CN-121997775-A - Icing prediction method and system based on multi-source data fusion

CN121997775ACN 121997775 ACN121997775 ACN 121997775ACN-121997775-A

Abstract

The application provides an icing prediction method and system based on multi-source data fusion, and relates to the technical field of data processing, wherein the method comprises the steps of constructing a historical data tensor field, constructing a real-time query tensor based on a residual effectiveness value of a real-time snow-melting agent, and calculating similarity scores of the real-time query tensor and the historical data tensor field to form a historical data subset; forming a discrete point cloud based on the historical data subset, generating a geometric distribution index to identify evolution states, obtaining a freezing point based on each evolution state, generating a physical icing probability according to the freezing point and the ground temperature, generating a predicted icing probability based on a pre-constructed machine learning model, and generating a final icing probability. The accurate quantification of the real ice resistance of the road surface is realized, and the accuracy and the reliability of ice probability prediction under complex meteorological conditions are remarkably improved.

Inventors

Pang Tengyu
WANG ENLI
LIANG ZHU
XI JIN
HUANG MIAO
GUO NIANSHENG
YUAN RUI
YE ZHIWEN
DU SAISAI

Assignees

安徽省交规院工程智慧养护科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260407

Claims (9)

1. The icing prediction method based on multi-source data fusion is characterized by comprising the following steps of: step 1, obtaining the ground temperature; Step 2, presetting an initial set efficacy value, collecting accumulated rainfall and accumulated vehicle flow data in real time, and calculating a total loss proportion; Step 3, constructing a historical data tensor field, constructing a real-time query tensor based on the residual effectiveness value of the real-time snow-melting agent, calculating similarity scores of the real-time query tensor and the historical data tensor field one by one, arranging all similarity scores in a descending order, intercepting the historical data tensor field with the highest k groups of similarity scores to form a similar historical data subset, mapping the historical data subset into a two-dimensional phase space taking the effectiveness value of the historical snow-melting agent as a horizontal axis and the historical measured freezing point as a vertical axis based on the historical data subset to form a discrete point cloud, generating a plurality of geometric distribution indexes based on the discrete point cloud, identifying evolution states based on the geometric distribution indexes, and obtaining the freezing point based on the magnitude relation between the geometric distribution indexes and preset target values based on the evolution states; Step 4, generating physical icing probability according to the freezing point and the ground temperature, constructing a forecast feature vector based on real-time weather forecast data, and inputting the forecast feature vector into a pre-constructed machine learning model to generate the forecast icing probability; and 5, generating a final icing probability according to the physical icing probability and the predicted icing probability.
2. The method of claim 1, wherein a historical meteorological feature vector, a historical snowmelt residual effectiveness value, and a historical measured freeze point are obtained and together form a multi-dimensional historical data tensor field.
3. The icing prediction method based on multi-source data fusion according to claim 1, wherein a real-time weather feature vector is obtained, a scalar value is obtained by normalizing a residual effectiveness value of a real-time snow-melting agent, and the scalar value and the real-time weather feature vector are spliced to generate a real-time query tensor.
4. The method for predicting ice based on multi-source data fusion as recited in claim 1, wherein the geometric distribution indexes include a linear concentration degree index, a distribution discrete degree index, an evolution trend direction index and a terminal bending characteristic index; Calculating a covariance matrix of the discrete point cloud, solving a maximum eigenvalue and a minimum eigenvalue of the covariance matrix, constructing an eigenvalue according to the maximum eigenvalue and the minimum eigenvalue, and calculating based on the eigenvalue to obtain a linear concentration degree index; Calculating the centroid coordinates of the discrete point cloud, calculating the Euclidean distance from each data point in the discrete point cloud to the centroid coordinates, calculating the arithmetic average value of all Euclidean distances, and taking the arithmetic average value as a distribution discrete degree index; Calculating the slope of the eigenvector, and taking the slope as an evolution trend direction index; The historical data subsets are sequenced from small to large according to time stamps, the last m data points after sequencing are selected, the m data points are fitted into a quadratic curve equation, the curvature absolute value of the quadratic curve equation at the residual effectiveness value of the real-time snow melting agent is calculated, and the end bending characteristic index is obtained after the curvature absolute value is normalized.
5. The method according to claim 4, wherein if the linear concentration degree index is greater than a first preset target value and the evolution trend direction index is negative, the method is in a dissolved state, and performs similarity calculation on the historical data subsets to obtain a similarity value, wherein the similarity value is used as a basic weight, and a weighted average is calculated with the historical measured freezing point of each historical data pair of the historical data subsets and is used as the freezing point.
6. The method according to claim 4, wherein if the distribution dispersion degree index is greater than a second preset target value and the terminal bending characteristic index is greater than a third preset target value, the method is in a saturated precipitation transition state, and an upper quartile of all history measured freezing points of the history data pairs of the history data subset is used as the freezing point.
7. The method for predicting ice formation based on multi-source data fusion according to claim 4, wherein if the evolution trend direction index is zero or more and the linear concentration degree index is smaller than a fourth preset target value, the method is in a failure accumulation state, and a maximum value of all history measured ice points of the history data pairs of the history data subset is used as the ice point.
8. The icing prediction method based on multi-source data fusion according to claim 1, wherein a first weight and a second weight are preset, and the final icing probability is obtained based on weighted summation of the physical icing probability, the first weight, the predicted icing probability and the second weight.
9. An icing prediction system based on multi-source data fusion for performing an icing prediction method based on multi-source data fusion as claimed in any of the claims 1-8, comprising the following modules: the first data acquisition module is used for acquiring the ground temperature; The second data acquisition module is used for presetting an initial set efficacy value, acquiring accumulated rainfall and accumulated vehicle flow data in real time, and calculating the total loss proportion; The data processing module is used for constructing a historical data tensor field, constructing a real-time query tensor based on the residual effectiveness value of the real-time snow-melting agent, calculating similarity scores of the real-time query tensor and the historical data tensor field one by one, descending the similarity scores, and arranging all the similarity scores in a descending order, intercepting the historical data tensor field with the highest k groups of similarity scores before intercepting to form a similar historical data subset, mapping the historical data subset into a two-dimensional phase space taking the effectiveness value of the historical snow-melting agent as a horizontal axis and the historical measured freezing point as a vertical axis to form a discrete point cloud based on the historical data subset, and generating a plurality of geometric distribution indexes based on the discrete point cloud; The prediction module is used for generating physical icing probability according to the freezing point and the ground temperature, constructing a prediction feature vector based on real-time weather prediction data, inputting the prediction feature vector into a pre-constructed machine learning model and generating predicted icing probability; And the fusion module is used for generating a final icing probability according to the physical icing probability and the predicted icing probability.

Description

Icing prediction method and system based on multi-source data fusion Technical Field The application relates to the technical field of data processing, in particular to an icing prediction method and system based on multi-source data fusion. Background The road surface icing in winter is one of main reasons for inducing malignant traffic accidents, and the accurate early warning of the road surface icing is a core link of road monitoring and maintenance. However, the conventional pavement icing prediction scheme has a fundamental technical defect in practical application that a prediction model only depends on real-time meteorological data, and completely ignores the dynamic change process of the residual effectiveness of the existing snow-melting agent on the pavement, so that the real ice resistance of the pavement cannot be accurately judged. Specifically, the existing method cannot identify the specific stage of the residual effectiveness of the snow-melting agent at present, namely, the road surface is in a 'dissolved state' in which salt is rapidly acting, a 'saturated transition state' in which the concentration reaches critical balance, a 'accumulated state' in which the salt is exhausted or fails, or an 'uncertain state' in which the salt is severely disturbed, so that the model cannot distinguish the huge difference of actual icing risks of different road surfaces when facing the same meteorological conditions. The loss of the internal state of the pavement directly causes serious distortion of the icing probability prediction result, namely false alarm of icing risk when salinity is still effective or false alarm of icing hidden danger when the salinity is invalid. Because of the lack of dynamic feedback on the real evolution state of the pavement, the existing prediction system is difficult to adapt to complex and changeable weather disturbance and traffic flow influence, and the accuracy and reliability of prediction are obviously reduced in extreme weather or long-duration ice and snow disasters. Disclosure of Invention The present application aims to solve at least one of the technical problems in the related art to some extent. To achieve the above objective, an embodiment of the present application provides an icing prediction method and system based on multi-source data fusion, including the following steps: step 1, obtaining the ground temperature; Step 2, presetting an initial set efficacy value, collecting accumulated rainfall and accumulated vehicle flow data in real time, and calculating a total loss proportion; Step 3, constructing a historical data tensor field, constructing a real-time query tensor based on the residual effectiveness value of the real-time snow-melting agent, calculating similarity scores of the real-time query tensor and the historical data tensor field one by one, arranging all similarity scores in a descending order, intercepting the historical data tensor field with the highest k groups of similarity scores to form a similar historical data subset, mapping the historical data subset into a two-dimensional phase space taking the effectiveness value of the historical snow-melting agent as a horizontal axis and the historical measured freezing point as a vertical axis based on the historical data subset to form a discrete point cloud, generating a plurality of geometric distribution indexes based on the discrete point cloud, identifying evolution states based on the geometric distribution indexes, and obtaining the freezing point based on the magnitude relation between the geometric distribution indexes and preset target values based on the evolution states; Step 4, generating physical icing probability according to the freezing point and the ground temperature, constructing a forecast feature vector based on real-time weather forecast data, and inputting the forecast feature vector into a pre-constructed machine learning model to generate the forecast icing probability; and 5, generating a final icing probability according to the physical icing probability and the predicted icing probability. Further, a historical meteorological feature vector, a historical snow-melting agent residual effectiveness value and a historical measured freezing point are obtained, and a multi-dimensional historical data tensor field is formed together. Further, a real-time meteorological feature vector is obtained, the residual effectiveness value of the real-time snow-melting agent is normalized to obtain a scalar value, and the scalar value and the real-time meteorological feature vector are spliced to generate a real-time query tensor. Further, the geometric distribution indexes comprise a linear concentration degree index, a distribution discrete degree index, an evolution trend direction index and a tail end bending characteristic index; Calculating a covariance matrix of the discrete point cloud, solving a maximum eigenvalue and a minimum eigenvalue of the covariance matrix, cons