Intelligent grading early warning method and system for road snow thickness

Abstract

The application discloses a road snow thickness intelligent grading early warning method and system, and relates to the technical field of road safety warning. The method comprises the following steps: measuring and screening the snow thickness of a target road at multiple points to determine a real-time representative value; perceiving the environment by using a multi-modal environment sensor array, acquiring an environment parameter sequence and performing time sequence iteration, and extracting trend characteristics; predicting the snow thickness based on the trend characteristics to obtain a prediction result; matching the prediction result and the trend characteristics to grading early warning information as analysis data, and combining a warning reference point information library to generate target grading early warning information. The technical effect of improving monitoring accuracy, providing a dynamic response and intelligent early warning mechanism based on prediction, and improving early warning accuracy is achieved.

Description

Technical Field

[0001] This invention relates to the field of road safety early warning technology, and in particular to a method and system for intelligent graded early warning of road snow thickness. Background Technology

[0002] In winter road maintenance scenarios, monitoring and early warning of snow thickness is crucial for ensuring traffic safety. Snow thickness significantly affects the traction of passing vehicles, deteriorates braking distance, and thus affects driving safety.

[0003] Existing snow depth warning technologies have the following drawbacks: they are based on single-point measurement data and cannot accurately reflect the snow depth distribution along the entire road; they do not consider the dynamic factors in the snow formation process, such as temperature changes and snowfall rate, on the snow depth; and the warning mechanism is relatively simple, only making threshold judgments based on the current snow depth and lacking the ability to predict future snow depth trends. Summary of the Invention

[0004] This invention provides an intelligent graded early warning method and system for road snow thickness, which solves the technical problems of single data sources, lack of dynamic response capability, and crude early warning mechanism in the prior art. It achieves the technical effect of improving monitoring accuracy, providing prediction-based dynamic response and intelligent early warning mechanism, and improving the accuracy of early warning.

[0005] In a first aspect, the present invention provides an intelligent classification and early warning method for road snow thickness, wherein the intelligent classification and early warning method for road snow thickness includes:

[0006] The snow thickness was measured at multiple points on the target road, and the measurement results were filtered to determine the representative value of the real-time road snow thickness.

[0007] The target road is continuously perceived using a multimodal environmental sensor array to obtain a multimodal environmental parameter array sequence. The multimodal environmental parameter array sequence is then iterated over time to determine the multimodal environmental trend characteristics.

[0008] Based on the multimodal environmental trend characteristics, the time-series trend prediction of the representative value of real-time road snow thickness is performed to determine the prediction result of road snow thickness.

[0009] The predicted road snow thickness and multimodal environmental trend characteristics are used as analysis data to match graded early warning information. Target information reference points are retrieved from the early warning reference point information database. The matched graded early warning information and target information reference points are then fused to determine the target graded early warning information.

[0010] In one feasible implementation, a multimodal environmental sensor array is used to continuously perceive the target road environment, obtain a multimodal environmental parameter array sequence, and perform time-series iteration on the multimodal environmental parameter array sequence to determine multimodal environmental trend characteristics, including:

[0011] By sequentially interacting with adjacent multimodal environmental parameter arrays in the multimodal environmental parameter array sequence, multiple iterative multimodal environmental parameter correlation matrices are obtained.

[0012] The correlation matrices of the multiple iterative multimodal environmental parameters are used as implicit information, and the last multimodal environmental parameter array in the multimodal environmental parameter array sequence is used as explicit information. Convolutional analysis is then performed to determine the trend characteristics of the multimodal environment.

[0013] In one feasible implementation, adjacent multimodal environment parameter arrays in a sequence are sequentially interacted to obtain multiple iterative multimodal environment parameter correlation matrices, including:

[0014] Extract any two adjacent multimodal environmental parameter arrays from the multimodal environmental parameter array sequence, and use them as the first multimodal environmental parameter array and the second multimodal environmental parameter array.

[0015] Environmental parameter mapping similarity recognition is performed on the first multimodal environmental parameter array and the second multimodal environmental parameter array, and the recognition results are normalized and filled into the initially empty matrix to obtain the first iterative multimodal environmental parameter correlation matrix.

[0016] Similarly, by interacting with adjacent multimodal environmental parameter arrays in the multimodal environmental parameter array sequence, multiple iterative multimodal environmental parameter correlation matrices are obtained.

[0017] In one feasible implementation, the real-time road snow thickness representative value is predicted based on the multimodal environmental trend characteristics to determine the road snow thickness prediction result, including:

[0018] A pre-built time-series trend predictor is used to learn a two-to-one mapping relationship between the real-time representative value of road snow thickness and multimodal environmental trend features and the road snow thickness prediction results.

[0019] The time-series trend predictor is used to identify the representative value of real-time road snow thickness and multimodal environmental trend characteristics to determine the predicted result of road snow thickness.

[0020] In one feasible implementation, the predicted road snow thickness and multimodal environmental trend characteristics are used as analytical data to perform hierarchical early warning information matching. Target information reference points are retrieved from the early warning reference point information database. The matched hierarchical early warning information and the target information reference points are then fused to determine the target hierarchical early warning information, including:

[0021] Acquire multiple sample analysis data, which include multiple information reference points.

[0022] The multiple sample analysis data are aggregated in the same category to obtain M aggregated sample analysis data sets, where M is a positive integer.

[0023] Based on the mapping relationship between sample analysis data and information reference points, the multiple information reference points are aggregated according to the M aggregated sample analysis data sets, and the mode of the aggregation results is calculated to obtain M aggregated information reference points.

[0024] The early warning reference point information database is constructed based on the M aggregated sample analysis data sets and the M aggregated information reference points.

[0025] One feasible implementation includes:

[0026] The credibility score of the multimodal environment trend characteristics is calculated, and the credibility interval is determined based on the credibility score results.

[0027] Using the aforementioned confidence interval as the warning period, periodic warnings are issued for the snow thickness of the target road.

[0028] In one feasible implementation, snow thickness is measured at multiple points on the target road, and the measurement results are filtered to determine a representative value of real-time road snow thickness, including:

[0029] The snow thickness is measured at multiple points on the target road to obtain the measurement results, and a snow thickness measurement space is constructed based on the measurement results.

[0030] Extract the mode of snow thickness from the measurement results and use it as the initial spatial center of the snow thickness measurement space.

[0031] The snow thickness measurement space is iteratively analyzed using a filtering iteration function until the preset constraints are met, the target space center is obtained, and the snow thickness corresponding to the target space center is used as the representative value of the real-time road snow thickness.

[0032] One feasible implementation includes:

[0033] Obtain the filtering iteration function, wherein the filtering iteration function is:

[0034] .

[0035] in, For the center of the iterative space, To measure the thickness of snow in space, To measure the thickness of the i-th snow layer in space, As the initial spatial center, This is the Gaussian kernel function.

[0036] The initial spatial center is iterated using the filtering iteration function to obtain the iterative spatial center. It is then determined whether the iterative spatial center satisfies the preset constraint conditions. If so, the iterative spatial center is taken as the target spatial center.

[0037] In one feasible implementation, the preset constraint is that the number of iterations satisfies a preset maximum number of iterations and / or the difference between the nearest neighbor quantities of the iteration space centers of two adjacent iterations within a preset scale is less than or equal to a preset difference.

[0038] Secondly, the present invention also provides an intelligent graded early warning system for road snow thickness, wherein the intelligent graded early warning system for road snow thickness includes:

[0039] The snow thickness measurement module is used to measure the snow thickness at multiple points on the target road, and to filter the measurement results to determine the representative value of the real-time road snow thickness.

[0040] The continuous sensing and trend acquisition module is used to continuously sense the target road using a multimodal environmental sensor array, obtain a multimodal environmental parameter array sequence, and perform time-series iteration on the multimodal environmental parameter array sequence to determine the multimodal environmental trend characteristics.

[0041] The thickness prediction module is used to perform time-series trend prediction on the representative value of real-time road snow thickness based on the multimodal environmental trend characteristics, and to determine the road snow thickness prediction result.

[0042] The graded early warning module is used to match graded early warning information with the predicted road snow thickness and multimodal environmental trend characteristics as analysis data, and to retrieve target information reference points in the early warning reference point information database. The matched graded early warning information and target information reference points are then fused to determine the target graded early warning information.

[0043] This invention discloses an intelligent graded early warning method and system for road snow thickness, comprising: measuring snow thickness at multiple points on a target road and filtering the measurement results to determine a representative value of real-time road snow thickness; then, continuously sensing the environment of the target road using a multimodal environmental sensor array to obtain a multimodal environmental parameter array sequence, and performing time-series iteration on the sequence to determine the trend characteristics of the multimodal environment; then, performing time-series trend prediction on the representative value of real-time road snow thickness based on these trend characteristics to obtain the predicted result of road snow thickness; finally, matching graded early warning information with the predicted road snow thickness and the multimodal environmental trend characteristics as analysis data, retrieving relevant target information reference points in the early warning reference point information database, fusing the matched graded early warning information with the target information reference points, and finally determining the target graded early warning information. The intelligent graded early warning method and system for road snow thickness disclosed in this invention solves the technical problems of single data sources, lack of dynamic response capabilities, and crude early warning mechanisms, achieving the technical effects of improving monitoring accuracy, providing prediction-based dynamic response and intelligent early warning mechanisms, and improving early warning accuracy. Attached Figure Description

[0044] Figure 1 This is a flowchart illustrating the intelligent classification and early warning method for road snow thickness according to the present invention.

[0045] Figure 2 This is a schematic diagram of the intelligent graded early warning system for road snow thickness of the present invention.

[0046] Figure labeling: Snow thickness measurement module 11, continuous sensing and trend acquisition module 12, thickness prediction module 13, graded early warning module 14. Detailed Implementation

[0047] The above technical solutions will now be described in detail with reference to the accompanying drawings and specific embodiments to provide a better understanding of them. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments used only to explain the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. Furthermore, it should be noted that, for ease of description, only the parts related to the present invention are shown in the drawings, not all of them.

[0048] Example 1, as Figure 1 This is a flowchart illustrating the intelligent classification and early warning method for road snow thickness according to the present invention, wherein the intelligent classification and early warning method for road snow thickness includes:

[0049] S100: Perform multi-point snow thickness measurement on the target road, filter the measurement results, and determine the representative value of real-time road snow thickness.

[0050] Specifically, multiple measurement points are first set up at different locations along the target road, and snow thickness data is acquired at each point using appropriate measuring equipment (such as ultrasonic sensors, laser rangefinders, or piezoelectric sensors). The locations of these measurement points can be strategically arranged based on factors such as road length, terrain features, and traffic flow distribution to ensure that the collected data fully reflects the snow thickness distribution along the entire road. For example, on a 10-kilometer-long highway, a measurement point is set up every kilometer, with the density of measurement points appropriately increased in sections with high traffic volume.

[0051] Specifically, after the data acquisition is completed, the original multi-point snow thickness data is further filtered and processed to remove outliers or invalid data (such as obvious erroneous readings caused by environmental interference, data that are outside the sensor's measurement range, or data that differ too much from other adjacent measurement points). Then, the valid data are summed using appropriate algorithms (such as the average method, median method, or weighted average method) to determine the real-time representative value of the road snow thickness that can represent the current road snow thickness.

[0052] Through the above process, the limitations of single-point measurement data can be effectively overcome, providing more comprehensive and accurate basic data for subsequent snow thickness prediction and early warning, ensuring that the early warning system can conduct analysis and decision-making based on reliable real-time snow thickness information.

[0053] In some embodiments, multi-point snow thickness measurements are performed on the target road, and the measurement results are filtered to determine a representative value of real-time road snow thickness, including:

[0054] The snow thickness is measured at multiple points on the target road to obtain measurement results, and a snow thickness measurement space is constructed based on the measurement results. The mode of snow thickness in the measurement results is extracted and used as the initial spatial center of the snow thickness measurement space. The snow thickness measurement space is iteratively analyzed using a filtering iteration function until the preset constraints are met to obtain the target spatial center. The snow thickness corresponding to the target spatial center is used as the representative value of the real-time road snow thickness.

[0055] Specifically, the measurement results are used to construct a snow thickness measurement space. This space is built based on a virtual three-dimensional coordinate system and includes multiple data points. Each data point contains two key pieces of information: the measurement location (represented by the horizontal and vertical coordinates of the road) and the measured snow thickness at that location. In this way, discrete measurement data can be integrated into a model with spatial relationships, providing a foundation for subsequent spatial analysis. For example, the constructed snow thickness measurement space is denoted as:

[0056] ;

[0057] Where x i h represents the coordinates of the measurement location. i This indicates the snow thickness at that location.

[0058] Specifically, the mode h of the snow thickness is extracted from the measurement results. mode The mode, or the snow thickness value that appears most frequently in the measurement results, reflects the typical snow thickness in most areas of the road. Therefore, the mode h can be considered the most frequently observed snow thickness value. mode It serves as the initial cluster center or initial reference region, and as the starting point for iterative analysis.

[0059] Furthermore, using this initial center as a starting point, an iterative analysis is performed in the snow thickness measurement space using a filtering iteration function: First, in each iteration, points that satisfy a certain spatial distance and thickness similarity are selected based on the distribution of points near the current center value; for example, data points whose thickness difference from the initial center exceeds 20% can be removed first. Then, the average thickness and position center of the remaining data points are recalculated, and this new center is used as the starting point for the next iteration. This process is repeated until preset constraints are met, such as 10 iterations (preset number constraint) or the change in the center position is less than 1 cm (result convergence constraint). The snow thickness corresponding to the target spatial center is then determined as the representative value of real-time road snow thickness.

[0060] Through the above process, not only are multi-point measurement data considered, but also the influence of abnormal data is effectively removed through spatial analysis and iterative filtering. This makes the determined representative value more accurately reflect the overall snow cover condition of the road, providing a reliable data foundation for subsequent snow thickness prediction and early warning, and helping to improve the accuracy and stability of road environment perception.

[0061] In some embodiments, obtaining the target space center includes:

[0062] Obtain the filtering iteration function, wherein the filtering iteration function is:

[0063] ;

[0064] in, For the center of the iterative space, To measure the thickness of snow in space, To measure the thickness of the i-th snow layer in space, As the initial spatial center, The Gaussian kernel function;

[0065] The initial spatial center is iterated using the filtering iteration function to obtain the iterative spatial center. It is then determined whether the iterative spatial center satisfies the preset constraint conditions. If so, the iterative spatial center is taken as the target spatial center.

[0066] Specifically, the screening iteration function is an iterative algorithm based on kernel density estimation, used to gradually approximate the spatial center point that best represents the snow thickness of the road in the snow thickness measurement space. The function assigns weights to the snow thickness of each measurement point through a Gaussian kernel function, where the weight depends on the distance between the point and the current center point.

[0067] Specifically, The current center point of the space (initially the mode); The i-th snow thickness value in the snow thickness measurement space; Let x be the set of points in the neighborhood centered at x, where the neighborhood is the space for measuring snow thickness; It is a Gaussian kernel function used to weight neighboring points, assigning higher weights to measurement points that are closer to the center point, thereby causing the new center point to move towards areas with higher data density; This is the new spatial center obtained in the current iteration.

[0068] The above process, through the non-parametric clustering characteristics of the Mean Shift algorithm, can automatically identify dense data areas without pre-setting the number of clusters, thereby effectively filtering outliers and abnormal measurements, obtaining more representative snow thickness centers, and exhibiting stronger robustness and adaptability.

[0069] In some embodiments, the preset constraint is that the number of iterations satisfies a preset maximum number of iterations and / or the difference between the nearest neighbor quantities of the iteration space centers of two adjacent iterations within a preset scale is less than or equal to a preset difference.

[0070] Specifically, preset constraints refer to the criteria used to determine whether to stop the iteration during the screening process, in order to ensure that the iteration process converges within a reasonable time and obtains stable results.

[0071] Specifically, during the iteration process, when the number of iterations reaches the preset maximum value, that is, when the number of iterations meets the preset maximum number of iterations, the iteration will stop regardless of whether the current iteration result has fully converged, and the result of the last iteration will be taken as the final result. The maximum number of iterations should be set according to the actual application scenario and computing resources. For example, it can be set to 10 or 20 iterations.

[0072] Specifically, if the difference between the nearest neighbor values ​​of the iteration space centers of two adjacent iterations within a preset scale is less than or equal to a preset difference, it means that during the iteration process, the difference between the neighbor values ​​of the current iteration space center and the previous iteration space center is compared. If the difference is within a preset scale range (such as the spatial distance being less than or equal to a certain threshold, for example, 0.5 cm), the iteration is considered to have converged and the iteration can be stopped.

[0073] For example, the difference between the nearest neighbors of the iteration space centers of two adjacent iterations within a preset scale is less than or equal to a preset difference, which can be expressed as:

[0074] ;

[0075] in, This represents the number of points in the neighborhood centered at x with radius r. Indicates the current spatial center; This indicates the center of the previous iteration; is the threshold for the change in the number of nearest neighbors (i.e., the preset difference); r is the preset scale mentioned above.

[0076] By combining the above constraints with the efficiency control of the algorithm (maximum number of iterations) and the perception of spatial density changes (difference in the number of nearest neighbors), the algorithm ensures that the iteration process can be completed within a reasonable time and obtain sufficiently accurate results, avoiding the problems of infinite loops or premature termination of iteration, thereby improving the stability and reliability of the algorithm.

[0077] S200: Continuously perceive the target road environment using a multimodal environment sensor array, obtain a multimodal environment parameter array sequence, and perform time-series iteration on the multimodal environment parameter array sequence to determine the multimodal environment trend characteristics.

[0078] Specifically, a multimodal environmental sensor array is a network system composed of various types of sensors that can monitor key environmental parameters related to snow accumulation in real time; preferably, the multimodal environmental sensor array includes temperature sensors, humidity sensors, snowfall sensors, wind speed sensors, and road condition sensors, etc.

[0079] Specifically, the multimodal environmental parameter array sequence is a sequence of environmental parameter data collected at different time points by the aforementioned multimodal environmental sensor array composed of various types of sensors. Each time point's dataset contains measurements from all sensors, reflecting the changes in environmental parameters over time. By performing time series analysis on the multimodal environmental parameter array sequence, the changing trends and patterns of environmental parameters can be extracted, outputting corresponding multimodal environmental trend features, such as rising or falling temperature trends, increasing or decreasing snowfall trends, and humidity fluctuations. These features can reveal the potential impact of environmental factors on changes in snow depth.

[0080] For example, on a 10-kilometer-long highway, an environmental monitoring station is deployed every kilometer. Each station is equipped with temperature, humidity, snowfall, wind speed, and road surface condition sensors. Data is collected in real time and transmitted to a central data processing center via wireless communication modules, forming a multimodal environmental parameter array sequence. The dataset at each time point contains measurements from all sensors. Then, time-series iterative analysis methods, such as moving averages, exponential smoothing, or Long Short-Term Memory (LSTM) networks, are used to process this data and extract the trends in environmental parameter changes. For instance, analyzing temperature data over the past 24 hours using an LSTM model reveals a decreasing trend, with a decrease of 0.2°C per hour. Analyzing snowfall data shows a gradual increase in snowfall over the past 3 hours, with a current snowfall rate of 3 mm / hour. The acquired trend features are then integrated into multimodal environmental trend features for subsequent snow depth prediction.

[0081] Through the above process, we can comprehensively monitor various environmental factors that affect snow cover changes, and extract their dynamic change characteristics through time series analysis, providing rich data support for predicting the trend of snow cover thickness changes and enhancing the accuracy and reliability of the prediction model.

[0082] In some embodiments, a multimodal environmental sensor array is used to continuously perceive the target road environment, obtain a multimodal environmental parameter array sequence, and perform time-series iteration on the multimodal environmental parameter array sequence to determine multimodal environmental trend characteristics, including:

[0083] By sequentially interacting with adjacent multimodal environmental parameter arrays in the multimodal environmental parameter array sequence, multiple iterative multimodal environmental parameter correlation matrices are obtained. The multiple iterative multimodal environmental parameter correlation matrices are used as implicit information, and the last multimodal environmental parameter array in the multimodal environmental parameter array sequence is used as explicit information. Convolutional analysis is then performed to determine the multimodal environmental trend characteristics.

[0084] Specifically, a multimodal environmental parameter array refers to a set of environmental parameter data collected by multiple heterogeneous sensors at a certain point in time. It is usually represented as a vector or tensor, containing values ​​of multiple dimensions such as temperature, humidity, visibility, road conditions, and traffic density. Two adjacent multimodal environmental parameter arrays refer to the multimodal environmental parameter arrays at two consecutive sampling times.

[0085] Specifically, by interacting with two adjacent multimodal environmental parameter arrays, the relationship between changes in environmental parameters at adjacent time points can be analyzed, thereby constructing an iterative multimodal environmental parameter correlation matrix to quantify the degree of change and correlation of environmental parameters between adjacent time points. Common interaction methods include vector difference (change amount), vector concatenation, outer product (constructing an intermodal influence matrix), and attention mechanisms (highlighting key modes). For example, the difference or rate of change of each environmental parameter between two time points can be calculated and organized into a matrix form.

[0086] Specifically, treating multiple such iterative multimodal environmental parameter correlation matrices as implicit information implies that these matrices contain a large number of potential patterns and trends in environmental parameter changes, while treating the last multimodal environmental parameter array in the multimodal environmental parameter array sequence as explicit information is a strategy that uses the environmental parameter data at the latest time point as a direct reflection of the current environmental state.

[0087] By constructing multiple iterative multimodal environmental parameter correlation matrices, the interaction relationships between different environmental factors evolving over time can be explicitly quantified and modeled. For example, during a cooling process, a decrease in temperature may be accompanied by an increase in humidity and a slippery road surface. Through matrix interaction modeling, these implicit dynamic coupling characteristics can be effectively captured, thereby improving the ability to extract trend features in subsequent convolutional analysis stages.

[0088] Furthermore, convolutional neural networks (CNNs) or graph convolutional networks (GCNs) are used to jointly model the above information to extract multimodal environmental trend features from implicit and explicit information, i.e., to predict the future direction of environmental change or risk status. Specifically, scanning and feature extraction of data using a sliding window approach can capture the local features and trends of environmental parameters in the time series.

[0089] For example, all the correlation matrices are stacked into a three-dimensional tensor M∈R (T-1)×d×d As implicit input; E T ∈R d As explicit input; then, through a pre-defined multi-channel convolutional network, spatial convolution is performed on M, and spatial convolution is performed on E. TChannel fusion is performed; then, feature concatenation or attention weighting is applied in the fusion layer, ultimately outputting a trend feature vector F∈R. k This k-dimensional feature vector can be used for subsequent risk prediction and control decisions.

[0090] By combining implicit historical evolution information with the current explicit state, the trend of environmental change can be extracted more comprehensively. At the same time, by combining convolutional analysis, not only is the nonlinear modeling capability of feature extraction improved, but the system's ability to understand the dynamics of complex environments is also enhanced, thereby improving prediction accuracy and response time.

[0091] In some embodiments, by sequentially interacting two adjacent multimodal environmental parameter arrays in a multimodal environmental parameter array sequence, multiple iterative multimodal environmental parameter correlation matrices are obtained, including:

[0092] Extract any two adjacent multimodal environmental parameter arrays from the multimodal environmental parameter array sequence as the first multimodal environmental parameter array and the second multimodal environmental parameter array; perform environmental parameter mapping similarity recognition on the first multimodal environmental parameter array and the second multimodal environmental parameter array, and normalize the recognition results to fill the initially empty matrix to obtain the first iterative multimodal environmental parameter association matrix; and so on, interact with any two adjacent multimodal environmental parameter arrays in the multimodal environmental parameter array sequence to obtain multiple iterative multimodal environmental parameter association matrices.

[0093] Specifically, the first multimodal environmental parameter array and the second multimodal environmental parameter array refer to any pair of adjacent environmental parameter vectors in the time series. Each vector contains multiple modal parameters (such as temperature, humidity, wind speed, visibility, etc.) and can be represented as a vector of dimension d.

[0094] Specifically, the modal parameters in the first and second parameter arrays are paired and their similarity or correlation (such as cosine similarity, inverse Euclidean distance function, Pearson coefficient, etc.) are calculated to form a d×d similarity matrix, which represents the relationship between each modality at the first time point and each modality at the second time point. Then, the similarity calculation results are standardized (such as Min-Max normalization to the [0,1] interval) to eliminate the influence of dimensions and enhance the comparability between different modalities. Finally, the normalized similarity values ​​are filled into a two-dimensional matrix initialized to zero or null values ​​to form the iterative multimodal environment parameter correlation matrix for that time step.

[0095] Furthermore, repeating the above operation for all adjacent pairs in the entire sequence will eventually yield T-1 iterative multimodal environmental parameter correlation matrices.

[0096] The above process, through intermodal similarity calculation and normalization, can explicitly construct the mutual influence matrix of different modal parameters during the time evolution process, forming a structured time-modal coupling modeling method. The normalized similarity matrix, as input, not only preserves the dynamic relationship between modalities but also has good numerical stability. It can be directly used as input for neural networks, which helps to improve model training efficiency and prediction performance.

[0097] Optionally, a modal weighting mechanism can be introduced to assign weights to different modes, thereby enhancing the influence of key modes (such as visibility and road surface slipperiness) in similarity calculation.

[0098] Optionally, the sequence of correlation matrices can be processed by a graph neural network (GNN), that is, each correlation matrix can be treated as a graph structure, thereby further extracting graph-level trend features.

[0099] It is feasible to introduce a time decay factor, that is, to apply a decay weight to the correlation matrix of earlier time steps, thereby enhancing the focus on the latest changes.

[0100] S300: Based on the multimodal environmental trend characteristics, perform time-series trend prediction on the representative value of real-time road snow thickness to determine the road snow thickness prediction result.

[0101] In some embodiments, based on the multimodal environmental trend characteristics, a time-series trend prediction is performed on the representative value of real-time road snow thickness to determine the road snow thickness prediction result, including:

[0102] A pre-built time-series trend predictor is used to learn a two-to-one mapping relationship between the real-time road snow thickness representative value and multimodal environmental trend features and the road snow thickness prediction result; the time-series trend predictor is used to identify the real-time road snow thickness representative value and multimodal environmental trend features to determine the road snow thickness prediction result.

[0103] Specifically, time-series trend predictors are built on machine learning or deep learning to analyze and predict time-series data. This particular predictor learns the complex relationship between input data (representative values ​​of real-time road snow depth and multimodal environmental trend characteristics) and output data (predicted road snow depth), thereby achieving accurate predictions of the output results.

[0104] Specifically, first, a time-series trend predictor is pre-built, such as based on a deep learning model like a Long Short-Term Memory (LSTM) network or a Gated Recurrent Unit (GRU). Then, historical time-series data is acquired, including real-time road snow thickness sequences, synchronized multimodal environmental trend feature sequences, and corresponding supervision labels (future target prediction values, which are considered true values ​​and thus accurate predictions), and corresponding training sample pairs are constructed. Next, the training sample pairs are divided into training and validation sets, and the time-series trend predictor is trained using supervised learning methods (such as the mean squared error loss function) to learn the mapping relationship between the input data (representative values ​​of real-time road snow thickness and multimodal environmental trend features) and the output data (road snow thickness prediction results). The prediction performance is then validated using the validation set until a preset prediction accuracy requirement is met (e.g., prediction error not exceeding 10%).

[0105] Furthermore, during the inference phase, the representative value and trend characteristics of the current snow thickness are input into the time-series trend predictor, which identifies the new real-time road snow thickness representative value and multimodal environmental trend characteristics to output the future snow thickness prediction results.

[0106] The above method and steps use a time-series trend predictor to predict road snow thickness, which can significantly improve the accuracy and foresight of the prediction. The predictor can simultaneously consider the representative value of real-time road snow thickness and multimodal environmental trend characteristics, and combine deep learning to more accurately capture the dynamic characteristics of snow changes and provide accurate road snow thickness prediction results.

[0107] S400: Using the predicted road snow thickness and multimodal environmental trend characteristics as analysis data, perform hierarchical early warning information matching, and retrieve target information reference points in the early warning reference point information database. Merge the matched hierarchical early warning information and target information reference points to determine the target hierarchical early warning information.

[0108] Specifically, the tiered early warning information matching process involves selecting the early warning information that best matches the current situation, such as the matching early warning level, from the early warning reference point information database based on the predicted road snow depth and multimodal environmental trend characteristics. The early warning reference point information database stores early warning information under different disaster scenarios, including various reference point information such as specific information reference points (precipitation, snow depth, etc.), social comparison reference points (comparison with historical disaster events), and negative impact reference points (potential secondary disasters).

[0109] Specifically, the target information reference point is the specific reference point information (such as early warning level matching) retrieved from the above information database that is most relevant to the current prediction results and environmental characteristics. By merging the matching level early warning information and the target information reference point, more targeted early warning content can be generated.

[0110] It is feasible to analyze user feedback on different reference point information to identify which reference point information is more effective in improving early warning. This allows for the definition of the early warning effect of reference point information, facilitating the selection of reference point information with better early warning effects based on matching tiered early warning information. In other words, priority should be given to reference point information that has been verified to significantly improve user perception and response, thereby optimizing the dissemination and reception of early warning information. The early warning effect refers to the impact of early warning information on user perception and behavior, including changes in users' perception of the severity of the disaster, the risk, and their motivation to act.

[0111] In some embodiments, the predicted road snow thickness and multimodal environmental trend characteristics are used as analysis data to perform hierarchical early warning information matching, and target information reference points are retrieved from the early warning reference point information database. The matched hierarchical early warning information and target information reference points are then fused to determine the target hierarchical early warning information, including:

[0112] Multiple sample analysis data are acquired, including multiple information reference points; the multiple sample analysis data are aggregated in the same category to obtain M aggregated sample analysis data sets, where M is a positive integer; based on the mapping relationship between the sample analysis data and the information reference points, the multiple information reference points are aggregated according to the M aggregated sample analysis data sets, and the mode of the aggregation results is calculated to obtain M aggregated information reference points; the early warning reference point information database is constructed based on the M aggregated sample analysis data sets and the M aggregated information reference points.

[0113] Specifically, firstly, multiple sample analysis data are collected, including road snow depth prediction results under different scenarios and multimodal environmental trend characteristics. For example, 50 sample analysis data are collected from winter snow warning records of the past three years. Each sample includes predicted snow depth values ​​(e.g., 5 cm, 10 cm, 15 cm, etc.), environmental trend characteristics (e.g., temperature drop, increased snowfall, etc.), and information reference points used (e.g., "similar to last year's heavy snow", "may cause road congestion", etc.). Then, a clustering algorithm (e.g., K-Means) is used to aggregate the acquired sample analysis data into similar clusters. That is, based on the similarity of the predicted snow depth values ​​and environmental trend characteristics, the 50 samples are divided into 5 aggregated sample analysis data sets (i.e., when M=5).

[0114] Furthermore, based on the mapping relationship between sample analysis data and information reference points, the information reference points used in each aggregated sample set are statistically analyzed. For example, in a certain set, 15 samples used the reference point "may cause road congestion", 10 samples used the reference point "similar to last year's heavy snow", and the remaining 5 samples used other reference points. Then, the multiple information reference points are aggregated and the mode is calculated to obtain the aggregated information reference points for each set. For example, "may cause road congestion" is used as the representative reference point for the above set.

[0115] Finally, based on the acquired aggregated sample analysis data set and aggregated information reference points, an early warning reference point information database is constructed. This involves associating each aggregated sample analysis data set with its corresponding aggregated information reference point to form a mapping table. This facilitates rapid matching and retrieval of reference point information, thereby providing accurate information support for early warning and ensuring the relevance and effectiveness of the early warning content.

[0116] For example, the reference point information in the early warning reference point information database is as follows:

[0117] Table 1. Reference point information of an exemplary early warning reference point information database.

[0118] T1 The meteorological observatory issued a red alert for heavy rain: In the next 3 hours, the rainfall in the urban area of ​​T city will reach more than 100 millimeters. Citizens are advised to take precautions. Original text none T2 The meteorological observatory issued a red alert for heavy rain: In the next 3 hours, the rainfall in the urban area of ​​T city will reach more than 100 millimeters (about 81 West Lakes). Citizens are advised to take precautions. Specific information reference points (precipitation) Approximately 81 West Lakes T3 The meteorological observatory has issued a red alert for heavy rain: Rainfall in the urban area of ​​T city is expected to exceed 100 millimeters in the next 3 hours. Residents are advised to avoid unnecessary outings and stay away from dangerous areas such as rivers and tunnels. Specific information reference points (preventive measures) Avoid going out unless absolutely necessary, and stay away from dangerous areas such as rivers and tunnels. T4 The meteorological observatory issued a red alert for heavy rain: In the next 3 hours, the rainfall in the urban area of ​​T city will reach more than 100 mm (slightly lower than the "July 20th rainstorm"). Citizens are advised to take precautions. Social Comparison Reference Points Slightly lower than the "July 20th rainstorm" T5 The meteorological observatory issued a red alert for heavy rain: In the next 3 hours, the rainfall in the urban area of ​​T city will reach more than 100 millimeters, which may cause water accumulation in low-lying areas, road blockage, and secondary disasters such as floods and landslides. Citizens are advised to take precautions. Negative impact reference points This could lead to waterlogging in low-lying areas, road blockages, and secondary disasters such as floods and landslides. T6 The meteorological observatory has issued a red alert for heavy rain: In the next 3 hours, rainfall in the urban area of ​​T city will exceed 100 millimeters (approximately 81 West Lakes), which may cause flooding in low-lying areas, road blockages, and secondary disasters such as floods and landslides. Residents are advised to take precautions. Avoid unnecessary travel and stay away from dangerous areas such as rivers and tunnels. Comprehensive reference points Approximately 81 locations around West Lake (see reference points for specific information) may experience flooding in low-lying areas, road blockages, and secondary disasters such as floods and landslides. Residents are advised to take precautions. Avoid unnecessary outings and stay away from dangerous areas such as rivers and tunnels (see reference points for specific information). T7 The meteorological observatory issued a red alert for heavy rain: In the next 3 hours, the rainfall in the urban area of ​​T city will reach more than 100 mm (slightly lower than the "July 20 rainstorm"), which may cause water accumulation in low-lying areas, road blockage, and secondary disasters such as floods and landslides. Citizens are advised to take precautions. Comprehensive reference points Slightly lower than the "July 20th rainstorm" (social comparison reference point) / May cause waterlogging in low-lying areas, road blockages, and secondary disasters such as floods and landslides.

[0119] In some embodiments, the intelligent classification and early warning method for road snow thickness further includes:

[0120] The confidence score of the multimodal environmental trend characteristics is calculated, and the confidence interval is determined based on the confidence score. The confidence interval is used as the early warning period to periodically issue early warnings on the snow thickness of the target road.

[0121] Specifically, the reliability score obtained through the reliability score quantitatively characterizes the stability of the multimodal environmental trend features, that is, the intensity of the change in the environmental trend. The better the reliability score, the more stable the environmental trend of the target scene is, and a larger reliability interval can be configured. Conversely, if the reliability score is worse, the environment of the target scene can be considered to be more variable, and a shorter reliability interval needs to be set, thereby providing a higher frequency (shorter period) snow thickness warning and improving the timeliness and effectiveness of the warning.

[0122] Specifically, a confidence interval is a time range determined based on a confidence score. Within this time range, multimodal environmental trend characteristics are considered to have high reliability and can be used as a basis for early warning. For example, a confidence score of 0.8 for a temperature decrease trend corresponds to a confidence interval of approximately 2 hours, meaning that a temperature decrease trend within the next 2 hours has high reliability.

[0123] Optionally, based on the degree or importance of different environmental indicators on traffic safety, a credibility score is performed on the multimodal environmental trend characteristics. First, by analyzing historical data and relevant research, the degree or importance of different environmental indicators on traffic safety is determined, defined as the importance weights of indicators such as temperature, snowfall, wind speed, and humidity. For example, assuming a weight of 0.4 for temperature, 0.3 for snowfall, 0.2 for wind speed, and 0.1 for humidity. Next, the volatility indicators for each environmental indicator are calculated, such as volatility and information entropy. For example, for temperature data, its volatility (e.g., standard deviation) and information entropy over the past 24 hours can be calculated. Volatility can be measured by calculating the standard deviation of the temperature data; a larger standard deviation indicates higher volatility. Information entropy is measured by calculating the probability distribution of the temperature data; a more uniform distribution indicates higher information entropy. Finally, based on the importance weights obtained from the analysis, the weighted volatility indicators are used as the credibility score.

[0124] Optionally, the continuity score result can be determined by separately scoring the continuity, synergy, noise interference, and historical fit of the multimodal environmental trend characteristics and then summing them by weight.

[0125] By scoring the credibility of multimodal environmental trend characteristics and determining the credibility interval, the reliability and effectiveness of early warning can be significantly improved, ensuring that early warning information is always based on the most reliable and up-to-date environmental trend characteristics, reflecting changes in snow accumulation in a timely manner, and providing an accurate time window for snow removal and traffic management.

[0126] In summary, the intelligent classification and early warning method for road snow thickness provided by this invention has the following technical effects:

[0127] By measuring snow thickness at multiple points along the target road and filtering the results, a representative value for real-time road snow thickness is determined. Next, a multimodal environmental sensor array is used to continuously perceive the target road environment, acquiring a sequence of multimodal environmental parameters. This sequence is then iterated over time to determine the trend characteristics of the multimodal environment. Based on these trend characteristics, a time-series trend prediction is performed on the representative value of real-time road snow thickness, yielding a predicted snow thickness. Finally, the predicted snow thickness and the multimodal environmental trend characteristics are used as analytical data for tiered early warning information matching. Relevant target information reference points are retrieved from the early warning reference point database. The matched tiered early warning information is then fused with the target information reference points to ultimately determine the target tiered early warning information. This approach improves monitoring accuracy, provides a prediction-based dynamic response and intelligent early warning mechanism, and enhances the technical effectiveness of early warning.

[0128] Example 2, as Figure 2 This is a schematic diagram of the intelligent graded early warning system for road snow thickness according to the present invention. For example, Figure 1 The flowchart of the intelligent graded early warning method for road snow thickness of the present invention can be seen as follows: Figure 2 The structure shown is implemented.

[0129] Based on the same concept as the intelligent graded early warning method for road snow thickness in the embodiments described above, the present invention also provides an intelligent graded early warning system for road snow thickness, comprising:

[0130] The snow thickness measurement module 11 is used to measure the snow thickness at multiple points on the target road and to filter the measurement results to determine the representative value of the real-time road snow thickness.

[0131] The continuous sensing and trend acquisition module 12 is used to continuously sense the target road using a multimodal environment sensor array, obtain a multimodal environment parameter array sequence, and perform time-series iteration on the multimodal environment parameter array sequence to determine the multimodal environment trend characteristics.

[0132] The thickness prediction module 13 is used to perform time-series trend prediction on the representative value of real-time road snow thickness based on the multimodal environment trend characteristics, and determine the road snow thickness prediction result.

[0133] The graded early warning module 14 is used to match graded early warning information with the predicted road snow thickness and multimodal environmental trend characteristics as analysis data, and to retrieve target information reference points in the early warning reference point information database. The matched graded early warning information and target information reference points are then fused to determine the target graded early warning information.

[0134] In some embodiments, the snow thickness measurement module 11 includes:

[0135] The snow thickness measurement and space construction unit is used to perform multi-point snow thickness measurement on the target road, obtain measurement results, and construct a snow thickness measurement space based on the measurement results.

[0136] The snow thickness mode extraction unit is used to extract the mode of snow thickness from the measurement results and use it as the initial spatial center of the snow thickness measurement space.

[0137] The iterative analysis and representative value determination unit is used to perform iterative analysis in the snow thickness measurement space using a screening iterative function until the preset constraint conditions are met, obtain the target space center, and take the snow thickness corresponding to the target space center as the representative value of the real-time road snow thickness.

[0138] In some implementations, the execution steps of the iterative analysis and representative value determination unit in the snow thickness measurement module 11 also include:

[0139] Obtain the filtering iteration function, wherein the filtering iteration function is:

[0140] ;

[0141] in, For the center of the iterative space, To measure the thickness of snow in space, To measure the thickness of the i-th snow layer in space, As the initial spatial center, This is the Gaussian kernel function.

[0142] The initial spatial center is iterated using the filtering iteration function to obtain the iterative spatial center. It is then determined whether the iterative spatial center satisfies the preset constraint conditions. If so, the iterative spatial center is taken as the target spatial center.

[0143] In some implementations, the preset constraint is that the number of iterations satisfies a preset maximum number of iterations and / or the difference between the nearest neighbors of the iteration space centers of two adjacent iterations within a preset scale is less than or equal to a preset difference.

[0144] In some embodiments, the continuous sensing and trend acquisition module 12 includes:

[0145] The iterative correlation matrix generation unit is used to sequentially interact with two adjacent multimodal environmental parameter arrays in the multimodal environmental parameter array sequence to obtain multiple iterative multimodal environmental parameter correlation matrices.

[0146] The trend feature determination unit is used to perform convolution analysis on the multiple iterative multimodal environmental parameter correlation matrices as implicit information and the last multimodal environmental parameter array in the multimodal environmental parameter array sequence as explicit information to determine the multimodal environmental trend features.

[0147] In some implementations, the iterative correlation matrix generation unit in the continuous sensing and trend acquisition module 12 includes:

[0148] An environmental parameter array extraction unit is used to extract any two adjacent multimodal environmental parameter arrays from the multimodal environmental parameter array sequence as the first multimodal environmental parameter array and the second multimodal environmental parameter array.

[0149] The environmental parameter mapping similarity recognition unit is used to perform environmental parameter mapping similarity recognition on the first multimodal environmental parameter array and the second multimodal environmental parameter array, and to normalize the recognition results and fill them into the initially empty matrix to obtain the first iterative multimodal environmental parameter association matrix.

[0150] The multimodal environment parameter interaction unit is used to interact between two adjacent multimodal environment parameter arrays in the multimodal environment parameter array sequence in a similar manner to obtain multiple iterative multimodal environment parameter correlation matrices.

[0151] In some embodiments, the thickness prediction module 13 includes:

[0152] A time-series trend predictor pre-construction unit is used to pre-construct a time-series trend predictor, wherein the time-series trend predictor is used to learn a two-to-one mapping relationship between the real-time road snow thickness representative value and multimodal environmental trend features and the road snow thickness prediction result.

[0153] The road snow thickness prediction unit is used to identify the real-time road snow thickness representative value and multimodal environmental trend characteristics using the time-series trend predictor, and to determine the road snow thickness prediction result.

[0154] In some embodiments, the graded early warning module 14 includes:

[0155] The sample analysis data acquisition unit is used to acquire multiple sample analysis data, which include multiple information reference points.

[0156] The sample analysis data aggregation unit is used to perform similar aggregation on the multiple sample analysis data to obtain M aggregated sample analysis data sets, where M is a positive integer.

[0157] The information reference point aggregation and mode calculation unit is used to aggregate the multiple information reference points based on the mapping relationship between sample analysis data and information reference points, according to the M aggregated sample analysis data sets, and to calculate the mode of the aggregation results to obtain M aggregated information reference points.

[0158] The early warning reference point information database construction unit is used to construct the early warning reference point information database based on the M aggregated sample analysis data sets and the M aggregated information reference points.

[0159] In some embodiments, the intelligent snow thickness classification and early warning system further includes:

[0160] The credibility scoring and interval determination unit is used to score the credibility of the trend features of the multimodal environment and determine the credibility interval based on the credibility scoring results.

[0161] A periodic snow thickness warning unit is used to provide periodic warnings of snow thickness for the target road, with the confidence interval as the warning period.

[0162] It should be understood that the focus of the embodiments mentioned in this specification is their difference from other embodiments. The specific embodiments in the aforementioned Embodiment 1 are also applicable to the intelligent graded early warning system for road snow thickness described in Embodiment 2. For the sake of brevity, they will not be elaborated further here.

[0163] It should be understood that the embodiments disclosed in this invention and the above description enable those skilled in the art to implement this invention. However, this invention is not limited to the embodiments mentioned above. It should be understood that those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this invention, and should all be included within the protection scope of this invention.

Claims

1. A method for intelligent classification and early warning of road snow depth, characterized in that, include: Multi-point snow thickness measurements were performed on the target road, and the measurement results were filtered to determine the representative value of real-time road snow thickness. The target road is continuously perceived using a multimodal environmental sensor array to obtain a multimodal environmental parameter array sequence, and the multimodal environmental parameter array sequence is iterated over time to determine the multimodal environmental trend characteristics. Based on the multimodal environmental trend characteristics, the time-series trend prediction of the representative value of real-time road snow thickness is performed to determine the prediction result of road snow thickness; The predicted road snow thickness and multimodal environmental trend characteristics are used as analysis data to match graded early warning information. Target information reference points are retrieved from the early warning reference point information database. The matched graded early warning information and target information reference points are then fused to determine the target graded early warning information.

2. The intelligent graded early warning method for road snow thickness as described in claim 1, characterized in that, Continuous environmental perception of the target road is achieved using a multimodal environmental sensor array, obtaining a multimodal environmental parameter array sequence. This sequence is then iterated over time to determine multimodal environmental trend characteristics, including: By sequentially interacting with adjacent multimodal environment parameter arrays in the multimodal environment parameter array sequence, multiple iterative multimodal environment parameter correlation matrices are obtained; The correlation matrices of the multiple iterative multimodal environmental parameters are used as implicit information, and the last multimodal environmental parameter array in the multimodal environmental parameter array sequence is used as explicit information. Convolutional analysis is then performed to determine the trend characteristics of the multimodal environment.

3. The intelligent graded early warning method for road snow thickness as described in claim 2, characterized in that, By sequentially interacting with adjacent multimodal environment parameter arrays in the sequence, multiple iterative multimodal environment parameter correlation matrices are obtained, including: Extract any two adjacent multimodal environmental parameter arrays from the multimodal environmental parameter array sequence, and use them as the first multimodal environmental parameter array and the second multimodal environmental parameter array; Environmental parameter mapping similarity recognition is performed on the first multimodal environmental parameter array and the second multimodal environmental parameter array, and the recognition results are normalized and filled into the initially empty matrix to obtain the first iterative multimodal environmental parameter association matrix; Similarly, by interacting with adjacent multimodal environmental parameter arrays in the multimodal environmental parameter array sequence, multiple iterative multimodal environmental parameter correlation matrices are obtained.

4. The intelligent graded early warning method for road snow thickness as described in claim 1, characterized in that, Based on the multimodal environmental trend characteristics, a time-series trend prediction is performed on the representative value of real-time road snow thickness to determine the road snow thickness prediction result, including: A pre-built time-series trend predictor is used to learn a two-to-one mapping relationship between the real-time representative value of road snow thickness and multimodal environmental trend features and the road snow thickness prediction result. The time-series trend predictor is used to identify the representative value of real-time road snow thickness and multimodal environmental trend characteristics to determine the predicted result of road snow thickness.

5. The intelligent graded early warning method for road snow thickness as described in claim 1, characterized in that, Using the predicted road snow thickness and multimodal environmental trend characteristics as analysis data, graded early warning information is matched, and target information reference points are retrieved from the early warning reference point information database. The matched graded early warning information and target information reference points are then fused to determine the target graded early warning information, including: Acquire multiple sample analysis data, which include multiple information reference points; The multiple sample analysis data are aggregated in the same category to obtain M aggregated sample analysis data sets, where M is a positive integer; Based on the mapping relationship between sample analysis data and information reference points, the multiple information reference points are aggregated according to the M aggregated sample analysis data sets, and the mode of the aggregation result is calculated to obtain M aggregated information reference points. The early warning reference point information database is constructed based on the M aggregated sample analysis data sets and the M aggregated information reference points.

6. The intelligent graded early warning method for road snow thickness as described in claim 1, characterized in that, include: The credibility score of the multimodal environmental trend characteristics is calculated, and the credibility interval is determined based on the credibility score results; Using the aforementioned confidence interval as the warning period, periodic warnings are issued for the snow thickness of the target road.

7. The intelligent graded early warning method for road snow thickness as described in claim 1, characterized in that, Multi-point snow thickness measurements were performed on the target road, and the measurement results were filtered to determine representative values ​​of real-time road snow thickness, including: The snow thickness is measured at multiple points on the target road to obtain the measurement results, and a snow thickness measurement space is constructed based on the measurement results. Extract the mode of snow thickness from the measurement results and use it as the initial spatial center of the snow thickness measurement space; The snow thickness measurement space is iteratively analyzed using a filtering iteration function until the preset constraints are met, the target space center is obtained, and the snow thickness corresponding to the target space center is used as the representative value of the real-time road snow thickness.

8. The intelligent graded early warning method for road snow thickness as described in claim 7, characterized in that, Obtaining the target space center includes: Obtain the filtering iteration function, wherein the filtering iteration function is: ； in, For the center of the iterative space, To measure the thickness of multiple snow layers in space, To measure the thickness of the i-th snow layer in space, As the initial spatial center, The Gaussian kernel function; The initial spatial center is iterated using the filtering iteration function to obtain the iterative spatial center. It is then determined whether the iterative spatial center satisfies the preset constraint conditions. If so, the iterative spatial center is taken as the target spatial center.

9. The intelligent graded early warning method for road snow thickness as described in claim 7, characterized in that, The preset constraints are that the number of iterations satisfies the preset maximum number of iterations and / or the difference between the nearest neighbor quantities of the iteration space centers of two adjacent iterations within a preset scale is less than or equal to a preset difference.

10. A road snow depth intelligent classification and early warning system, characterized in that, The method for implementing the intelligent graded early warning method for road snow thickness as described in any one of claims 1-9 includes: The snow thickness measurement module is used to measure the snow thickness at multiple points on the target road, and to filter the measurement results to determine the representative value of the real-time road snow thickness. The continuous sensing and trend acquisition module is used to continuously sense the target road using a multimodal environment sensor array, obtain a multimodal environment parameter array sequence, and perform time-series iteration on the multimodal environment parameter array sequence to determine multimodal environment trend characteristics. The thickness prediction module is used to perform time-series trend prediction on the representative value of real-time road snow thickness based on the multimodal environmental trend characteristics, and determine the road snow thickness prediction result. The graded early warning module is used to match graded early warning information with the predicted road snow thickness and multimodal environmental trend characteristics as analysis data, and to retrieve target information reference points in the early warning reference point information database. The matched graded early warning information and target information reference points are then fused to determine the target graded early warning information.

Images