A road surface icing state monitoring system and method based on multi-source data fusion

Abstract

The application belongs to the technical field of road traffic safety monitoring, and discloses a road surface icing state monitoring system and method based on multi-source data fusion, which comprises a vibration sensing module, a meteorological and road surface state sensing module and a data fusion and processing module. The meteorological and road surface state sensing module is used for identifying potential icing risk of a road surface. The vibration sensing module is used for detecting the icing state of the road surface with potential icing risk. The data fusion and processing module is used for estimating the thickness of an ice layer after confirming icing. The application creatively fuses direct contact sensing based on structural dynamics, indirect models based on meteorological physics, and introduces a data fusion algorithm, so as to solve the inherent defects of a single technical path by using a lightweight system.

Description

Technical Field

[0001] This invention belongs to the field of road traffic safety monitoring technology, specifically relating to a road surface icing status monitoring system and method based on multi-source data fusion. Background Technology

[0002] Road icing, especially "black ice" on bridge surfaces in winter, poses a serious traffic safety hazard. Existing road icing monitoring technologies have the following limitations: Pure model-based early warning systems (such as the dew point temperature difference method) lack reliability: they rely solely on meteorological parameters (temperature and humidity) for calculation, making them indirect predictions. Their accuracy is greatly affected by factors such as local microclimate, road surface materials, and thermal inertia, making them prone to false alarms or missed alarms, and they cannot provide crucial information such as ice thickness.

[0003] Direct measurement sensors are costly and complex to deploy: high-precision microwave radar, laser thickness gauges and other non-contact equipment are expensive; although embedded capacitor and fiber optic sensors can measure directly, they require road surface damage for installation, resulting in high single-point costs and difficult maintenance, making it difficult to popularize them in large-scale road networks.

[0004] Existing sensing methods have poor environmental adaptability: cameras are easily affected by rain, snow, fog, haze, and nighttime light; the performance of some ultrasonic or infrared sensors drops sharply under complex conditions such as road surface dirt and snow cover.

[0005] Therefore, the industry urgently needs a cost-effective monitoring solution that can balance economy, reliability, and maintainability, and can achieve a progressive approach from "risk warning" to "status confirmation" and then to "thickness estimation". Summary of the Invention

[0006] To address the contradiction in existing technologies where "low-cost solutions are unreliable and reliable solutions are costly," this invention provides a road surface icing status monitoring system and method based on multi-source data fusion. It creatively integrates direct contact sensing based on structural dynamics and indirect models based on meteorological physics, and introduces a data fusion algorithm to solve the inherent defects of a single technical approach with a lightweight system.

[0007] To achieve the above objectives, the present invention provides the following solution: A road surface icing status monitoring system based on multi-source data fusion, the system comprising: a vibration sensing module, a meteorological and road surface status sensing module, and a data fusion and processing module; The weather and road condition sensing module is used to identify potential road icing risks; The vibration sensing module is used to detect the icing status of road surfaces with potential icing risks. The data fusion and processing module is used to estimate the ice thickness after icing is confirmed.

[0008] Preferably, the weather and road condition sensing module includes a temperature and humidity sensor deployed on the roadside guardrail and a temperature sensor buried under the road surface. The temperature and humidity sensor deployed on the roadside guardrail is used to acquire air temperature and humidity. The temperature sensor embedded in the road surface is used to obtain the road surface temperature.

[0009] Preferably, the vibration sensing module includes a metal rod-shaped structure fixed to the road surface and a vibration sensor, a waterproof sealing layer, a signal transmission line, and a fixing base attached thereto; The vibration sensor is used to sense the micro-vibrations of the rod. The waterproof sealing layer is used to cover the sensor; The signal transmission line is used to transmit the vibration electrical signal to the data acquisition unit; The fixed base is used to rigidly fix the bottom of the pole to the roadbed through flanges and chemical anchors.

[0010] This invention also provides a method for monitoring road surface icing status based on multi-source data fusion. The method is implemented using the aforementioned system and includes: Identify potential road surface icing risks; Detect the icing status of road surfaces with potential icing risk; After confirming icing, the ice thickness is estimated.

[0011] Preferred methods for identifying potential road surface icing risks include: Real-time calculation of the difference between the road surface temperature T_s and the dew point temperature T_d estimated from the air temperature and humidity. ; When satisfied When the temperature is below the preset threshold and T_s < 1℃, a Level 1 warning is triggered, indicating a risk of icing.

[0012] Preferred methods for calculating the dew point temperature T_d from air temperature and humidity include: T_d=(243.5 ln(e / 6.112)) / (17.67-ln(e / 6.112)); Where e is the actual water vapor pressure; Real-time calculation of the difference between the road surface temperature T_s and the dew point temperature T_d estimated from the air temperature and humidity. The methods include: ΔT = T_s - T_d.

[0013] Preferably, methods for detecting the icing status of road surfaces at potential icing risk include: After the first-level warning is triggered, the vibration spectrum of the metal rod is analyzed and its fundamental frequency is extracted; If the current base frequency f_current is found to be significantly higher than the reference frequency f_baseline calibrated under ice-free and dry conditions (i.e., frequency offset Δf = f_current - f_baseline > threshold), then it is determined that ice is wrapped around the rod, triggering a level two warning to confirm icing.

[0014] Preferably, after confirming icing, the methods for estimating ice thickness include: calibration-assisted mode and pure model fusion mode; The calibration assistance mode involves installing a capacitive ice thickness sensor in parallel at key locations during system deployment as a "ruler" to measure the change in vibration frequency. The characteristics of ambient temperature and wind speed are correlated with the actual thickness measured by the "ruler". A localized "frequency-environmental feature-thickness" estimation model is trained by machine learning. Based on the "frequency-environmental feature-thickness" estimation model, the thickness of the ice layer is estimated. Pure model fusion mode: In the absence of a direct "benchmark", the system will measure the change in vibration frequency. As a strong feature, along with the dew point temperature difference ΔT, the duration of low temperature, and the historical cooling rate, it is input into an improved energy balance model to estimate the ice thickness trend. The specific expression is as follows: ; Where: H n For the first The thickness of the ice layer at step H n+1 Let Δt be the ice thickness for the next step, ρ be the time step, and L be the density of the ice. f For the latent heat of fusion of ice, h0(W) s,n ) is the basic convective heat transfer coefficient, W s,n For the first The wind speed of the step, g(Δf) n ) is the vibration characteristic correction function, Δf n f(ΔT) represents the offset of the current frequency relative to the reference frequency. n ) is the dew point temperature difference correction function, ΔT n Φ is the difference between the road surface temperature and the dew point temperature. n T is a correction factor for the cumulative effect of historical low temperatures. s,n For the first The surface temperature of the step, T a,n For the first The air temperature at step, h0 is the basic convective heat transfer coefficient.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention is the first to use "structural vibration frequency change" as a direct contact criterion for road icing, and to conduct collaborative verification with the "dew point temperature difference model" to form a double-insurance judgment mechanism.

[0016] A lightweight, low-cost, and highly robust implementation scheme (metal rod + MEMS sensor) for a vibration sensing module is proposed, making it feasible for engineering.

[0017] A multi-source data fusion ice layer thickness estimation model with "vibration frequency change" as one of its core features was constructed, realizing a functional leap from qualitative judgment to quantitative estimation.

[0018] The entire system architecture embodies the idea of ​​"combining points and surfaces": using inexpensive networked temperature and humidity sensors to achieve risk scanning on the "surface", and using low-cost vibration sensing units to perform precise verification and feature enhancement on the "points", thus achieving the best overall cost-effectiveness of the system. Attached Figure Description

[0019] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram illustrating the overall system composition and deployment according to an embodiment of the present invention; Figure 2 This is a detailed structural diagram of the vibration sensing module according to an embodiment of the present invention; Figure 3 This is a flowchart illustrating the data processing and early warning process of the method in an embodiment of the present invention. Detailed Implementation

[0021] 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.

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0023] Example 1 like Figure 1 , Figure 2As shown, the present invention provides a road surface icing status monitoring system based on multi-source data fusion. Specifically, it relates to a road surface monitoring system for bridges, elevated roads and slopes and other road sections prone to icing, which is cost-effective, flexible in deployment, highly adaptable to the environment, and can simultaneously achieve highly reliable icing early warning and ice thickness estimation. The system includes: a vibration sensing module, a meteorological and road surface status sensing module, and a data fusion and processing module. The weather and road condition perception module is used to identify potential road icing risks; Vibration sensing module is used to detect the icing status of road surfaces with potential icing risk; The data fusion and processing module is used to estimate the ice thickness after icing is confirmed.

[0024] In this embodiment, the meteorological and road condition perception module serves as the foundation for the networked risk early warning system, including temperature and humidity sensors deployed on the roadside guardrail and temperature sensors buried under the road surface. Temperature and humidity sensors deployed on roadside guardrails are used to acquire air temperature and humidity. Temperature sensors embedded under the road surface are used to obtain road surface temperature.

[0025] In this embodiment, the "vibration sensing module" is the core, original component of this invention used for directly and physically detecting the state of road icing. It is fundamentally different from all existing technologies that rely on electromagnetic waves, optics, or purely environmental models.

[0026] The vibration sensing module is a miniaturized, mechatronic monitoring probe embedded within the road surface. Its core design concept is to transform the environmental problem of "the presence and impact of ice"—difficult to measure directly—into the mechanical problem of "whether structural boundary conditions and vibration characteristics change," which is easier to measure precisely. Its core components include: a metal rod-like structure fixed to the road surface, a vibration sensor (such as a piezoelectric ceramic plate or MEMS accelerometer) attached to it, a waterproof sealing layer, a signal transmission line, and a mounting base. Rod (core oscillator): A metal rod (such as stainless steel) protruding from the road surface, serving as the main body of vibration.

[0027] Vibration sensor: A piezoelectric ceramic sheet or MEMS accelerometer that is tightly attached to the rod to sense the microscopic vibration of the rod.

[0028] Waterproof sealing layer: Covers the sensor to ensure its long-term stability in humid and salty environments.

[0029] Signal transmission line: transmits the vibration electrical signal to the data acquisition unit.

[0030] Fixed base: The bottom of the pole is rigidly fixed to the roadbed by flanges and chemical anchors.

[0031] The vibration sensing module, as the core of the system's direct contact verification, has the core function of monitoring changes in the natural frequency of the members. The specific process is as follows: Step 1: Initial Calibration After system installation and in a stable, ice-free, dry state, multiple sets of data were continuously measured. Subsequent algorithms were used to determine the reference natural frequency f0 of the member and its probability distribution (mean and standard deviation). This f0 serves as the "healthy baseline" for all subsequent comparisons.

[0032] Step 2: Signal Preprocessing The original signal x(t) is processed to eliminate interference and linear drift. High-frequency noise and low-frequency interference are filtered out.

[0033] Step 3: Core Frequency Domain Analysis The time-domain signal is converted to the frequency domain using a Fast Fourier Transform (FFT), and the power spectral density (PSD) is calculated to identify frequencies where energy is concentrated.

[0034] Step 4: Peak detection and change calculation Within the expected frequency band of PSD, find the peak with the largest amplitude; the frequency corresponding to this peak is the currently estimated natural frequency f_current. Calculate the frequency offset Δf.

[0035] Step 5: Statistical significance test: Not all Δf indicates icing. It is crucial to distinguish between normal fluctuations caused by environmental noise and systematic shifts due to ice. This invention employs the Z-score test for determination. Z = (f_current - μ0) / σ0 Here, μ0 and σ0 are the mean and standard deviation of the reference frequency obtained during the calibration phase. If |Z|>3 (i.e., the deviation exceeds 3 times the standard deviation), according to statistical principles, this event is a low-probability event (confidence level > 99.7%), and can be determined as a significant change in the physical state of the rod. Combined with the risk warning from the meteorological module, it can be attributed to ice layer encasement.

[0036] In this embodiment, "structural vibration frequency change" is used for the first time as a direct contact criterion for road icing, and is jointly verified with the "dew point temperature difference model" to form a double-insurance judgment mechanism. The specific process is as follows: First layer of protection: Initial screening and early warning based on meteorological physical models The goal of this phase is to identify potential icing risks in a low-cost, wide-ranging manner, based on atmospheric physical laws.

[0037] 1. Core Formula: Dew point temperature (T_d) is the temperature at which air cools to water vapor saturation (relative humidity RH = 100%). It can be accurately calculated using the current air temperature T_a and relative humidity RH, and is commonly approximated using the Magnus formula: T_d = (b α(T_a, RH)) / (a ​​- α(T_a, RH)), Where, α(T_a, RH) = (a T_a) / (b + T_a) + ln(RH / 100), where α is a composite index of the current air's "logarithm of saturated vapor pressure" and "logarithm of actual water vapor content". (a T_a) / (b + T_a): Represents the logarithmic scale of the saturated vapor pressure at the current temperature T_a. It quantifies the current air's "water-holding capacity". ln(RH / 100): Represents the logarithm of relative humidity. It is a negative number (because RH < 100), indicating the "distance" between the current actual water vapor content and the saturation value at the current temperature. The essence of α is to map the current air's "thermal state" (T_a) and "humidity state" (RH) onto a single variable, which directly determines how much the air needs to be cooled to reach saturation. It is an intermediate variable that simplifies the complex exponential equation into a simple algebraic equation to solve for T_d. Simply put, α is the measure of the current air's "comprehensive distance" from saturation in a specific mathematical coordinate system. The closer the distance (i.e., RH close to 100%, or T_a close to T_d), the closer α is to (a + T_d). T_d) / (b + T_d).

[0038] a = 17.27, b = 237.7 (°C, applicable above 0°C). (Note: This is a simplified formula; the actual code will use more precise iterative calculations or table lookup methods.)

[0039] The surface-dew point temperature difference (ΔT) is a direct criterion for early warning: ΔT = T_surface - T_d, Where T_surface is the measured value of the road surface temperature sensor.

[0040] 2. Judgment Mechanism An L1 level warning (ice-freezing risk) is triggered when both of the following conditions are met simultaneously: Condition 1: ΔT < δ (e.g., δ = 2℃). Condition 2: T_surface < 1℃, The second layer of protection: physical verification based on structural dynamics The goal of this phase is to "verify the truth" of L1 warnings, based on the changes in the mechanical properties of solid structures caused by ice layers.

[0041] Core principle: Ice layer encapsulation alters boundary conditions The metal pole fixed to the road surface is simplified as a cantilever beam with its base fixed. Its first natural frequency (f) is determined by the beam's geometry, material properties, and boundary conditions. When ice wraps around the pole, it is equivalent to adding additional mass and elastic constraints locally, significantly altering the system's equivalent stiffness and mass distribution.

[0042] Key formula: Frequency variation characterizes the influence of ice layer Ice-free reference state: Under dry, ice-free conditions, the natural frequency of the rod is measured as the reference value f_0.

[0043] Icing state: When the height of the ice layer enveloping the rod is h_ice, the current natural frequency is measured as f_current.

[0044] Core variable: Define frequency offset Δf = f_current - f_0.

[0045] Simplified mechanical model: For a rod fixed at the bottom and partially constrained by ice at the top, its frequency variation has a complex relationship with the ice-bound height and stiffness. This can be simplified to a single function: Δf / f_0 = G( h_ice / L, E_ice, ... ), Where L is the rod length and E_ice is the elastic modulus of ice (affected by temperature and purity).

[0046] Ultimate Synergistic Effect: Effective Early Warning = L1 Warning && L2 Confirmation. This mechanism functions like a rigorous "scout-sniper" team: the scout (dew point model) uses binoculars to identify suspicious targets (risk areas), while the sniper (vibration sensor) uses a high-precision scope (physical contact) for observation and confirmation. Only when both are in agreement does the system fire (issue a high-level warning). This ensures the system has both high sensitivity and high reliability, solving the industry challenge of a single sensor or model being unable to achieve both.

[0047] In this embodiment, the data fusion and processing module is typically a gateway with edge computing capabilities. It is responsible for running the dew point temperature difference model, vibration signal analysis algorithm, and multi-source data fusion estimation model, and executing the tiered early warning logic. The specific process is as follows: 1. Data preprocessing and synchronization Input: Raw data stream (temperature, humidity, road surface temperature, vibration waveform, wind speed, etc.).

[0048] process: Cleaning: Remove obviously physically unreliable outliers (such as humidity > 100%).

[0049] Alignment: All data is stamped with a uniform, high-precision timestamp to resolve timing misalignment issues caused by transmission delays.

[0050] Feature extraction: Real-time calculation of key features, such as ΔT (dew point temperature difference), Δf (change in vibration frequency and its Z-score), and cooling rate over the past hour.

[0051] Output: A time-aligned, clean array of feature vectors for subsequent engine use.

[0052] 2. Core Judgment and Fusion Engine This engine is the "brain" of the module, consisting of three layers of logic: a. Rule-based reasoning layer (executes double-insurance logic) This layer uses hard-coded logic to implement fast and reliable primary decisions, which is the system's "conditioned reflex".

[0053] b. Model fusion layer (performs thickness estimation) This layer is activated when an L2 warning is triggered. Its core is a regression model used to estimate ice thickness (H_ice).

[0054] Input feature vector (X): [Δf, ΔT, road surface temperature, air temperature, wind speed, duration of low temperature,...] Estimation model (Y): H_ice = F(X) Two implementation paths for model F: Physical model driven: Based on a simplified energy balance equation, Δf is used as a feedback parameter of the ice layer binding strength to dynamically correct the heat loss calculation.

[0055] Data-driven (machine learning): Algorithms such as Gradient Boosting Regression Tree (GBRT) are used. In the initial deployment phase, a basic model can be trained using limited historical or simulation data. The introduction of vibration characteristics Δf is crucial to this model, providing features directly related to the physical contact of the ice layer that are unavailable in pure meteorological models, significantly improving estimation accuracy.

[0056] Output: Estimated thickness value and its confidence interval (e.g., 1.5 ± 0.3 mm).

[0057] c. Confidence Management The credibility of the final output is comprehensively evaluated.

[0058] If the vibration signal is strong (Z_score is large) and the weather conditions remain severe, the confidence level of the fusion estimation results will increase.

[0059] If a weather warning is strong but the vibration signal is weak or contradictory, a warning will be issued but marked "low confidence level" to prompt manual verification.

[0060] 3. Final Decision and Output The results of combining rule-based reasoning and model fusion are used to generate and issue the final structured instructions.

[0061] Example 2 like Figure 3 As shown, the present invention also provides a method for monitoring road surface icing status based on multi-source data fusion. This method is implemented using the system described in Embodiment 1, and includes: Identify potential road surface icing risks; Detect the icing status of road surfaces with potential icing risk; After confirming icing, the ice thickness is estimated.

[0062] In this embodiment, the method for identifying potential road surface icing risks includes: The difference ΔT between the road surface temperature T_s and the dew point temperature T_d estimated from the air temperature and humidity is calculated in real time. When ΔT < preset threshold and T_s < 1℃, a Level 1 warning is triggered, indicating the risk of icing.

[0063] In this embodiment, the method for calculating the difference ΔT between the road surface temperature T_s and the dew point temperature T_d derived from the air temperature and humidity in real time includes: Step 1: Data Acquisition and Synchronization enter: Air temperature and humidity: The current air temperature (T_a, in °C) and relative humidity (RH, in %) are read from sensors deployed on the guardrail.

[0064] Road surface temperature: The current road surface temperature (T_s, in °C) is read from a sensor embedded in the road surface.

[0065] Synchronization: Assign the same millisecond-level timestamp to the three data points T_a, RH, and T_s to ensure that they represent the physical state at the same moment.

[0066] Step 2: Data validity verification and cleaning (ensuring robustness) Before calculation, the raw data is "checked" to prevent calculation errors or false alarms caused by invalid data.

[0067] Step 3: Core Calculation - Dew Point Temperature (T_d) The modified August-Roche-Magnus formula recommended by the World Meteorological Organization is adopted: First, calculate the saturated vapor pressure (es, unit: hPa). es = 6.112 exp((17.67 T_a) / (T_a + 243.5)), Then, calculate the actual water vapor pressure (e, unit: hPa). e = es (RH / 100.0), Finally, calculate the dew point temperature (T_d, unit: °C). T_d = (243.5 ln(e / 6.112)) / (17.67 - ln(e / 6.112)), Where: T_a is the current air temperature (°C), RH is the current relative humidity (%), exp() and ln() are the natural exponent and natural logarithm functions respectively, and the constants 17.67 and 243.5 are empirical coefficients for water.

[0068] Step 4: Calculate the key feature - road surface dew point temperature difference (ΔT) ΔT = T_s - T_d, ΔT>0: The road surface temperature is higher than the dew point, and water vapor in the air will not condense on the road surface.

[0069] ΔT ≈ 0: The road surface temperature is close to the dew point, and the road surface may be damp.

[0070] ΔT<0: The road surface temperature is below the dew point, and water vapor in the air will condense on the road surface. If T_s is also below the freezing point at this time, the condensate will freeze.

[0071] Step 5: Output and Transmission The calculated ΔT value, along with the original values ​​such as T_s and T_d, is packaged into a structured data packet and transmitted to the subsequent hierarchical early warning logic unit.

[0072] In this embodiment, the method for detecting the icing state of road surfaces with potential icing risk includes: Upon triggering a Level 1 warning, the system simultaneously activates high-frequency sampling of the vibration sensing module. The vibration spectrum of the metal rod is analyzed, and its fundamental frequency (f) is extracted. If a statistically significant increase is found in the current fundamental frequency (f_current) compared to the reference frequency (f_baseline) calibrated under ice-free and dry conditions (i.e., Δf = f_current - f_baseline > threshold), it is determined that ice is encasing the rod, triggering a Level 2 warning (ice confirmation). This step is a key innovation; it utilizes the physical fact that ice alters the boundary conditions of the rod (increasing constraints and shortening the effective vibration length) to provide irrefutable direct evidence for model-based warnings, significantly reducing the false alarm rate.

[0073] In this embodiment, after confirming icing, the system initiates the thickness estimation process. Two optional fusion paths are provided: Path A (Calibration-Assisted Mode): During system deployment, a commercially available capacitive ice thickness sensor can be installed in parallel at a key point (the coldest, wettest, most ventilated, and least disturbed point on the bridge, typically at the base of the guardrail on the leeward side (or downwind of the prevailing winter wind) of the cross section) as a "benchmark." The data processing module correlates the vibration frequency change (Δf), ambient temperature, wind speed, and other characteristics with the actual thickness measured by the "benchmark," and trains a localized "frequency-environmental feature-thickness" estimation model using machine learning (such as random forest or gradient boosting tree). After long-term operation, the reliance on the "benchmark" can be reduced. The specific process is as follows: I. Feature Engineering and Data Preparation The model takes a multidimensional feature vector X as input and outputs the target variable y (ice thickness). The feature vector X consists of the following two parts: Vibration characteristics: Δf: The offset of the current frequency relative to the reference frequency, in Hz. This feature directly reflects the grip strength of the ice layer on the rod.

[0074] Z Δf Δf is the statistically standardized value, or Z-score, used to characterize the significance of the change.

[0075] Environmental characteristics: T_s: Road surface temperature (°C).

[0076] T_a: Air temperature (°C).

[0077] RH: Relative humidity (%).

[0078] W_s: Wind speed (m / s).

[0079] ΔT: Temperature difference between road surface and dew point (°C).

[0080] t low : The duration (in minutes) during which the road surface temperature is below 1°C.

[0081] All features need to be normalized to eliminate the influence of dimensions. For example, Z-score standardization can be used: Where μ i ,σ i For the first The mean and standard deviation of each feature on the training set.

[0082] II. Model Training: Supervised Learning Using capacitive ice thickness sensors installed at key locations as a "ruler," the actual ice thickness value is collected. The feature vectors at the same moment (or within a time window) and Composition of training samples The training set should cover as wide a range of icing events as possible (e.g., combinations of different temperatures, wind speeds, and ice thicknesses). A machine learning regression algorithm is used to build the model. This invention preferably uses Random Forest or Gradient Boosting Decision Tree (GBDT) because: they can automatically handle nonlinear relationships and interactions between features; they have good robustness to outliers and noisy data; they can output feature importance rankings, facilitating model interpretation and feature selection; and they are suitable for small to medium-sized datasets and are less prone to overfitting.

[0083] Taking random forest as an example, the model consists of Tree of return Each tree is constructed based on Bootstrap samples from the training set and a random subset of features. The final prediction is the mean of all tree predictions. The training objective is to minimize the mean squared error (MSE): in This represents the number of training samples.

[0084] III. Model Validation and Update After training, the model accuracy needs to be evaluated on a separate test set, with metrics including: Mean Absolute Error (MAE): ,in, This represents the number of test sets.

[0085] Root Mean Square Error (RMSE): .

[0086] Coefficient of determination : Measures the extent to which the model explains the target variable.

[0087] If the model meets the accuracy requirements (e.g., MAE < 0.5 mm), it is deployed to an edge computing gateway for real-time estimation. The system has self-optimization capabilities: as runtime accumulates and new benchmark data is continuously added, the model can be retrained periodically (e.g., weekly), and the updated parameters are distributed from the cloud to the edge, making the model increasingly adaptable to local microclimates and structural characteristics.

[0088] Path B (Pure Model Fusion Mode): In the absence of a direct "scale," the system uses the change in vibration frequency (Δf) as a strong feature, along with features such as dew point temperature difference (ΔT), duration of low temperature, and historical cooling rate, and inputs them into an improved energy balance model. This model treats ice layer growth as a thermodynamic process, and Δf provides quantitative feedback on the strength of ice-structure interaction, thereby dynamically correcting model parameters and achieving a more accurate thickness trend estimation than a pure temperature and humidity model. The specific process is as follows: I. Basic Energy Balance Equation The growth of ice on roads is primarily controlled by heat exchange between the road surface and the air. According to the law of conservation of energy, the latent heat of phase change released by the thickening of the ice layer should equal the heat flux lost from the ice surface. in: The thickness of the ice layer is (m). The density of ice (usually taken as 917 kg / m³) 3 ); The latent heat of fusion of ice (3.34 × 10⁻⁶) 5 J / kg), It is the heat flux per unit area (W / m²), which is the rate at which heat is transferred from the surface of the ice layer to the air.

[0089] II. Parameterization of Heat Flux and Basic Heat Transfer Coefficient Heat flux mainly consists of convective heat transfer and radiative heat transfer, and in practical applications it is often simplified to a linear form that is proportional to the temperature difference: in: The surface temperature (°C) of the road surface (ice layer). Air temperature (°C); The comprehensive heat transfer coefficient (W / (m²·℃)) reflects the combined effects of convection, radiation, and other factors.

[0090] Mainly affected by wind speed The influence, according to the classic empirical formula, is: in For empirical constants (e.g.) (This can be calibrated through experiments or historical data). The basic convective heat transfer coefficient represents the amount of heat carried away by airflow per unit time, per unit temperature difference, and per unit area. It is related to wind speed. The function.

[0091] III. Based on vibration characteristics Dynamic correction Change in vibration frequency This directly reflects the grip strength and adhesion of the ice layer to the pole. When the ice layer is dense and tightly bonded to the pole,... As the surface roughness of the ice layer increases, its effective thermal conductivity and surface roughness may also change, thus affecting heat transfer efficiency. This invention introduces a correction function. The basic heat transfer coefficient is dynamically adjusted. Correction function It must satisfy monotonically increasing ( The larger the value, the stronger the correction. (No correction when there is no ice). Optional formats include: Linear: index: Piecewise functions: Based on experimental calibration, determine different... The correction factor corresponding to the interval.

[0092] in These are parameters to be determined and can be optimized using a small number of historical icing events (no scale is needed; manual measurement or empirical values ​​can be used).

[0093] IV. Refined Modification Integrating Other Environmental Features To further improve model accuracy, dew point temperature difference was introduced. and duration of low temperature As an auxiliary correction factor.

[0094] 1. Dew point temperature difference correction when Near freezing temperatures, water vapor in the air easily condenses on the ice surface, releasing latent heat and slowing down ice growth. Define the correction function: in For empirical parameters, make hour Reduced (heat exchange efficiency decreased).

[0095] 2. Low temperature duration correction Ice growth has a cumulative effect; the duration of historical low temperatures influences the internal structure and density of the ice layer. Introducing an integral term: in, This is a dimensionless correction factor for the cumulative effect of historical low temperatures, used to amplify or reduce the heat transfer coefficient at the current moment to reflect the cumulative impact of past low temperatures. The reference temperature (unit: °C) is usually taken as the freezing point temperature (0 °C), but it can be finely adjusted according to the physical properties of the ice layer (e.g., -1 °C to account for the freezing point decrease). Road surface temperature (unit: °C), time The function. Historical low temperature influence coefficient (unit: ℃) -1 ·s -1 or ℃ -1 ·min -1 ), which characterizes the contribution of low-temperature duration to ice growth. The integration start time is usually taken as the start time of the current icing process (such as the L2 warning trigger time). This refers to the current moment.

[0096] In practical applications, it can be discretized as follows: in, For the first The road surface temperature, in °C, is measured in real time by an embedded road surface temperature sensor. For time step, In summary, the improved expression for heat flux is: V. Ice Thickness Recursive Formula and Model Discretization The formula Substitution The rate of change of ice thickness was obtained: Discretization is performed using the forward Euler method, with a time step of 1. (e.g., 60 seconds), the Thick ice step Updated to: In the formula: For the first The ice thickness at this step, in mm or m, represents the estimated ice thickness at the current moment, initially... . The next step is to measure the ice thickness, in mm or m, indicating the updated ice thickness. The time step, in seconds or minutes, represents the time interval for discretization calculations. It is typically set to 60 seconds (1 minute) to balance computational accuracy with edge computing load. This is the density of ice, expressed in kg / m³. The density of pure ice is approximately 917 kg / m³, while road ice containing air bubbles or impurities may have a slightly lower density. The latent heat of fusion of ice is expressed in J / kg, representing the amount of heat required for ice to melt into water. The standard value is 3.34 × 10⁵. (Ws,n) is the basic convective heat transfer coefficient, with units of W / (m²·℃), which represents the heat exchange coefficient between the road surface and the air when there is no ice cover. For the first The wind speed at each step, in m / s, is measured in real time by the bridge meteorological station. This is a dimensionless correction function for vibration characteristics, based on the change in vibration frequency. The dynamic correction reflects the influence of the bond strength between the ice layer and the rod on heat transfer. This represents the offset of the current frequency relative to the reference frequency, in Hz, extracted in real time by the vibration sensing module. A positive value indicates that icing causes the natural frequency of the rod to increase. The dew point temperature difference correction function is dimensionless and considers the effect of air humidity on the heat transfer of the ice surface. When the road surface is close to the dew point, water vapor condenses and releases latent heat, which slows down the growth of the ice layer. The difference between the road surface temperature and the dew point temperature, in °C. It is calculated in real time by the meteorological module. This is a dimensionless correction factor for the cumulative effect of historical low temperatures, reflecting the cumulative impact of the duration of low temperatures on the internal structure and growth rate of the ice layer since the beginning of freezing. For the first The road surface temperature, in °C, is measured in real time by an embedded road surface temperature sensor. For the first The air temperature at the step, in °C, is measured in real time by the temperature and humidity sensor at the weather station. The temperature difference between the road surface and the air, measured in °C, is the core driving force for heat exchange; the greater the temperature difference, the stronger the heat exchange.

[0097] Initial conditions .when or When the ice layer is below the threshold, it may stop growing or begin to melt. In this case, the model needs to introduce a melting term (similar principle but with opposite signs).

[0098] VI. Parameter Calibration and Model Initialization Empirical parameters in the model ( (etc.) can be determined by fitting historical meteorological data of the bridge location and a small number of manually observed icing events (without the need for a high-precision scale).

[0099] In summary, the present invention achieves the following beneficial effects: High reliability: Through the dual mechanism of "model early warning + physical verification", the problem of high false alarm rate of pure model prediction is fundamentally solved, which greatly improves the confidence of icing judgment.

[0100] Economic efficiency and accessibility: The core verification unit (metal rod + vibration sensor) has extremely low cost and a robust and durable structure. The overall system hardware cost is far lower than that of high-precision remote sensing or all-road-buried sensor solutions, making it suitable for large-scale deployment in road networks.

[0101] Environmental robustness: Vibration sensing is insensitive to optical interference such as rain, snow, fog, and dust; the metal rod structure is simple, corrosion-resistant, and crush-resistant, making it suitable for harsh road environments.

[0102] Advanced Functionality: It not only determines whether ice is present or not, but also estimates the thickness of the ice layer through the fusion of multi-source data, providing a more valuable basis for maintenance decisions.

[0103] Flexible deployment: The vibration sensing unit can be used independently to upgrade old systems with existing weather stations, or it can be used as a core component of new systems. It is easy to install and does not require large-scale road excavation.

[0104] Example 3 The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0105] 1. Implementation of the vibration sensing module: Pole body: A solid round pole made of 304 stainless steel with a diameter of 20mm and a height of 150mm above the road surface is selected. A flange is welded to the bottom of the pole body, and it is fixed by drilling holes in the shoulder or guardrail foundation and casting high-strength chemical anchors to ensure that the bottom is an ideal fixed end.

[0106] Vibration sensing: A low-power, wide-temperature MEMS digital accelerometer (such as ADI's ADXL345) is selected. It is encapsulated in a waterproof housing and bonded to the root of the exposed portion of the rod (the point of maximum stress and strain) with high-strength epoxy resin. Vibration data is transmitted to the edge gateway via an I2C bus.

[0107] Excitation and Acquisition: The edge gateway can control a small electromagnetic exciter (or use ambient wind or vehicle traffic to excite) to transiently excite the pole, and acquire its free decaying vibration waveform through an accelerometer, and extract the fundamental frequency through FFT transformation.

[0108] 2. Implementation of data processing algorithms: Frequency change detection: During dry, ice-free periods, the system learns and stores a baseline frequency f_baseline and its normal fluctuation range. During operation, it calculates the standard score (Z-Score) of the current frequency f_current compared to f_baseline. A significant change is defined as a Z-Score that consistently exceeds 3 (i.e., 99.7% confidence level).

[0109] Example of a fusion estimation model (using a machine learning approach): Feature vector: [Δf (frequency change), T_s (surface temperature), T_air (air temperature), RH (humidity), wind_speed (wind speed), ΔT (dew point temperature difference), duration (duration of low temperature)].

[0110] Training objective: The true ice thickness measured by a capacitive sensor.

[0111] Model: Trained using GradientBoostingRegressor from the Scikit-learn library. When running online, the model takes real-time feature vectors as input and outputs estimated ice thickness and confidence intervals.

[0112] 3. System Deployment: A complete system is deployed at typical cross-sections of the target bridge (such as the highest point or a shaded area). Vibration sensing units and capacitive ice thickness sensors (if applicable) are installed side-by-side, approximately 0.5 meters apart. Temperature and humidity sensors are installed on the guardrail upstream in the crosswind direction. All sensors are connected to an edge computing gateway located in the bridgehead electrical distribution box via RS-485 bus or LoRa wireless connection.

[0113] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A road surface icing status monitoring system based on multi-source data fusion, characterized in that, The system includes: a vibration sensing module, a meteorological and road surface condition sensing module, and a data fusion and processing module; The weather and road condition sensing module is used to identify potential road icing risks; The vibration sensing module is used to detect the icing status of road surfaces with potential icing risks. The data fusion and processing module is used to estimate the ice thickness after icing is confirmed.

2. The system according to claim 1, characterized in that, The meteorological and road condition sensing module includes a temperature and humidity sensor deployed on the roadside guardrail and a temperature sensor buried under the road surface. The temperature and humidity sensor deployed on the roadside guardrail is used to acquire air temperature and humidity. The temperature sensor embedded in the road surface is used to obtain the road surface temperature.

3. The system according to claim 1, characterized in that, The vibration sensing module includes a metal rod-shaped structure fixed to the road surface and a vibration sensor, a waterproof sealing layer, a signal transmission line, and a fixing base attached thereto. The vibration sensor is used to sense the micro-vibrations of the rod. The waterproof sealing layer is used to cover the sensor; The signal transmission line is used to transmit the vibration electrical signal to the data acquisition unit; The fixed base is used to rigidly fix the bottom of the pole to the roadbed through flanges and chemical anchors.

4. A method for monitoring road surface icing status based on multi-source data fusion, wherein the method is implemented using the system described in any one of claims 1-3, characterized in that, The method includes: Identify potential road surface icing risks; Detect the icing status of road surfaces with potential icing risk; After confirming icing, the ice thickness is estimated.

5. The method according to claim 4, characterized in that, Methods for identifying potential road surface icing risks include: Real-time calculation of the difference between the road surface temperature T_s and the dew point temperature T_d estimated from the air temperature and humidity. ; When satisfied When the temperature is below the preset threshold and T_s < 1℃, a Level 1 warning is triggered, indicating a risk of icing.

6. The method according to claim 5, characterized in that, Methods for calculating the dew point temperature T_d from air temperature and humidity include: T_d=(243.5 ln(e / 6.112)) / (17.67-ln(e / 6.112)) Where e is the actual water vapor pressure; Real-time calculation of the difference between the road surface temperature T_s and the dew point temperature T_d estimated from the air temperature and humidity. The methods include: ΔT = T_s - T_d.

7. The method according to claim 6, characterized in that, Methods for detecting the icing status of road surfaces at potential icing risk include: After the first-level warning is triggered, the vibration spectrum of the metal rod is analyzed and its fundamental frequency is extracted; If a statistically significant increase is found between the current fundamental frequency f_current and the reference frequency f_baseline calibrated under ice-free and dry conditions, i.e., a frequency offset, then... If f_current - f_baseline > threshold, it is determined that ice is wrapped around the rod, triggering a level 2 warning to confirm icing.

8. The method according to claim 7, characterized in that, After confirming icing, methods for estimating ice thickness include: calibration-assisted mode and pure model fusion mode; The calibration assistance mode involves installing a capacitive ice thickness sensor in parallel at key locations during system deployment as a "ruler" to measure the change in vibration frequency. The characteristics of ambient temperature and wind speed are correlated with the actual thickness measured by the "ruler". A localized "frequency-environmental feature-thickness" estimation model is trained by machine learning. Based on the "frequency-environmental feature-thickness" estimation model, the thickness of the ice layer is estimated. Pure model fusion mode: In the absence of a direct "scale", the system will measure the change in vibration frequency. As a strong characteristic, it is related to the dew point temperature difference. The duration of low temperatures and historical cooling rates are input together into an improved energy balance model to estimate the trend of ice thickness. The specific expression is as follows: ； Where: H n For the first The thickness of the ice layer at step H n+1 Let Δt be the ice thickness for the next step, ρ be the time step, and L be the density of the ice. f For the latent heat of fusion of ice, h0 (W) s,n ) is the basic convective heat transfer coefficient, W s,n For the first The wind speed of the step, g(Δf) n ) is the vibration characteristic correction function, Δf n f(ΔT) represents the offset of the current frequency relative to the reference frequency. n ) is the dew point temperature difference correction function, ΔT n Φ is the difference between the road surface temperature and the dew point temperature. n T is a correction factor for the cumulative effect of historical low temperatures. s,n For the first The surface temperature of the step, T a,n For the first The air temperature at step, h0 is the basic convective heat transfer coefficient.

Images