Disclosure of Invention
Aiming at the problem of lack of geological disaster early warning in the driving process in the prior art, the invention provides a vehicle-mounted geological disaster early warning system and method based on environment awareness.
In order to achieve the above object, the present invention is realized by the following technical scheme:
a vehicle-mounted geological disaster early warning system based on environmental awareness, the system comprising:
The information acquisition module is used for acquiring geological disaster information, vehicle information and external environment information in real time, wherein the geological disaster information comprises seismic waveform data and topographic image data, the vehicle information comprises a navigation route, a real-time position of the vehicle, a real-time speed of the vehicle, historical driving track data of the vehicle and driver preference data, and the external environment information comprises historical traffic data, real-time traffic information and weather data;
The judgment analysis module comprises a driving path prediction unit, a geological disaster prediction unit and a risk assessment unit;
The travel path prediction unit is used for predicting a vehicle travel path based on vehicle information and external environment information;
The geological disaster prediction unit is used for inputting geological disaster information into a pre-trained geological disaster early warning model and predicting probability values of occurrence of geological disasters of each monitoring point or region;
The geological disaster early warning model adopts a multi-task learning architecture, can learn a plurality of targets of earthquake intensity prediction, landslide risk assessment, debris flow early warning and collapse early warning at the same time, and promotes knowledge migration by sharing a bottom layer characteristic layer;
The risk assessment unit is used for assessing the risk level of the influence of the geological disaster on the vehicle according to the probability value of the occurrence of the geological disaster of each monitoring point or region;
and the early warning execution module is used for matching the safety suggestion according to the risk level and carrying out early warning prompt.
As a preferable mode of the present invention, the method of predicting a vehicle travel path based on vehicle information and external environment information includes:
data cleaning, denoising and feature extraction are carried out on the vehicle information and the external environment information, and a structured data set for training is generated;
training the historical driving track data and the real-time traffic information of the vehicle through a recurrent neural network or a long-short-term memory network, constructing a time sequence prediction model, and predicting the driving path of the vehicle based on the time sequence data;
data training is carried out locally by utilizing the federal learning technology, and model parameters are shared, but original data are not shared directly;
Optimizing the prediction result by adopting a deep Q learning algorithm, and simulating feedback mechanisms of different path selections;
based on the real-time traffic data and the weather data, the prediction result is dynamically adjusted and updated in real time by using a self-attention mechanism and timely fed back to a driver so as to adapt to the traffic condition which changes in real time.
As a preferable scheme of the invention, the construction method of the geological disaster early warning model comprises the following steps:
carrying out data preprocessing, namely carrying out normalization processing on multi-source geological disaster information, and constructing an adjacent matrix according to the geographical position relation to represent the connection strength between monitoring points;
Feature extraction, namely extracting features from seismic waveform data by using 1D CNN, and extracting topographic information from topographic image data by using 2D CNN;
Input feature map by DCT transformationDividing the channel dimension into a plurality of feature vectors, and applying DCT (discrete cosine transform) of different frequency components to each feature vector, wherein the formula is as follows:
;
In the formula,Is the firstFrequency representation of the individual feature vectors after DCT processing; The index of the feature vector is represented,,Representing the number of feature vectors; Representing a characteristic diagramMiddle (f)A feature vector; the index of the frequency is represented and,,Length for DCT transform; indexing elements in the feature vector; Representing a characteristic diagramMiddle (f)The first feature vectorAn element; is the basis function of the DCT;
FIG. neural network block operations Using ChebConv and GCSConv layers, spatial and temporal relationships in the encoded graph structure are conveyed according to the adjacency matrix, the FIG. neural network block operations are represented as:
;
In the formula,Is the firstFeature vectors of the individual nodes; the number of nodes; is an adjacency matrix; representing the input feature vectorAdjacency matrixIs output by a graph neural network block;
Self-supervised contrast learning, namely, by constructing positive samples and negative samples and defining contrast loss functions, the model can learn the internal relation between data without depending on a large amount of marked data, and the contrast loss functions are expressed as follows:
;
In the formula,Is a contrast loss; is a positive sample pairAndA similarity function between the two; as a set of negative examples of the way,Is a negative sample setAny one of the elements; Is a positive sampleAnd negative sampleA similarity function between the two; Is a temperature parameter for adjusting the output of the similarity function;
the state updating of the convolution long expression memory model is realized by integrating convolution operation into a long expression memory architecture and is used for capturing a multi-scale space-time structure, and the state updating equation is as follows:
;
;
In the formula,AndRespectively representing a fast hidden state and a slow hidden state; Is a time stepIs a measurement of the observed value of (2); AndIs a gain factor for adjusting respectivelyAndRate of change of (2)AndRespectively areAndIs used to update the function of the (c),AndRespectively update functionsAndFor adjusting the behavior of the function; Representing the Hadamard product;
Adopting an implicit-explicit time stepping scheme in a convolution long expression memory model, and solving the state updating equation by using an implicit Euler method or a Longer-Kutta method;
and (3) multi-task learning, namely introducing a multi-task learning mechanism, and simultaneously learning a plurality of targets such as earthquake intensity prediction, landslide risk assessment and debris flow early warning, and promoting knowledge migration by sharing a bottom characteristic layer.
As a preferred solution of the present invention, in the multitasking learning mechanism, the probability of converting the output into each class using the softmax function is expressed as:
;
In the formula,Is the firstA score for each category; Is the firstRaw output values of the individual classes (logits, i.e., the unnormalized value that the model outputs at the last layer); Is the predicted total number of categories.
As a preferred solution of the present invention, in the risk assessment unit, weighted average is performed on probability values of occurrence of geological disasters of each monitoring point or region on a vehicle driving path, so as to obtain comprehensive risk probability of the whole vehicle driving path, and the risk is classified into different grades according to the comprehensive risk probability:
when the comprehensive risk probability is 0-0.2, the risk level is low risk;
When the comprehensive risk probability is 0.2-0.6, the risk grade is medium risk;
when the comprehensive risk probability is 0.6-1.0, the risk grade is high risk;
The weight is determined according to the historical disaster frequency of each monitoring point or area, the terrain and the distance from the vehicle driving path.
As a preferable scheme of the invention, the risk assessment unit further comprises a step of collecting a vehicle-mounted camera picture, judging whether trees, billboards, high buildings, high mountains, boulders or other objects with falling risks exist around the vehicle through a target detection algorithm, and if the trees, billboards, high buildings, boulders or other objects with falling risks exist, adding falling risk weights into the comprehensive risk probability to further divide the risk grades.
As a preferred embodiment of the present invention, the early warning execution module includes:
the geological information display unit is used for displaying or voice broadcasting geological disaster information through the central control screen and generating early warning information to remind a user;
And the safety suggestion matching unit is used for matching the safety suggestions according to the current risk level of the vehicle.
An on-vehicle geological disaster early warning method based on environment perception, based on the on-vehicle geological disaster early warning system based on environment perception as described above, the method includes:
Acquiring geological disaster information, vehicle information and external environment information in real time, wherein the geological disaster information comprises seismic waveform data and topographic image data, the vehicle information comprises a navigation route, a real-time position of the vehicle, a real-time vehicle speed, historical driving track data of the vehicle and driver preference data, and the external environment information comprises historical traffic data, real-time traffic information and weather data;
predicting a vehicle travel path based on the vehicle information and the external environment information;
inputting geological disaster information into a pre-trained geological disaster early warning model, and predicting the occurrence probability value of geological disasters of each monitoring point or area;
according to the probability value of occurrence of the geological disaster of each monitoring point or region, evaluating the risk level of the geological disaster affecting the vehicle;
and carrying out early warning prompt according to the risk level matching safety suggestion.
A computer readable storage medium storing a computer program which when executed by a processor implements a vehicle-mounted geological disaster warning method based on environmental awareness as described above.
Compared with the prior art, the method has the advantages that diversified information can be obtained in real time, so that the system can predict the geological disaster risk more accurately and respond in time, the vehicle information and the external environment information are combined to predict the driving path of the vehicle, the system is facilitated to identify potential geological disaster risk areas in advance so as to perform targeted early warning, a multi-task learning architecture geological disaster early warning model is adopted, multiple targets such as earthquake intensity prediction, landslide risk assessment, debris flow early warning and collapse early warning can be learned at the same time, the generalization performance and prediction accuracy of the model are improved, the occurrence probability of geological disasters can be predicted more accurately by the system, the risk level of the influence of the geological disasters on the vehicle is estimated in real time according to the occurrence probability value of the geological disasters, the real-time risk assessment capability enables the system to respond in time before the occurrence of the disasters, safety suggestions are provided for drivers, and correct geological disasters are facilitated to make correct decisions when the drivers face potential geological disasters, so that driving safety is remarkably improved.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. It will be apparent that the described embodiments are some, but not all, embodiments of the invention. All other embodiments, which are obtained by a person skilled in the art based on the described embodiments of the invention, fall within the scope of protection of the invention.
As shown in fig. 1, in an embodiment of the present invention, there is provided a vehicle-mounted geological disaster early warning system based on environmental awareness, including:
(1) Information acquisition module
The information acquisition module is used for acquiring geological disaster information, vehicle information and external environment information in real time, wherein the geological disaster information comprises seismic waveform data, topographic image data and the like, the vehicle information comprises navigation routes, real-time positions of vehicles, real-time speeds of vehicles, historical driving track data of vehicles, driver preference data (driving habits, preferences and the like) and the like, and the external environment information comprises historical traffic data, real-time traffic information (such as traffic jams, road construction, traffic accidents and the like), weather data and the like;
(2) Judgment and analysis module
The system comprises a driving path prediction unit, a geological disaster prediction unit and a risk assessment unit;
1) The travel path prediction unit is configured to predict a travel path of a vehicle based on vehicle information and external environment information, and the method includes:
data cleaning, denoising and feature extraction are carried out on the vehicle information and the external environment information, and a structured data set for training is generated;
training the historical driving track data and the real-time traffic information of the vehicle through a recurrent neural network or a long-short-term memory network, constructing a time sequence prediction model, and predicting the driving path of the vehicle based on the time sequence data;
Data training is carried out locally by utilizing the federal learning technology, model parameters are shared, and original data is not shared directly, so that data privacy protection is enhanced, and generalization capability of a path prediction model is improved through cooperative training of different data sources;
Optimizing the prediction result by adopting a deep Q-network (DQN) algorithm, and enabling the prediction result to be more accurate and adapt to a complex traffic environment by simulating feedback mechanisms of different path selections;
based on the real-time traffic data and the weather data, the prediction result is dynamically adjusted and updated in real time by using a self-attention mechanism and timely fed back to a driver so as to adapt to the traffic condition which changes in real time.
2) The geological disaster prediction unit is used for inputting geological disaster information into a pre-trained geological disaster early warning model and predicting probability values of occurrence of geological disasters of each monitoring point or region;
The geological disaster early warning model adopts a multi-task learning architecture, can learn a plurality of targets such as earthquake intensity prediction, landslide risk assessment, debris flow early warning and collapse early warning and the like at the same time, and promotes knowledge migration by sharing a bottom characteristic layer;
A long expression memory architecture is a recurrent neural network element designed to process data with complex spatio-temporal dependencies, and LEM architecture enhances the capabilities of traditional recurrent elements (such as LSTM or GRU) by introducing multi-scale temporal dynamics, thereby better capturing long-term dependencies and multi-scale patterns in the data. By integrating convolution operations into the LEM architecture, the processing power of the model on spatial information is further enhanced. This enables the model to better capture multi-scale spatial patterns in the input data.
The construction method of the geological disaster early warning model comprises the following steps:
1. Data preprocessing:
Carrying out normalization processing on multi-source geological disaster information (such as seismic waveform data, topographic image data, meteorological data and the like), and constructing an adjacent matrix according to geographic position relations to represent the connection strength between monitoring points;
2. Feature extraction:
Convolution feature extraction, namely extracting features from seismic waveform data by using a 1D CNN (one-dimensional convolution neural network), and extracting topographic information from topographic image data by using a 2D CNN (two-dimensional convolution neural network);
frequency enhanced channel attention mechanism-feature map input by DCT transformThe method comprises the steps of dividing the channel dimension into a plurality of feature vectors, and applying DCT (discrete cosine transformation) of different frequency components to each feature vector to capture more frequency information and avoid the Gibbs phenomenon in the Fourier transform, wherein the formula is as follows:
;
In the formula,Is the firstFrequency representation of the individual feature vectors after DCT processing; The index of the feature vector is represented,,Representing the number of feature vectors; Representing a characteristic diagramMiddle (f)A feature vector; the index of the frequency is represented and,,Length for DCT transform; indexing elements in the feature vector; Representing a characteristic diagramMiddle (f)The first feature vectorAn element; is the basis function of the DCT;
3. graph Neural Network (GNN) block operation:
using ChebConv and GCSConv layers, spatial and temporal information is conveyed according to adjacency matrices, encoding spatio-temporal relationships in the graph structure, the graph neural network block operation is expressed as:
;
In the formula,Is the firstFeature vectors of the individual nodes; the number of nodes; is an adjacency matrix; representing the input feature vectorAdjacency matrixIs output by a graph neural network block;
ChebConv (Chebyshev Convolution ) is a convolution operation used in a Graph Neural Network (GNN) that approximates a graph filter based on chebyshev polynomials that can effectively process local and non-local information in graph structure data to better capture complex relationships between nodes in a graph.
GCSConv (GRAPH SKIP Convolution) can be expressed as "graph-jump convolution," which refers to the ability of the layer to enhance local structure and feature diffusion through trainable jump connections (skip connections). In particular, the GCSConv layer combines spectral domain diffusion (focusing on the global structure of the graph) and spatial domain diffusion (emphasizing the local neighborhood structure of each node) to better capture local and global information in the graph data. In the application of earthquake early warning and other geological disaster prediction, GCSConv layers can help the model to more effectively encode complex space-time relations between monitoring points, and the accuracy and reliability of prediction are improved.
4. Self-supervision contrast learning:
The generalization capability of the model is enhanced by introducing self-supervision contrast learning, and the model can learn the internal relation between data without depending on a large amount of labeling data by constructing positive samples and negative samples and defining a proper contrast loss function, wherein the contrast loss function is expressed as follows:
;
In the formula,For comparison loss, the method is used for measuring the similarity difference between the positive sample pair and the negative sample pair; is a positive sample pairAndSimilarity functions between the two, common similarity functions include cosine similarity (Cosine Similarity) or Dot Product (Dot Product); as a set of negative examples of the way,Is a negative sample setAny one of the elements; Is a positive sampleAnd negative sampleA similarity function between the two; Is a temperature parameter for adjusting the output of the similarity function, typically a small positive number, for controlling the sharpness of the similarity function;
5. State update of convolution long expression memory model:
The method is used for capturing a multi-scale space-time structure by integrating convolution operation into a long expression memory architecture, and a state update equation is as follows:
;
;
In the formula,AndRespectively representing a fast hidden state and a slow hidden state; Is a time stepIs a measurement of the observed value of (2); AndIs a gain factor for adjusting respectivelyAndIs a rate of change of (2); AndRespectively areAndIs used to update the function of the (c),AndRespectively update functionsAndFor adjusting the behavior of the function; Representing the Hadamard product;
in a convolutional long-term expression memory model, an implicit-explicit (IMEX) time-stepping scheme is employed to improve numerical stability, for example, the implicit Euler method or the Dragon-Kutta method can be used to solve the state update equation.
6. Multitasking learning:
a multi-task learning mechanism is introduced, so that the model can learn a plurality of targets such as earthquake intensity prediction, landslide risk assessment, debris flow early warning and the like at the same time, and knowledge migration is promoted by sharing a bottom layer characteristic layer;
7. outputting a predicted probability value:
The final output of the model may be a continuous value (e.g., seismic intensity), a classification label (e.g., risk level), or other form of prediction, for multi-classification tasks, the probability of converting the output to each class directly using the softmax function is expressed as:
;
In the formula,Is the firstA score for each geological disaster category; Is the firstOriginal output values of individual geologic hazard categories (logits, i.e., un-normalized values of the model output at the last layer); is the total number of predicted geological disaster categories.
3) The risk assessment unit is used for assessing the risk level of the influence of the geological disaster on the vehicle according to the probability value of the occurrence of the geological disaster of each monitoring point or region;
In the risk assessment unit, weighted average is carried out on probability values of geological disasters of all monitoring points or areas on a vehicle driving path to obtain comprehensive risk probability of the whole vehicle driving path, and the risk is divided into different grades according to the comprehensive risk probability:
when the comprehensive risk probability is 0-0.2, the risk level is low risk;
When the comprehensive risk probability is 0.2-0.6, the risk grade is medium risk;
when the comprehensive risk probability is 0.6-1.0, the risk grade is high risk;
the weight is determined according to the historical disaster frequency of each monitoring point or area, the terrain and the distance between the monitoring point or area and the vehicle driving path;
The risk assessment unit further comprises a vehicle-mounted camera picture collection unit, a target detection algorithm is used for judging whether trees, billboards, high buildings, high mountains, boulders or other objects with falling risks exist around the vehicle, and if the trees, billboards, high buildings, boulders or other objects with falling risks exist, falling risk weights are added into the comprehensive risk probability to further divide risk grades.
Optionally, the seismic risk assessment criteria may also be set to:
High risk of earthquake with a level greater than 6, a depth of 0-60 km of the source, high building, mountain, boulders, bridges, poles, etc. within 80 km from the center of the earthquake, and nearby (if the earthquake is within 50 km of the ocean floor and the level is above 6, it increases: within 200 meters from the coastline, within 10 meters from the altitude);
The medium risk is that the earthquake is grade 4 to grade 6, the depth of the earthquake focus is 60 to 300 km, the distance from the earthquake focus is 80 to 160 km, street lamps, trees, high walls, fences, billboards and the like are arranged nearby (if the earthquake is in the sea, the earthquake increases by 200 to 1000 meters from the coastline, and the altitude is 10 to 30 meters);
Low risk, earthquake grade less than grade 4, earthquake source depth more than 300 km, distance more than 160 km from earthquake center, etc.
(3) Early warning execution module
The early warning execution module is used for matching the safety suggestion according to the risk level and carrying out early warning prompt, and comprises the following steps:
the geological information display unit is used for displaying or voice broadcasting geological disaster information through the central control screen and generating early warning information to remind a user;
The early warning prompt mode can be that red represents high risk, yellow represents medium risk and blue represents low risk, so that the risk prompt is more visual;
And the safety suggestion matching unit is used for matching the safety suggestions according to the current risk level of the vehicle.
As shown in fig. 2, another embodiment of the present invention provides a vehicle-mounted geological disaster early warning method based on environmental awareness, which is based on the vehicle-mounted geological disaster early warning system based on environmental awareness as described above, and includes the following steps:
S1, acquiring geological disaster information, vehicle information and external environment information in real time, wherein the geological disaster information comprises seismic waveform data and topographic image data, the vehicle information comprises a navigation route, a real-time position of the vehicle, a real-time vehicle speed, historical driving track data of the vehicle and driver preference data, and the external environment information comprises historical traffic data, real-time traffic information and weather data;
s2, predicting a vehicle driving path based on vehicle information and external environment information;
S3, inputting geological disaster information into a pre-trained geological disaster early warning model, and predicting probability values of occurrence of geological disasters of each monitoring point or area;
s4, evaluating the risk level of the geological disaster affecting the vehicle according to the probability value of the geological disaster of each monitoring point or region;
And S5, matching safety suggestions according to the risk level, and carrying out early warning prompt.
Portions of the present application may be implemented in hardware, software, firmware, or a combination thereof. In the above-described embodiments, the various steps or methods may be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. All or part of the steps of the methods of the embodiments described above may be performed by a program that, when executed, comprises one or a combination of the steps of the method embodiments, instructs the associated hardware to perform the method. Embodiments of the present application thus also provide a computer readable storage medium, which may be a read-only memory, a magnetic disk or an optical disk, etc., having stored thereon a computer program which, when executed by a processor, implements an on-board geological disaster warning method based on environmental awareness as described above.
In summary, the method and the system can acquire diversified information in real time, so that the system can more accurately predict the geological disaster risk and respond in time, the vehicle information and the external environment information are combined to predict the driving path of the vehicle, the system is facilitated to identify potential geological disaster risk areas in advance, and accordingly targeted early warning is carried out, a geological disaster early warning model of a multi-task learning framework is adopted, a plurality of targets such as earthquake intensity prediction, landslide risk assessment, debris flow early warning and collapse early warning can be learned at the same time, the generalization performance and prediction precision of the model are improved, the system can more accurately predict the occurrence probability of the geological disaster, the risk level of the influence of the geological disaster on the vehicle is evaluated in real time according to the occurrence probability value of the geological disaster, the real-time risk assessment capability enables the system to respond in time before the occurrence of the disaster, safety advice is provided for a driver, correct decisions are facilitated when the driver faces the potential geological disaster are facilitated, and therefore the driving safety is remarkably improved.
It should be noted that the above embodiments are only for illustrating the technical solution of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention may be modified or substituted without departing from the spirit and scope of the technical solution of the present invention, which is intended to be covered in the scope of the claims of the present invention.