Road network abnormal event identification and risk evaluation early warning method based on vibration sensing

Abstract

The application discloses a road network abnormal event identification and risk evaluation early warning method based on vibration sensing, and relates to the technical field of urban road safety monitoring. The method comprises the following steps: collecting road vibration data of a road network through a sensor; pre-processing the road vibration data through Kalman filtering and multi-dimensional feature extraction; identifying abnormal events from the pre-processed road vibration data through a differentiation algorithm, wherein the abnormal events include vehicle overload, illegal construction and road structure damage; and scoring the abnormal events through dynamic weight distribution, dividing the risk scores according to risk levels and responding to the risk scores in stages. The application adopts a hybrid model architecture and a staged response mechanism through the design of the differentiation algorithm, realizes abnormal event identification, risk level quantitative evaluation and differentiated early warning disposal, and monitors the safety state of the urban road network in real time, thereby providing early warning information.

Description

Technical Field

[0001] This invention relates to the field of urban road safety monitoring technology, and in particular to a method for identifying and assessing risks and providing early warnings of abnormal events in road networks based on vibration sensing. Background Technology

[0002] With the acceleration of urbanization, urban roads face multiple safety threats, including overloaded vehicles, illegal construction disturbances, and structural aging and cracking. Traditional manual inspections and single-point sensor monitoring are insufficient to meet the needs due to incomplete coverage and poor real-time performance. Existing technologies suffer from three bottlenecks: multi-source data (traffic flow, vibration, geology) lacks a spatiotemporal synchronization mechanism, making it difficult to construct a unified risk model; traditional algorithms (such as SVM) are poorly adaptable to complex working conditions and are easily affected by environmental noise; static weight evaluation models cannot dynamically reflect risk evolution; and while vibration sensing technology has the advantages of non-contact and full-section coverage, its engineering application still faces challenges: MEMS sensors are susceptible to temperature drift, low-frequency signals (<10Hz) are easily drowned out by traffic noise, and multi-scale feature fusion algorithms are lacking.

[0003] Therefore, there is an urgent need for a method for identifying and assessing risks of abnormal events in road networks based on vibration sensing, in order to solve the above problems. Summary of the Invention

[0004] The purpose of this invention is to provide a method for identifying and assessing risks of abnormal events in road networks based on vibration sensing, so as to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0006] A method for identifying and assessing risks of abnormal events in road networks based on vibration sensing, the method comprising:

[0007] Step S100: Collect road vibration data of the road network using sensors;

[0008] Step S200: Preprocess the road vibration data using Kalman filtering and multi-dimensional feature extraction;

[0009] Step S300: The processed road vibration data is subjected to anomaly event identification using a differential algorithm. The anomaly events include vehicle overloading, illegal construction, and road structure damage.

[0010] Step S400: Assign risk scores to abnormal events through dynamic weight allocation, classify the risk scores according to the risk level, and respond in a graded manner.

[0011] Furthermore, the specific implementation process for collecting road vibration data of the road network includes,

[0012] The vibration signals of the road network are monitored in real time by sensors. The road vibration data is generated by synchronizing the road network vibration signals with GPS via the PTP protocol. The sensors include an accelerometer, a seismic wave sensor, and a temperature sensor. The temperature sensor is used to collect the ambient temperature and correct the sensor output signal using an adaptive baseline correction algorithm.

[0013] It should be noted that the accelerometer housing has an IP67 protection rating, uses a MEMS triaxial accelerometer, has a measurement range of ±20g, and a noise density of [missing information]. It is installed on the top of the light pole using a magnetic base, 6-8m above the ground, and the installation position avoids areas directly run over by vehicles.

[0014] The seismic wave sensor specifically uses the Geospace GS-11D model, employing a low-frequency detector with a frequency response of 0.1-100Hz. It is buried 30cm deep beside the roadbed, with a distance of ≤5m from the accelerometer. The temperature compensation section uses a PT1000 thermistor to collect ambient temperature data in real time and correct the sensor output signal. The GPS specifically uses the u-blox NEO-M8N model. The edge computing node specifically uses an NVIDIA Jetson Xavier NX.

[0015] The adaptive baseline correction algorithm in the anti-interference design can monitor and compensate for the impact of ambient temperature changes on sensor output in real time, ensuring the accuracy and stability of sensor data. The multi-source data synchronization process achieves precise time synchronization between sensors through the PTP protocol, ensuring that the data collected by different sensors are consistent in time, with the error controlled within 1ms. Combined with the absolute timestamp provided by the GPS module, the occurrence time and location of vibration events can be accurately determined. The LoRaWAN gateway deployed on the edge computing node supports multiple self-organizing network transmissions, enabling wide-area wireless coverage in urban environments with obstructions, with a coverage radius of up to 2km. This effectively solves the data transmission problem, ensuring that the data collected by the sensors can be transmitted to the data center for processing and analysis in a timely and stable manner.

[0016] Furthermore, the specific implementation process of preprocessing the road vibration data includes,

[0017] Kalman filtering is used to preprocess the road vibration data collected by the accelerometer, and features of the road vibration data are extracted from multiple dimensions such as time domain, frequency domain and time-frequency domain. Principal component analysis is used to reduce the dimensionality and extract the first three principal components as fusion features.

[0018] The Kalman filtering process includes:

[0019] Constructing the state vector:

[0020] x k=[a k ,a' k ] T

[0021] Where, x k Let a represent the state vector. k a' represents the real-time measurement value from the accelerometer. k Indicates the rate of change of acceleration;

[0022] According to the state equation x k|k-1 =Fx k-1 Predict the current state, where x k|k-1 Let x represent the predicted state at time k. k-1 Let x represent the state at the previous time step k-1, and F represent the state x used to convert the previous time step k-1. k-1 The predicted state x mapped to the current time k k|k-1 The state transition matrix;

[0023] State transition matrix Where Δt represents the sampling interval;

[0024] Predict the covariance at the current moment: P k|k-1 =FP k-1 F T +Q, where P k|k-1 P represents the predicted covariance at the current time k. k-1 Let Q represent the covariance at the previous time k-1, and let Q represent the process noise covariance matrix used to characterize the dynamic uncertainty of the system, Q=diag(0.01,0.1);

[0025] Combined with the observed value z k Calculate the Kalman gain D k =P k|k-1 H T HP k|k-1 H T +R k ) -1 , where D k Let K represent the Kalman gain at the current time k, H represent the observation matrix, and H = [1, 0], R k Represents the observation noise covariance;

[0026] For the observation noise covariance R k Dynamic adjustments are made using the following formula:

[0027] R k =αR k-1 +(1-α)(z k -Hx k|k-1 ) 2

[0028] Where α represents the smoothing factor, α = 0.95, R k-1 z represents the observation noise covariance at the previous time k-1; k This represents the observation value at the current time k;

[0029] Update state estimate x k =x k|k-1 +D k (z k -Hx k|k-1 ) and covariance matrix P * k =(ID) k H)P k|k-1 Output the smoothed acceleration and rate of change of acceleration, where x k P represents the updated state estimate. * k Let I represent the updated covariance, and let I represent the identity matrix.

[0030] The specific implementation process of the multi-dimensional feature extraction includes:

[0031] Temporal features are extracted along the time axis to construct a temporal feature set. These features include statistical features and waveform indices. The statistical features include mean, peak value, variance, root mean square (RMS), kurtosis, and skewness. The waveform indices include a peak factor C used to quantify the signal impulse intensity. r and the pulse factor L used to identify transient abnormal events r ;

[0032] The formula for the peak factor is as follows: Among them, C r Let Pvl represent the peak value and Rms represent the root mean square value; the formula for the impulse factor is as follows: Among them, L r Me represents the impulse factor, and Me represents the mean.

[0033] The road vibration data is converted to the frequency domain, and the spectrum is obtained through Fourier transform to extract frequency domain features.

[0034] The energy distribution of the 1 / 3 octave band was calculated using Fourier transform, with the center frequency of the band covering 10-500Hz. The energy E of each band was calculated. m for:

[0035]

[0036] Among them, E m Band represents the energy of the m-th frequency band. m X(f) represents the frequency coverage range of the center frequency of the m-th frequency band, and X(f) represents the power spectral density value at the f-th frequency point of the m-th frequency band.

[0037] Extract the power spectral entropy (PSE) in the low-frequency band (0.1-10Hz) that characterizes the spectral complexity of the signal. f :

[0038]

[0039] Among them, PSE f The power spectral entropy in the low-frequency band is represented by p, where M represents the total number of frequency points in the low-frequency band. i E represents the normalized probability distribution of the low-frequency power spectrum. i E represents the energy at the i-th frequency point within the low-frequency band. h This represents the energy at the h-th frequency point within the low-frequency band;

[0040] By combining time and frequency information, time-frequency domain features are extracted:

[0041] Wavelet basis functions are used to perform a 5-level decomposition of the wavelet, and the energy proportion of the detail coefficients in each level is extracted:

[0042]

[0043] Among them, RA j d represents the proportion of the energy of the detail coefficients at level j to the total energy of the detail coefficients. j (t) represents the detail coefficient of the j-th layer, s j (t) represents the energy of the detail coefficients at the j-th level, and t represents the index of the wavelet coefficients;

[0044] A time-frequency energy matrix is ​​constructed, and the first three principal components are extracted as fusion features through principal component analysis to reduce dimensionality.

[0045] It's important to explain that Principal Component Analysis (PCA) is a classic unsupervised linear dimensionality reduction method. It projects high-dimensional data into a low-dimensional space by extracting principal components (i.e., the directions with the largest variance) from the data, while preserving as much key information as possible. The specific implementation process includes: centering each feature (subtracting the mean) and scaling (dividing by the standard deviation) to eliminate the influence of dimensions; calculating the covariance matrix to reflect the correlation between features; decomposing the covariance matrix to obtain eigenvalues ​​and corresponding eigenvectors; sorting the eigenvalues ​​from largest to smallest, and the corresponding eigenvectors are the principal component directions; selecting the eigenvectors corresponding to the largest number of eigenvalues ​​based on the cumulative variance contribution rate (e.g., retaining 95% of the variance) or business needs (e.g., visualization requiring reduction to 2-3 dimensions); and projecting the original data onto the selected principal components to obtain the dimensionality-reduced data.

[0046] Furthermore, the specific implementation process of identifying abnormal events in the processed road vibration data using a differential algorithm includes:

[0047] Abnormal events include vehicle overloading, illegal construction, and road structure damage;

[0048] To address the differences in vibration characteristics across various anomalous events, differentiated algorithms are designed. An optimized SVM is employed to handle the nonlinear classification of overloaded vehicles, and the model's adaptability is enhanced through hybrid kernel functions and incremental learning. For unstructured vibration signals from illegal construction, an improved K-Medoids clustering method combined with Dynamic Time Warping (DTW) is proposed to address the noise sensitivity of traditional K-Means. For the temporal characteristics of road structural damage, a Bi-LSTM+Attention network is constructed, integrating transfer learning to achieve trend prediction under small sample sizes. These three algorithms, from the perspectives of pattern classification, cluster matching, and temporal prediction, respectively, form a comprehensive analytical capability for multi-source road vibration data across all scenarios.

[0049] (1) Use support vector machine to detect overloaded vehicles; Support vector machine (SVM) is a classification method based on statistical learning theory. Its core idea is to find an optimal hyperplane in the feature space to separate data points of different categories as much as possible.

[0050] Feature selection is performed on the processed road vibration data, extracting the root mean square value in the time domain (reflecting the energy of the signal), the energy proportion in the 50-200Hz frequency domain (this frequency band is related to the vibration characteristics of vehicles under overload), and the energy of the wavelet third-level detail coefficients as input features;

[0051] The features are mapped to a high-dimensional feature space using a hybrid kernel function. The specific formula for the hybrid kernel function is as follows:

[0052]

[0053] Among them, xp q XP j xp represents the standardized feature vector. q T xp j Denotes the vector dot product, ||xp q -xp j || 2 γ represents the square of the Euclidean distance. poly and γ rbf σ represents the weights of the polynomial kernel and the radial basis kernel, respectively, and σ represents the parameters of the radial basis kernel;

[0054] The optimal hyperplane is found using the following formula:

[0055] yp im (w T xl im +b)≥1-ξ im ,ξim ≥0, im=1,...,n

[0056] Where w represents the normal vector of the hyperplane, b represents the intercept, C represents the penalty parameter, and ξ represents the normal vector of the hyperplane. i Denotes slack variable, xl im and yp im These represent the feature vector and class label of the i-th sample, respectively; i represents the total number of samples in the hyperplane.

[0057] Deploy an online support vector machine, and incrementally train the model based on newly collected road vibration data in each cycle to update the model parameters. When new data arrives, use the online learning mechanism of SVM to incrementally train the model so that it can adapt to the dynamic changes in vehicle load distribution.

[0058] (2) Detection of illegal construction by combining K-Medoids clustering with dynamic time warping:

[0059] The low-frequency seismic wave signal was resampled, increasing the frequency from 10Hz to 100Hz to obtain richer details. Simultaneously, Gaussian noise (signal-to-noise ratio SNR = 20dB) was added to expand the dataset size and enhance the model's generalization ability.

[0060] The dataset is divided into K clusters, and actual points in the dataset are selected as cluster centers. Manhattan distance is used as the similarity measure, which is more robust to outliers than Euclidean distance. During the clustering process, the selection of medoids is optimized through continuous iteration to minimize the objective function.

[0061] Define a construction feature template, such as a 1-3Hz periodic pulse signal generated by an impact drill. Use a dynamic time warping algorithm to obtain the optimal alignment path between two time series and obtain the similarity between the signal to be detected and the construction feature template.

[0062] When the similarity between the signal to be detected and the construction feature template exceeds the similarity threshold, it is determined to be an illegal construction event;

[0063] (3) Detection of road structure damage by constructing a Bi-LSTM+Attention network:

[0064] Long Short-Term Memory (LSTM) networks are a special type of recurrent neural network (RNN) that solves the gradient vanishing and gradient exploding problems of traditional RNNs by introducing a gating mechanism, enabling them to effectively handle long-term sequential data. An LSTM unit contains an input gate, a forget gate, and an output gate.

[0065] The input layer receives the root mean square value of temporal vibration per unit time; a two-layer bidirectional LSTM is constructed, with each layer containing 64 LSTM units. Weights are assigned to different time steps through a self-attention mechanism, and the output layer outputs the vibration prediction value at future time nodes.

[0066] It needs to be explained that the input layer receives the RMS value of the 60-second time series vibration; a two-layer bidirectional LSTM (Bi-LSTM) is built, with each layer containing 64 LSTM units. The Bi-LSTM can process both forward and reverse time series information at the same time, enhancing the ability to capture contextual information; an attention mechanism is introduced to weight key time steps and highlight important information; the output layer predicts the vibration trend in the next 10 seconds.

[0067] Using Huber as the loss function, the formula is as follows:

[0068]

[0069] Where y represents the true value. The value represents the predicted value, and δ represents the threshold parameter, where δ = 0.1;

[0070] Pre-training is performed based on publicly available bridge vibration datasets to obtain general vibration features; during the fine-tuning stage, the parameters of the two-layer bidirectional LSTM are frozen, and only the Attention layer and the fully connected layer are trained. Using road structure damage data, the road structure damage results are output.

[0071] The specific implementation process of risk scoring for abnormal events through dynamic weight allocation, and the classification and tiered response based on risk levels, includes: constructing a dynamic weight allocation model and a tiered response system to achieve closed-loop management from risk quantification to handling. First, the entropy weight method-AHP fusion model is used to address the dynamic adjustment of weights for three types of risks: overload, construction, and structural damage, introducing a time decay factor to reflect the timeliness of risk. Second, a four-level early warning response mechanism is designed, matching differentiated handling strategies through threshold classification (<30 / 30-60 / 60-90 / ≥90). Finally, relying on a GIS visualization dashboard, multi-dimensional information such as spatial location, event type, and historical trends is integrated to form a full-link technical architecture of "data collection - feature analysis - risk assessment - visual decision-making," supporting the refined management of urban road safety.

[0072] Dynamic weight allocation in risk assessment comprehensively considers both time factors and subjective / objective weights. Risk weights are adjusted based on a time decay factor; taking overload event weights as an example, the formula is w. overload =w0e -λtWhere w0 is the initial weight, λ = 0.01 / hour is the decay coefficient, and t is time. The time decay factor reflects the time-sensitivity of the risk; as time goes on, the impact of the risk gradually decreases.

[0073] The objective weight of the index is calculated using the entropy weight method. The basic principle of the entropy weight method is to determine the weight based on the degree of variation of the index. The greater the degree of variation, the higher the weight.

[0074] Based on the results of abnormal event identification, the objective weights of the indicators are calculated using the entropy weight method:

[0075]

[0076] Among them, pp uo g represents the proportion of the u-th sample under the o-th indicator. uo Let g represent the standardized index of the u-th sample under the 0-th index. Ψo Let represent the standardized index of the Ψ-th sample under the 0-th index, where zi represents the sample size; e o This represents the entropy value of the o-th index; pp τo HL represents the proportion of the τ-th sample under the ω-th indicator. entropy,o Oi represents the objective weight of the o-th indicator, and e represents the number of indicators. η This represents the entropy value of the ηth index;

[0077] By combining the analytic hierarchy process (AHP) to determine subjective weights, constructing a judgment matrix, and comparing each indicator pairwise, the relative weights HL are obtained. AHP,o ;

[0078] The final weight is calculated by combining subjective and objective weights: LG o =βHL entropy,o +(1-β)HL AHP,o Among them, LG o β represents the final weight of the 0th indicator, and β represents the balance coefficient.

[0079] A risk score is obtained based on the final weighting, such as: Where PFS represents the risk score, X * uo This represents the standardized value of the o-th indicator for the u-th sample;

[0080] The risk score is divided into risk levels, including low risk, medium risk, high risk, and very high risk.

[0081] When the risk level is low (score <30), relevant information will be generated into a log and stored in the database without issuing an alert, so as to facilitate subsequent analysis and tracing.

[0082] When the risk level is medium risk (30≤score<60), the risk information will be sent to the maintenance personnel in the jurisdiction via SMS so that they can be informed of the road conditions in a timely manner and take corresponding preventive measures.

[0083] When the risk level is high (60≤score<90), an audible and visual alarm, such as a road LED warning screen, is triggered to visually alert people in the surrounding area to road risks and to draw the attention of relevant departments.

[0084] When the risk level is extremely high (score ≥ 90), not only will an audible and visual alarm be triggered, but the traffic signal system will also be activated to restrict vehicle passage, prevent potential dangers from escalating further, and promptly report to the emergency management platform so that relevant departments can coordinate and handle the emergency.

[0085] It should also be explained that users can perform drill-down analysis on the dashboard using multiple dimensions such as vibration energy, event type, and historical trends. For example, by viewing the vibration energy heatmap, users can intuitively understand the vibration intensity distribution in different areas of the road and quickly locate potential risk areas; filtering by event type allows users to clearly see the location and frequency of various abnormal events, providing a basis for decision-making in road safety management; and analyzing historical trends helps to discover patterns in changes in road safety conditions, enabling users to take targeted measures for prevention and maintenance in advance.

[0086] A road network abnormal event identification and risk assessment and early warning system based on vibration sensing includes a sensor deployment and data acquisition module, a data preprocessing module, an abnormal event module, and a risk assessment and early warning module.

[0087] The sensor deployment and data acquisition module is used to collect road vibration data; it deploys accelerometers, seismic wave sensors, and temperature sensors. The accelerometers and seismic wave sensors monitor the vibration signals of the road network in real time, and the temperature sensors collect the ambient temperature and correct the sensor output signals using an adaptive baseline correction algorithm; it generates road vibration data by synchronizing the road network vibration signals with GPS via the PTP protocol.

[0088] The data preprocessing module is used to preprocess road vibration data by improving the Kalman filter parameter design and multi-dimensional feature extraction; it preprocesses road vibration data collected by acceleration sensors through Kalman filtering, and extracts features of road vibration data from the time domain, frequency domain, and time-frequency domain through multi-dimensional feature extraction.

[0089] The abnormal event identification module is used to identify abnormal events in the processed road vibration data using a differential algorithm; to detect overloaded vehicles using an optimized support vector machine; to detect illegal construction using K-Medoids clustering combined with dynamic time warping; and to detect road structure damage by constructing a Bi-LSTM+Attention network.

[0090] The risk assessment and early warning module is used to score abnormal events through dynamic weight allocation, classify the risk scores according to the risk level, and respond in a graded manner.

[0091] A terminal device includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor. The memory stores the computer program that can run on the processor. When the processor loads and executes the computer program, it employs a vibration-sensing-based method for identifying and assessing and warning of abnormal road network events.

[0092] A storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform the vibration-sensing-based road network anomaly event identification and risk assessment early warning method.

[0093] Compared with the prior art, the beneficial effects achieved by the present invention are: by optimizing SVM, improving K-Medoids clustering and Bi-LSTM network, it can realize the classification of overloaded vehicles, matching of illegal construction patterns and prediction of road structure damage trends;

[0094] By integrating the entropy weight method-AHP dynamic weight model with a graded response mechanism, we can achieve quantitative assessment of risk levels and differentiated early warning and handling.

[0095] It also has a high level of intelligence, enabling it to monitor the safety status of urban road networks in real time, provide early warning information, and is suitable for road risk assessment in a variety of complex scenarios. Attached Figure Description

[0096] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0097] Figure 1 This is a schematic diagram illustrating the steps of the method for identifying and assessing risks of abnormal road network events based on vibration sensing, as described in this invention. Detailed Implementation

[0098] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0099] Please see Figure 1 This invention provides a technical solution: a method for identifying and assessing risks of abnormal events in road networks based on vibration sensing, the method comprising:

[0100] Road vibration data of the road network is collected using sensors;

[0101] Road vibration data were preprocessed using Kalman filtering and multi-dimensional feature extraction.

[0102] The processed road vibration data is analyzed using a differential algorithm to identify abnormal events, including vehicle overloading, illegal construction, and road structure damage.

[0103] Anomalies are scored using dynamic weight allocation, and risk scores are categorized and responded to in stages based on risk level.

[0104] Specifically, the implementation process for collecting road vibration data of the road network includes:

[0105] The vibration signals of the road network are monitored in real time by sensors. The road vibration data is generated by synchronizing the road network vibration signals with GPS via the PTP protocol. The sensors include an accelerometer, a seismic wave sensor, and a temperature sensor. The temperature sensor is used to collect the ambient temperature and correct the sensor output signal using an adaptive baseline correction algorithm.

[0106] Specifically, the preprocessing of road vibration data includes the following steps:

[0107] Kalman filtering is used to preprocess the road vibration data collected by the accelerometer, and features of the road vibration data are extracted from multiple dimensions such as time domain, frequency domain and time-frequency domain. Principal component analysis is used to reduce the dimensionality and extract the first three principal components as fusion features.

[0108] The Kalman filtering process includes:

[0109] Constructing the state vector:

[0110] x k =[a k ,a' k ] T

[0111] Where, x k Let a represent the state vector. k a' represents the real-time measurement value from the accelerometer. k Indicates the rate of change of acceleration;

[0112] According to the state equation x k|k-1 =Fx k-1 Predict the current state, where x k|k-1 Let x represent the predicted state at time k. k-1 Let x represent the state at the previous time step k-1, and F represent the state x used to convert the previous time step k-1. k-1 The predicted state x mapped to the current time k k|k-1 The state transition matrix;

[0113] State transition matrix Where Δt represents the sampling interval;

[0114] Predict the covariance at the current moment: P k|k-1 =FP k-1 F T +Q, where P k|k-1 P represents the predicted covariance at the current time k. k-1 Let Q represent the covariance at the previous time k-1, and let Q represent the process noise covariance matrix used to characterize the dynamic uncertainty of the system, Q=diag(0.01,0.1);

[0115] Combined with the observed value z k Calculate the Kalman gain D k =P k|k-1 H T HP k|k-1 H T +R k ) -1 , where D k Let K represent the Kalman gain at the current time k, H represent the observation matrix, and H = [1, 0], R k Represents the observation noise covariance;

[0116] For the observation noise covariance R k Dynamic adjustments are made using the following formula:

[0117] R k =αR k-1 +(1-α)(z k -Hx k|k-1 ) 2

[0118] Where α represents the smoothing factor, α = 0.95, R k-1 z represents the observation noise covariance at the previous time k-1; kThis represents the observation value at the current time k;

[0119] Update state estimate x k =x k|k-1 +D k (z k -Hx k|k-1 ) and covariance matrix P * k =(ID) k H)P k|k-1 Output the smoothed acceleration and rate of change of acceleration, where x k P represents the updated state estimate. * k Let I represent the updated covariance, and let I represent the identity matrix.

[0120] Specifically, the implementation process of the multi-dimensional feature extraction includes:

[0121] Temporal features are extracted along the time axis to construct a temporal feature set. These features include statistical features and waveform indices. The statistical features include mean, peak value, variance, root mean square (RMS), kurtosis, and skewness. The waveform indices include a peak factor C used to quantify the signal impulse intensity. r and the pulse factor L used to identify transient abnormal events r ;

[0122] The formula for the peak factor is as follows: Among them, C r Let Pvl represent the peak value and Rms represent the root mean square value; the formula for the impulse factor is as follows: Among them, L r Me represents the impulse factor, and Me represents the mean.

[0123] The road vibration data is converted to the frequency domain, and the spectrum is obtained through Fourier transform to extract frequency domain features.

[0124] The energy distribution of the 1 / 3 octave band was calculated using Fourier transform, with the center frequency of the band covering 10-500Hz. The energy E of each band was calculated. m for:

[0125]

[0126] Among them, E m Band represents the energy of the m-th frequency band. m X(f) represents the frequency coverage range of the center frequency of the m-th frequency band, and X(f) represents the power spectral density value at the f-th frequency point of the m-th frequency band.

[0127] Extract the power spectral entropy (PSE) in the low-frequency band (0.1-10Hz) that characterizes the spectral complexity of the signal. f:

[0128]

[0129] Among them, PSE f The power spectral entropy in the low-frequency band is represented by p, where M represents the total number of frequency points in the low-frequency band. i E represents the normalized probability distribution of the low-frequency power spectrum. i E represents the energy at the i-th frequency point within the low-frequency band. h This represents the energy at the h-th frequency point within the low-frequency band;

[0130] By combining time and frequency information, time-frequency domain features are extracted:

[0131] Wavelet basis functions are used to perform a 5-level decomposition of the wavelet, and the energy proportion of the detail coefficients in each level is extracted:

[0132]

[0133] Among them, RA j d represents the proportion of the energy of the detail coefficients at level j to the total energy of the detail coefficients. j (t) represents the detail coefficient of the j-th layer, s j (t) represents the energy of the detail coefficients at the j-th level, and t represents the index of the wavelet coefficients;

[0134] A time-frequency energy matrix is ​​constructed, and the first three principal components are extracted as fusion features through principal component analysis to reduce dimensionality.

[0135] Specifically, the implementation process of identifying abnormal events in processed road vibration data using a differential algorithm includes:

[0136] Abnormal events include vehicle overloading, illegal construction, and road structure damage;

[0137] To address the differences in vibration characteristics across various abnormal events, differentiated algorithms were designed. An optimized support vector machine was used to detect overloaded vehicles; K-Medoids clustering combined with dynamic time warping was used to detect illegal construction; and a Bi-LSTM+Attention network was constructed to detect road structure damage.

[0138] An optimized support vector machine is used to detect overloaded vehicles;

[0139] Feature selection was performed on the processed road vibration data, and the root mean square value in the time domain, the energy proportion in the 50-200Hz frequency domain, and the energy of the third level detail coefficients of the wavelet were extracted as input features.

[0140] The features are mapped to a high-dimensional feature space using a hybrid kernel function. The specific formula for the hybrid kernel function is as follows:

[0141]

[0142] Among them, xp q XP j xp represents the standardized feature vector. q T xp j Denotes the vector dot product, ||xp q -xp j || 2 γ represents the square of the Euclidean distance. poly and γ rbf σ represents the weights of the polynomial kernel and the radial basis kernel, respectively, and σ represents the parameters of the radial basis kernel;

[0143] The optimal hyperplane is found using the following formula:

[0144] yp im (w T xl im +b)≥1-ξ im ,ξ im ≥0, im=1,...,n

[0145] Where w represents the normal vector of the hyperplane, b represents the intercept, C represents the penalty parameter, and ξ represents the normal vector of the hyperplane. i Denotes slack variable, xl im and yp im These represent the feature vector and class label of the i-th sample, respectively; i represents the total number of samples in the hyperplane.

[0146] Deploy an online support vector machine, and incrementally train the model each cycle based on newly collected road vibration data to update the model parameters;

[0147] Detection of illegal construction using K-Medoids clustering combined with dynamic time warping:

[0148] The low-frequency seismic wave signal was resampled and Gaussian noise was added to expand the dataset;

[0149] The dataset is divided into K clusters, and actual points in the dataset are selected as cluster centers. Manhattan distance is used as a similarity measure. Through continuous iteration, the selection of cluster centers is optimized to minimize the objective function.

[0150] A construction feature template is defined, and the optimal alignment path between two time series is obtained through a dynamic time warping algorithm, thus obtaining the similarity between the signal to be detected and the construction feature template.

[0151] When the similarity between the signal to be detected and the construction feature template exceeds the similarity threshold, it is determined to be an illegal construction event;

[0152] By constructing a Bi-LSTM+Attention network, road structural damage can be detected.

[0153] The input layer receives the root mean square value of temporal vibration per unit time; a two-layer bidirectional LSTM is constructed, with each layer containing 64 LSTM units. Weights are assigned to different time steps through a self-attention mechanism, and the output layer outputs the vibration prediction value at future time nodes.

[0154] Using Huber as the loss function, the formula is as follows:

[0155]

[0156] Where y represents the true value. The value represents the predicted value, and δ represents the threshold parameter, where δ = 0.1;

[0157] Pre-training is performed based on publicly available bridge vibration datasets to obtain general vibration features; during the fine-tuning stage, the parameters of the two-layer bidirectional LSTM are frozen, and only the Attention layer and the fully connected layer are trained. Using road structure damage data, the road structure damage results are output.

[0158] Specifically, the process of dynamically assigning weights to score the risk of abnormal events, classifying the risk scores according to risk levels, and implementing tiered responses includes:

[0159] Based on the results of abnormal event identification, the objective weights of the indicators are calculated using the entropy weight method:

[0160]

[0161] Among them, pp uo g represents the proportion of the u-th sample under the o-th indicator. uo Let g represent the standardized index of the u-th sample under the 0-th index. Ψo Let represent the standardized index of the Ψ-th sample under the 0-th index, where zi represents the sample size; e o This represents the entropy value of the o-th index; pp τo HL represents the proportion of the τ-th sample under the ω-th indicator. entropy,o Oi represents the objective weight of the o-th indicator, and e represents the number of indicators. η This represents the entropy value of the ηth index;

[0162] By combining the analytic hierarchy process (AHP) to determine subjective weights, constructing a judgment matrix, and comparing each indicator pairwise, the relative weights HL are obtained. AHP,o ;

[0163] The final weight is calculated by combining subjective and objective weights: LGo =βHL entropy,o +(1-β)HL AHP,o Among them, LG o β represents the final weight of the 0th indicator, and β represents the balance coefficient.

[0164] A risk score is obtained based on the final weighting.

[0165] The risk score is categorized based on risk level, which includes low risk, medium risk, high risk, and very high risk.

[0166] When the risk level is low (score <30), relevant information will be generated into a log and stored in the database without issuing an alert, so as to facilitate subsequent analysis and tracing.

[0167] When the risk level is medium risk (30≤score<60), the risk information will be sent to the maintenance personnel in the jurisdiction via SMS so that they can be informed of the road conditions in a timely manner and take corresponding preventive measures.

[0168] When the risk level is high (60≤score<90), an audible and visual alarm, such as a road LED warning screen, is triggered to visually alert people in the surrounding area to road risks and to draw the attention of relevant departments.

[0169] When the risk level is extremely high (score ≥ 90), not only will an audible and visual alarm be triggered, but the traffic signal system will also be activated to restrict vehicle passage, prevent potential dangers from escalating further, and promptly report to the emergency management platform so that relevant departments can coordinate and handle the emergency.

[0170] A road network abnormal event identification and risk assessment and early warning system based on vibration sensing includes a sensor deployment and data acquisition module, a data preprocessing module, an abnormal event module, and a risk assessment and early warning module.

[0171] The sensor deployment and data acquisition module is used to collect road vibration data; it deploys accelerometers, seismic wave sensors, and temperature sensors. The accelerometers and seismic wave sensors monitor the vibration signals of the road network in real time, and the temperature sensors collect the ambient temperature and correct the sensor output signals using an adaptive baseline correction algorithm; it generates road vibration data by synchronizing the road network vibration signals with GPS via the PTP protocol.

[0172] The data preprocessing module is used to preprocess road vibration data by improving the Kalman filter parameter design and multi-dimensional feature extraction; it preprocesses road vibration data collected by acceleration sensors through Kalman filtering, and extracts features of road vibration data from the time domain, frequency domain, and time-frequency domain through multi-dimensional feature extraction.

[0173] The abnormal event identification module is used to identify abnormal events in the processed road vibration data using a differential algorithm; to detect overloaded vehicles using an optimized support vector machine; to detect illegal construction using K-Medoids clustering combined with dynamic time warping; and to detect road structure damage by constructing a Bi-LSTM+Attention network.

[0174] The risk assessment and early warning module is used to score abnormal events through dynamic weight allocation, classify the risk scores according to the risk level, and respond in a graded manner.

[0175] Specifically, the aforementioned sensor deployment and data acquisition module, data preprocessing module, abnormal event identification module, and risk assessment and early warning module can be embedded into a computer processing system. The computer, based on the vibration-based road network abnormal event identification and risk assessment and early warning method provided above, calls upon each of the above modules to complete the task of abnormal event identification and risk assessment and early warning. The aforementioned sensor deployment and data acquisition module, data preprocessing module, abnormal event identification module, and risk assessment and early warning module can perform operations according to the specific steps given in the vibration-based road network abnormal event identification and risk assessment and early warning method.

[0176] It should be noted that the division of the various modules in the above system is merely a division of logical functions. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. These modules can be implemented entirely in software through processing element calls; they can be fully implemented in hardware; or some modules can be implemented by processing element calls to software, while others are implemented in hardware. For example, the receiving module can be a separate processing element or integrated into a chip in the aforementioned device. Alternatively, it can be stored as program code in the device's memory, called and executed by a processing element. The implementation of other modules is similar. Furthermore, these modules can be fully or partially integrated together or implemented independently. The processing element here can be an integrated circuit with signal processing capabilities. During implementation, the steps of the above method or the various modules can be completed through integrated logic circuits in the processor element or software instructions.

[0177] For example, these modules can be one or more integrated circuits configured to implement the above methods, such as one or more Application Specific Integrated Circuits (ASICs), one or more Digital Signal Processors (DSPs), or one or more Field Programmable Gate Arrays (FPGAs). As another example, when a module is implemented using processing element scheduler code, the processing element can be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor capable of calling program code. Furthermore, these modules can be integrated together to form a system-on-a-chip (SOC).

[0178] A terminal device includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor. The memory stores the computer program that can run on the processor. When the processor loads and executes the computer program, it employs a vibration-sensing-based method for identifying and assessing and warning of abnormal road network events.

[0179] It should be noted that the terminal device can be a computer device such as a desktop computer, a laptop computer, or a cloud server, and the terminal device includes, but is not limited to, a processor and a memory. For example, the terminal device may also include input / output devices, network access devices, and buses.

[0180] Specifically, the processor can be a central processing unit (CPU). Of course, depending on the actual use, other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. can also be used. The general-purpose processor can be a microprocessor or any conventional processor, etc. This application does not limit this.

[0181] A storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform the aforementioned vibration-sensing-based road network anomaly event identification and risk assessment early warning method.

[0182] The computer program can be stored in a computer-readable medium. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or certain middleware. The computer-readable medium includes any entity or device capable of carrying computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the computer-readable medium includes, but is not limited to, the above-mentioned components.

[0183] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0184] For those skilled in the art, the specific meaning of the above terms in this invention can be understood according to the specific circumstances. When an element is referred to as being "assembled on," "mounted on," "fixed to," or "set on" another element, it may be directly on the other element or there may be an intermediate element present. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be an intermediate element present. The terms "vertical," "horizontal," "upper," "lower," "left," "right," and similar expressions used herein are for illustrative purposes only and do not represent the only possible embodiments.

[0185] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0186] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

Claims

1. A method for identifying and assessing risks and providing early warning of abnormal events in road networks based on vibration sensing, characterized in that, The method includes: Road vibration data of the road network is collected using sensors; Road vibration data were preprocessed using Kalman filtering and multi-dimensional feature extraction. Anomaly events, including vehicle overloading, illegal construction, and road structure damage, are identified in the processed road vibration data using a differential algorithm. Overloaded vehicles are detected using an optimized support vector machine. Illegal construction is detected using K-Medoids clustering combined with dynamic time warping. Road structure damage is detected by constructing a Bi-LSTM+Attention network. Optimized support vector machines are used to detect overloaded vehicles, including: Feature selection was performed on the processed road vibration data, and the root mean square value in the time domain, the energy proportion in the 50-200Hz frequency domain, and the energy of the third level detail coefficients of the wavelet were extracted as input features. The features are mapped to a high-dimensional feature space using a hybrid kernel function. The specific formula for the hybrid kernel function is as follows: ； in, This represents the standardized feature vector. Represents the vector dot product. This represents the square of the Euclidean distance. and These represent the weights of the polynomial kernel and the radial basis kernel, respectively. Parameters representing the radial basis core; The optimal hyperplane is found using the following formula: ; ； Where w represents the normal vector of the hyperplane, b represents the intercept, and C represents the penalty parameter. Denotes slack variables, and These represent the feature vector and class label of the i-th sample, respectively; n represents the total number of samples in the hyperplane. Deploy an online support vector machine, and incrementally train the model each cycle based on newly collected road vibration data to update the model parameters; Illegal construction is detected using K-Medoids clustering combined with dynamic time warping, including: The low-frequency seismic wave signal was resampled and Gaussian noise was added to expand the dataset; The dataset is divided into K clusters, and actual points in the dataset are selected as cluster centers. Manhattan distance is used as a similarity measure. Through continuous iteration, the selection of cluster centers is optimized to minimize the objective function. A construction feature template is defined, and the optimal alignment path between two time series is obtained through a dynamic time warping algorithm, thus obtaining the similarity between the signal to be detected and the construction feature template. When the similarity between the signal to be detected and the construction feature template exceeds the similarity threshold, it is determined to be an illegal construction event; By constructing a Bi-LSTM+Attention network, road structural damage is detected, including: The input layer receives the root mean square value of temporal vibration per unit time; a two-layer bidirectional LSTM is constructed, with each layer containing 64 LSTM units. Weights are assigned to different time steps through a self-attention mechanism, and the output layer outputs the vibration prediction value at future time nodes. Using Huber as the loss function, the formula is as follows: ； Where y represents the true value. Indicates the predicted value. Indicates the threshold parameter. =0.1; Pre-training is performed based on publicly available bridge vibration datasets to obtain general vibration features; during the fine-tuning stage, the parameters of the two-layer bidirectional LSTM layer are frozen, and only the Attention layer and the fully connected layer are trained. Road structure damage data is used to output road structure damage results. Anomalies are scored using dynamic weight allocation, and risk scores are categorized and responded to in stages based on risk level.

2. The method for identifying and assessing risks of abnormal road network events based on vibration sensing according to claim 1, characterized in that, The specific implementation process for collecting road vibration data of the road network includes: The vibration signals of the road network are monitored in real time by sensors. The road vibration data is generated by synchronizing the road network vibration signals with GPS via the PTP protocol. The sensors include an accelerometer, a seismic wave sensor, and a temperature sensor. The temperature sensor is used to collect the ambient temperature and correct the sensor output signal using an adaptive baseline correction algorithm.

3. The method for identifying and assessing risks of abnormal road network events based on vibration sensing according to claim 2, characterized in that, The specific implementation process for preprocessing road vibration data includes: Kalman filtering is used to preprocess the road vibration data collected by the accelerometer, and features of the road vibration data are extracted from multiple dimensions such as time domain, frequency domain and time-frequency domain. Principal component analysis is used to reduce the dimensionality and extract the first three principal components as fusion features. The Kalman filtering process includes: Constructing the state vector: ； in, Represents the state vector. This represents the real-time measurement value from the accelerometer. Indicates the rate of change of acceleration; According to the state equation Predict the current state, where, This represents the predicted state at the current time k. This represents the state at the previous time step k-1, and F represents the state used to set the previous time step k-1. Mapping to the predicted state at time k The state transition matrix; State transition matrix ,in, Indicates the sampling interval; Predict the covariance at the current moment: ,in, This represents the prediction covariance at the current time k. Let Q represent the covariance at the previous time step k-1, and let Q represent the process noise covariance matrix used to characterize the dynamic uncertainty of the system. ; Combined with observations Calculate Kalman gain ,in, H represents the Kalman gain at time k, and H represents the observation matrix. Represents the observation noise covariance; observation noise covariance Dynamic adjustments are made using the following formula: ； in, Represents the smoothing factor. This represents the observation noise covariance at the previous time step k-1; This represents the observation value at the current time k; Update state estimation Covariance Matrix Output the smoothed acceleration and rate of change of acceleration, where, Let I represent the updated covariance, and let I represent the identity matrix.

4. The method for identifying and assessing risks of abnormal road network events based on vibration sensing according to claim 3, characterized in that, The specific implementation process of the multi-dimensional feature extraction includes: Temporal features are extracted along the time axis to construct a temporal feature set. These features include statistical features and waveform indices. The statistical features include mean, peak value, variance, root mean square (RMS), kurtosis, and skewness. The waveform indices include a peak factor used to quantify the signal impulse intensity. and the pulse factor used to identify transient abnormal events ; The formula for the peak factor is as follows: ,in, Let Pvl represent the peak value and Rms represent the root mean square value; the formula for the impulse factor is as follows: ,in, Me represents the impulse factor, and Me represents the mean. The road vibration data is converted to the frequency domain, and the spectrum is obtained through Fourier transform to extract frequency domain features. The energy distribution of the 1 / 3 octave band was calculated using Fourier transform, with the center frequency of the band covering 10-500Hz. The energy distribution of each band was calculated. for: ； in, This represents the energy of the m-th frequency band. This represents the frequency coverage range of the center frequency of the m-th frequency band. This represents the power spectral density value at the f-th frequency point in the m-th frequency band; Extracting the power spectral entropy in the low-frequency band that characterizes the spectral complexity of a signal. : ； ； in, This represents the power spectral entropy in the low-frequency band, where M represents the total number of frequency points in the low-frequency band. This represents the normalized probability distribution of the power spectrum in the low-frequency band. This represents the energy at the i-th frequency point within the low-frequency band. This represents the energy at the h-th frequency point within the low-frequency band; By combining time and frequency information, time-frequency domain features are extracted: Wavelet basis functions are used to perform a 5-level decomposition of the wavelet, and the energy proportion of the detail coefficients in each level is extracted: ； in, This represents the proportion of the energy of the level j detail coefficients to the total energy of the detail coefficients. Represents the detail coefficients of the j-th layer. Let represent the energy of the detail coefficients at the j-th level, and t represent the index of the wavelet coefficients; A time-frequency energy matrix is ​​constructed, and the first three principal components are extracted as fusion features through principal component analysis to reduce dimensionality.

5. The method for identifying and assessing risks of abnormal road network events based on vibration sensing according to claim 4, characterized in that, The specific implementation process of scoring abnormal events through dynamic weight allocation, classifying risk scores according to risk levels, and responding in a graded manner includes: Based on the results of abnormal event identification, the objective weights of the indicators are calculated using the entropy weight method: ； ； ； in, This represents the proportion of the u-th sample under the 0-th indicator. This represents the standardized index of the u-th sample under the 0-th index. This indicates the 0th indicator. The standardized index of a sample, where zi represents the number of samples; This represents the entropy value of the 0th index; This indicates the 0th indicator. The proportion of each sample Oi represents the objective weight of the 0th indicator, and Oi represents the number of indicators. Indicates the first The entropy value of each indicator; By combining the analytic hierarchy process (AHP) to determine subjective weights, constructing a judgment matrix, and comparing each indicator pairwise to obtain relative weights, we can obtain the relative weights. ; The final weight is calculated by combining subjective and objective weights. ,in, This represents the final weight of the o-th indicator. Represents the balance coefficient. Indicates relative weight; A risk score is obtained based on the final weighting. The risk score is divided into risk levels, including low risk, medium risk, high risk and very high risk. When the risk level is low risk, the relevant information is generated into a log and recorded in the database. When the risk level is medium, the risk information will be sent to the maintenance personnel in the jurisdiction; When the risk level is high, the audible and visual alarm will be triggered. When the risk level is extremely high, an audible and visual alarm is triggered, the traffic signal system is activated, and the incident is reported to the emergency management platform.

6. A road network anomaly event identification and risk assessment early warning system based on vibration sensing, using the road network anomaly event identification and risk assessment early warning method based on vibration sensing as described in any one of claims 1-5, characterized in that, The system includes a sensor deployment and data acquisition module, a data preprocessing module, an anomaly identification module, and a risk assessment and early warning module. The sensor deployment and data acquisition module is used to collect road vibration data; it deploys accelerometers, seismic wave sensors, and temperature sensors. The accelerometers and seismic wave sensors monitor the vibration signals of the road network in real time, the temperature sensors detect the ambient temperature, and the sensor output signals are corrected using an adaptive baseline correction algorithm; the vibration signals of the road network are synchronized with GPS via the PTP protocol to generate road vibration data. The data preprocessing module is used to preprocess road vibration data by improving the Kalman filter parameter design and multi-dimensional feature extraction; it preprocesses road vibration data collected by acceleration sensors through Kalman filtering and extracts features of road vibration data from the time domain, frequency domain, and time-frequency domain through multi-dimensional feature extraction; The abnormal event identification module is used to identify abnormal events in the processed road vibration data using a differential algorithm; to detect overloaded vehicles using an optimized support vector machine; to detect illegal construction using K-Medoids clustering combined with dynamic time warping; and to detect road structure damage by constructing a Bi-LSTM+Attention network. The risk assessment and early warning module is used to score abnormal events through dynamic weight allocation, classify the risk scores according to the risk level, and respond in a graded manner.

7. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that, The memory stores a computer program that can run on the processor. When the processor loads and executes the computer program, it employs the vibration-sensing-based road network abnormal event identification and risk assessment early warning method as described in any one of claims 1 to 6.

8. A storage medium containing computer-executable instructions, characterized in that, The computer-executable instructions, when executed by a computer processor, are used to perform the vibration-sensing-based road network abnormal event identification and risk assessment early warning method as described in any one of claims 1 to 6.

Images