040] S41, obtaining real-time traffic monitoring data, and generating a modified real-time potential feature representation through a diffusion process and a reverse process;
[0041] S42: input the corrected real-time potential feature representation into the trained risk assessment model, and obtain the traffic accident risk assessment value through prediction by the risk assessment model.
[0042] Further, comparing the traffic accident risk assessment value with the warning threshold, when the risk assessment value exceeds the preset threshold, generating a traffic accident warning signal includes:
[0043] S51, setting a warning threshold, and comparing the real-time traffic accident risk assessment value with the warning threshold;
[0044] S52: If the real-time traffic accident risk assessment value exceeds a preset threshold, there is a potential traffic accident risk, and a traffic accident warning signal is generated.
[0045] According to another aspect of the present invention, a traffic accident warning system based on a neural network algorithm is also provided. The traffic accident warning system based on a neural network algorithm includes a preprocessing module, a potential feature acquisition module, a model training module, a risk assessment value prediction module and a warning judgment module.
[0046] The preprocessing module is used to acquire and preprocess traffic monitoring data including vehicle behavior data and traffic flow data, and configure corresponding traffic accident labels. The potential feature acquisition module is used to extract the potential feature representation in the traffic monitoring data using the mechanism of diffusion process and reverse process, and to modify the potential feature representation in combination with environmental impact. The model training module is used to build a risk assessment model based on neural network, and use the modified potential feature representation and corresponding traffic accident labels to train and optimize the risk assessment model. The risk assessment value prediction module is used to acquire real-time traffic monitoring data and generate a modified real-time potential feature representation, and input the modified real-time potential feature representation into the trained risk assessment model to obtain the traffic accident risk assessment value. The warning judgment module is used to compare the traffic accident risk assessment value with the warning threshold. When the risk assessment value exceeds the preset threshold, a traffic accident warning signal is generated.
[0047] The beneficial effects of the present invention are:
[0048] (1) The present invention extracts potential feature representations from traffic monitoring data through the mechanism of diffusion process and reverse process, and modifies the potential features in combination with environmental influences, effectively capturing the deep and essential characteristics of the data, which can improve the prediction accuracy of the model and avoid errors caused by data noise and redundant information, thereby providing a more reliable traffic accident risk assessment.
[0049] (2) The present invention obtains real-time traffic monitoring data and generates a modified real-time potential feature representation, which is input into a trained risk assessment model for prediction. Through this real-time data processing and model prediction, the system can quickly respond to current traffic conditions and provide timely warning signals.
[0050] (3) The present invention sets a warning threshold, compares the traffic accident risk assessment value with the warning threshold in real time, and generates a traffic accident warning signal when the assessment value exceeds the threshold. The warning mechanism can identify and prompt potential traffic accident risks in advance, prompting relevant personnel or systems to take preventive measures, thereby effectively reducing the frequency and severity of traffic accidents, reducing casualties and property losses, and improving road traffic safety and management efficiency. BRIEF DESCRIPTION OF THE DRAWINGS
[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0052] Figure 1 is a flow chart of a traffic accident early warning method based on a neural network algorithm according to an embodiment of the present invention;
[0053] Figure 2 It is a principle block diagram of a traffic accident warning system based on a neural network algorithm according to an embodiment of the present invention.
[0054] In the figure:
[0055] 1. Preprocessing module; 2. Potential feature acquisition module; 3. Model training module; 4. Risk assessment value prediction module; 5. Early warning judgment module. DETAILED DESCRIPTION
[0056] To further illustrate each embodiment, the present invention provides drawings, which are part of the disclosure of the present invention and are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, ordinary technicians in the field should be able to understand other possible implementations and advantages of the present invention. The components in the figures are not drawn to scale, and similar component symbols are generally used to represent similar components.
[0057] According to an embodiment of the present invention, a traffic accident warning method and system based on a neural network algorithm are provided.
[0058] The present invention is further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to one embodiment of the present invention, a traffic accident early warning method based on a neural network algorithm is provided, and the traffic accident early warning method based on a neural network algorithm includes:
[0059] S1. Acquire and preprocess traffic monitoring data including vehicle behavior data and traffic flow data, and configure corresponding traffic accident labels.
[0060] In one embodiment, acquiring and preprocessing traffic monitoring data including vehicle behavior data and traffic flow data, and configuring corresponding traffic accident tags include:
[0061] S11. Acquire vehicle behavior data and traffic flow data to form traffic monitoring data, and perform data cleaning and normalization processing on the vehicle behavior data and traffic flow data.
[0062] S12. Associating traffic monitoring data and traffic accident labels according to timestamp and location. Timestamp refers to the specific time point of data recording, which can align and associate data from different sources in chronological order. Location refers to the geographical location or specific road segment of data recording, which can align and associate data related to the same geographical location.
[0063] S2. Use the diffusion process and the reverse process mechanism to extract the potential feature representation in the traffic monitoring data, and modify the potential feature representation in combination with environmental influences.
[0064] In one embodiment, the potential feature representation in the traffic monitoring data is extracted by using the diffusion process and the reverse process mechanism, and the potential feature representation is modified in combination with the environmental impact, including:
[0065] S21, selecting a diffusion model, such as a Gaussian diffusion model, and setting diffusion steps and noise parameters;
[0066] Setting the diffusion steps includes setting the number of steps: determining the number of steps in the diffusion process, such as 100 steps. Setting the step size: the proportion of noise added in each step, such as 0.01.
[0067] The noise parameters include the noise mean: generally set to 0. Noise variance: set according to the data characteristics, such as 0.1.
[0068] S22. Gradually add noise to the traffic monitoring data, that is, simulate the transformation of real data into potential features, and record the feature representation at each step.
[0069] S23. Use the opposite method of the diffusion process to gradually remove the noise in the traffic monitoring data, restore the original distribution of the data, and modify the feature representation according to the environmental impact coefficient at each step of denoising.
[0070] S24. Obtain the revised latent feature representation from the last step of the reverse process.
[0071] In one embodiment, during each step of denoising, the feature representation is modified according to the environmental impact coefficient, including:
[0072] S231. Analyze the impact of different environmental factors on vehicle behavior and traffic flow based on traffic monitoring data and environmental data to obtain candidate environmental factors;
[0073] S232. Establishing a regression model between the candidate environmental factors and the vehicle behavior and traffic flow indicators through a regression analysis algorithm, and calculating the environmental impact coefficient of each candidate environmental factor;
[0074] S233, calculating a correction value according to the environmental impact coefficient and the current environmental conditions;
[0075] S234. Apply the correction value to the current feature representation to obtain the current corrected feature representation.
[0076] In one embodiment, based on traffic monitoring data and environmental data, the influence of different environmental factors on vehicle behavior and traffic flow is analyzed, and candidate environmental factors are obtained, including:
[0077] S2311. Use statistical charts, such as scatter plots, box plots, etc., to obtain the relationship between environmental factors and vehicle behavior and traffic flow;
[0078] S2312. Calculate the Pearson correlation coefficient between environmental factors and vehicle behavior and traffic flow;
[0079] The value range of the Pearson correlation coefficient is between -1 and 1. Values close to 1 or -1 indicate strong correlation, and 0 indicates no correlation. The strength of the correlation is determined by the absolute value of the correlation coefficient, for example, 0.1 to 0.3 is a weak correlation, 0.3 to 0.5 is a moderate correlation, and above 0.5 is a strong correlation.
[0080] S2313. Determine the degree of influence of different environmental factors on vehicle behavior and traffic flow based on the absolute value of the Pearson correlation coefficient, and screen out highly correlated candidate environmental factors.
[0081] In one embodiment, a regression model between environmental factors and vehicle behavior and traffic flow indicators is established through a regression analysis algorithm, and the environmental impact coefficient of each candidate environmental factor is calculated to include:
[0082] S2321. Construct a regression model by taking candidate environmental factors as independent variables and vehicle behavior or traffic flow as dependent variables.
[0083] S2322. Fit the regression model to obtain the regression coefficient of each candidate environmental factor, and use the regression coefficient as the environmental impact coefficient, that is, use regression analysis to determine the specific impact of each candidate environmental factor on traffic flow and vehicle behavior. Specifically, it involves estimating the regression coefficient, which quantifies the expected change in the dependent variable when the independent variable changes by one unit.
[0084] For example, the temperature influence coefficient is -0.5 (for every 1 degree increase in temperature, the vehicle speed decreases by 0.5 km / h).
[0085] Humidity influence coefficient: -0.1 (for every 1% increase in humidity, the vehicle speed decreases by 0.1km / h).
[0086] Precipitation impact coefficient: -2.0 (for every 1mm increase in precipitation, the vehicle speed decreases by 2.0km / h).
[0087] S3. Build a risk assessment model based on neural network, and use the modified latent feature representation and corresponding traffic accident labels to train and optimize the risk assessment model.
[0088] In one embodiment, constructing a risk assessment model based on a neural network and using the modified latent feature representation and the corresponding traffic accident labels to train and optimize the risk assessment model includes:
[0089] S31. Divide the corrected latent feature representation and traffic accident labels into a training set, a validation set, and a test set.
[0090] S32. Construct a risk assessment model based on a neural network. When selecting a neural network, include a multi-layer perceptron, a convolutional neural network, or a recurrent neural network. Determine the number of layers of the neural network architecture, the number of neurons in each layer, the activation function, etc., and compile the model and set hyperparameters for the risk assessment model.
[0091] Model compilation:
[0092] Select a loss function, such as binary cross entropy. Select an optimizer, such as Adam. Set an evaluation metric, such as accuracy.
[0093] Hyperparameter settings:
[0094] Set hyperparameters such as learning rate, batch size, and training rounds.
[0095] S33. Use the training set to train the risk assessment model, and use the validation set to validate the risk assessment model during the training process to prevent overfitting.
[0096] S34. Use the test set to evaluate the performance of the neural network model, calculate indicators such as accuracy, precision, recall, F1 score, and adjust the architecture and hyperparameters of the risk assessment model based on the evaluation results.
[0097] S4. Acquire real-time traffic monitoring data and generate a revised real-time potential feature representation. Input the revised real-time potential feature representation into the trained risk assessment model to obtain a traffic accident risk assessment value, that is, the traffic accident risk probability at each moment.
[0098] In one embodiment, real-time traffic monitoring data is obtained, and a modified real-time potential feature representation is generated. The modified real-time potential feature representation is input into a trained risk assessment model to obtain a traffic accident risk assessment value including:
[0099] S41. Obtain real-time traffic monitoring data, and generate a modified real-time potential feature representation through a diffusion process and a reverse process. S42. Input the modified real-time potential feature representation into a trained risk assessment model, and obtain a traffic accident risk assessment value through the risk assessment model prediction. In addition, interpret the obtained risk assessment value. Display the results in a monitoring interface or report for relevant personnel to understand.
[0100] S5. Compare the traffic accident risk assessment value with the warning threshold, and when the risk assessment value exceeds the preset threshold, generate a traffic accident warning signal.
[0101] In one embodiment, comparing the traffic accident risk assessment value with the warning threshold, and when the risk assessment value exceeds the preset threshold, generating a traffic accident warning signal includes:
[0102] S51. According to historical data and actual needs, a warning threshold is set (the threshold may be a fixed value, for example, it may also be dynamically adjusted according to different time periods or road sections), and the real-time traffic accident risk assessment value is compared with the warning threshold.
[0103] S52: If the real-time traffic accident risk assessment value exceeds a preset threshold, there is a potential traffic accident risk, and a traffic accident warning signal is generated.
[0104] In addition, multiple warning thresholds can be set, that is, according to the comparison results, the risk assessment value is classified into: low risk, medium risk, high risk, etc. According to low risk, medium risk, high risk, etc., corresponding traffic accident warning signals are generated.
[0105] like Figure 2 As shown, according to another embodiment of the present invention, a traffic accident warning system based on a neural network algorithm is also provided. The traffic accident warning system based on a neural network algorithm includes a preprocessing module 1, a potential feature acquisition module 2, a model training module 3, a risk assessment value prediction module 4 and a warning judgment module 5.
[0106] Preprocessing module 1 is used to acquire and preprocess traffic monitoring data including vehicle behavior data and traffic flow data, and configure corresponding traffic accident labels. Potential feature acquisition module 2 is used to extract potential feature representations in traffic monitoring data using the mechanism of diffusion process and reverse process, and to modify the potential feature representations in combination with environmental impact. Model training module 3 is used to construct a risk assessment model based on neural network, and use the modified potential feature representations and corresponding traffic accident labels to train and optimize the risk assessment model. Risk assessment value prediction module 4 is used to acquire real-time traffic monitoring data and generate a modified real-time potential feature representation, and input the modified real-time potential feature representation into the trained risk assessment model to obtain a traffic accident risk assessment value. Warning judgment module 5 is used to compare the traffic accident risk assessment value with the warning threshold. When the risk assessment value exceeds the preset threshold, a traffic accident warning signal is generated.
[0107] In summary, the present invention extracts potential feature representations in traffic monitoring data through the mechanism of diffusion process and reverse process, and corrects the potential features in combination with environmental influences, effectively captures the deep and essential features in the data, can improve the prediction accuracy of the model, avoid errors caused by data noise and redundant information, and thus provide a more reliable traffic accident risk assessment. The present invention obtains real-time traffic monitoring data and generates a corrected real-time potential feature representation, which is input into a trained risk assessment model for prediction. Through this real-time data processing and model prediction, the system can quickly respond to current traffic conditions and provide timely warning signals. The present invention sets a warning threshold, compares the traffic accident risk assessment value with the warning threshold in real time, and generates a traffic accident warning signal when the assessment value exceeds the threshold. The warning mechanism can identify and prompt potential traffic accident risks in advance, prompting relevant personnel or systems to take preventive measures, thereby effectively reducing the frequency and severity of traffic accidents, reducing casualties and property losses, and improving road traffic safety and management efficiency.
[0108] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims
1. A traffic accident early warning method based on a neural network algorithm, characterized in that: The traffic accident early warning method based on the neural network algorithm includes: S1. Acquire and preprocess traffic monitoring data including vehicle behavior data and traffic flow data, and configure corresponding traffic accident labels; S2, using the diffusion process and the reverse process mechanism to extract the potential feature representation in the traffic monitoring data, and modify the potential feature representation in combination with environmental influences; S3. Construct a risk assessment model based on a neural network, and use the modified latent feature representation and corresponding traffic accident labels to train and optimize the risk assessment model; S4, obtaining real-time traffic monitoring data, and generating a corrected real-time potential feature representation, and inputting the corrected real-time potential feature representation into the trained risk assessment model to obtain a traffic accident risk assessment value; S5. Compare the traffic accident risk assessment value with the warning threshold, and when the risk assessment value exceeds the preset threshold, generate a traffic accident warning signal.
2. A traffic accident early warning method based on a neural network algorithm according to claim 1, characterized in that: The acquisition and preprocessing of traffic monitoring data including vehicle behavior data and traffic flow data, and configuration of corresponding traffic accident labels include: S11, obtaining vehicle behavior data and traffic flow data to form traffic monitoring data, and performing data cleaning and normalization processing on the vehicle behavior data and traffic flow data; S12. Associating traffic monitoring data and traffic accident labels according to timestamps and locations.
3. The traffic accident early warning method based on a neural network algorithm according to claim 1, characterized in that: The method of extracting potential feature representations from traffic monitoring data by using the diffusion process and the reverse process mechanism, and correcting the potential feature representations in combination with environmental influences includes: S21, select a diffusion model, and set diffusion steps and noise parameters; S22, gradually adding noise to the traffic monitoring data and recording the feature representation of each step; S23, using the opposite method of the diffusion process to gradually remove the noise in the traffic monitoring data, and in each step of denoising, modifying the feature representation according to the environmental impact coefficient; S24. Obtain the revised latent feature representation from the last step of the reverse process.
4. A traffic accident early warning method based on a neural network algorithm according to claim 3, characterized in that: At each step of denoising, the feature representation is modified according to the environmental impact coefficient, including: S231. Analyze the impact of different environmental factors on vehicle behavior and traffic flow based on traffic monitoring data and environmental data to obtain candidate environmental factors; S232. Establishing a regression model between the candidate environmental factors and the vehicle behavior and traffic flow indicators through a regression analysis algorithm, and calculating the environmental impact coefficient of each candidate environmental factor; S233, calculating a correction value according to the environmental impact coefficient and the current environmental conditions; S234. Apply the correction value to the current feature representation to obtain the current corrected feature representation.
5. A traffic accident early warning method based on a neural network algorithm according to claim 4, characterized in that: Based on the traffic monitoring data and environmental data, the influence of different environmental factors on vehicle behavior and traffic flow is analyzed, and the candidate environmental factors include: S2311. Use statistical charts to obtain the relationship between environmental factors and vehicle behavior and traffic flow; S2312. Calculate the Pearson correlation coefficient between environmental factors and vehicle behavior and traffic flow; S2313. Determine the degree of influence of different environmental factors on vehicle behavior and traffic flow based on the absolute value of the Pearson correlation coefficient, and screen out candidate environmental factors.
6. A traffic accident early warning method based on a neural network algorithm according to claim 5, characterized in that: The regression analysis algorithm is used to establish a regression model between environmental factors and vehicle behavior and traffic flow indicators, and the environmental impact coefficient of each candidate environmental factor is calculated, including: S2321. Construct a regression model by taking candidate environmental factors as independent variables and vehicle behavior or traffic flow as dependent variables; S2322. Fit the regression model to obtain the regression coefficient of each candidate environmental factor, and use the regression coefficient as the environmental impact coefficient.
7. The traffic accident early warning method based on a neural network algorithm according to claim 1, characterized in that: The construction of a risk assessment model based on a neural network and the use of the modified latent feature representation and the corresponding traffic accident labels to train and optimize the risk assessment model include: S31, dividing the modified latent feature representation and traffic accident labels into a training set, a validation set and a test set; S32. Construct a risk assessment model based on a neural network, and compile the model and set hyperparameters for the risk assessment model; S33, using the training set to train the risk assessment model, and using the validation set to validate the risk assessment model during the training process; S34. Use the test set to evaluate the performance of the neural network model and adjust the architecture and hyperparameters of the risk assessment model based on the evaluation results.
8. The traffic accident early warning method based on a neural network algorithm according to claim 7 is characterized in that: The real-time traffic monitoring data is obtained, and a modified real-time potential feature representation is generated, and the modified real-time potential feature representation is input into a trained risk assessment model to obtain a traffic accident risk assessment value, including: S41, obtaining real-time traffic monitoring data, and generating a modified real-time potential feature representation through a diffusion process and a reverse process; S42: input the corrected real-time potential feature representation into the trained risk assessment model, and obtain the traffic accident risk assessment value through prediction by the risk assessment model.
9. A traffic accident early warning method based on a neural network algorithm according to claim 8, characterized in that: The comparing the traffic accident risk assessment value with the warning threshold, and generating a traffic accident warning signal when the risk assessment value exceeds the preset threshold, comprises: S51, setting a warning threshold, and comparing the real-time traffic accident risk assessment value with the warning threshold; S52: If the real-time traffic accident risk assessment value exceeds a preset threshold, there is a potential traffic accident risk, and a traffic accident warning signal is generated.
10. A traffic accident warning system based on a neural network algorithm, used to implement the traffic accident warning method based on a neural network algorithm according to any one of claims 1 to 9, characterized in that: The traffic accident warning system based on the neural network algorithm includes a preprocessing module, a potential feature acquisition module, a model training module, a risk assessment value prediction module and a warning judgment module; The preprocessing module is used to acquire and preprocess traffic monitoring data including vehicle behavior data and traffic flow data, and configure corresponding traffic accident labels; The potential feature acquisition module is used to extract the potential feature representation in the traffic monitoring data by using the mechanism of diffusion process and reverse process, and to modify the potential feature representation in combination with environmental influence; The model training module is used to construct a risk assessment model based on a neural network, and use the modified potential feature representation and corresponding traffic accident labels to train and optimize the risk assessment model; The risk assessment value prediction module is used to obtain real-time traffic monitoring data and generate a modified real-time potential feature representation, and input the modified real-time potential feature representation into the trained risk assessment model to obtain a traffic accident risk assessment value; The warning judgment module is used to compare the traffic accident risk assessment value with the warning threshold, and generate a traffic accident warning signal when the risk assessment value exceeds the preset threshold.
Citation Information
Cited By
Intelligent traffic accident prediction and emergency response system and method based on vehicle-mounted data
CN119721395A
Accurate traffic accident early warning system based on multi-sensor fusion
CN121811659A
