CN-121998381-A - Machine learning-based municipal road maintenance demand prediction method and system

CN121998381ACN 121998381 ACN121998381 ACN 121998381ACN-121998381-A

Abstract

The application provides a method and a system for predicting maintenance requirements of municipal roads based on machine learning, which relate to the technical field of machine learning; the method comprises the steps of carrying out differential calculation on service starting time and sampling time points in each sampling period to construct a space-time correlation matrix, taking attenuation coefficients in the space-time correlation matrix as dynamic adjustment factors to obtain a bit stream sequence, mapping the bit stream sequence as an observation sequence into a preset road raster graph, determining an optimal transfer path through optimal path planning, determining a raster step length from the optimal transfer path from a road state at the current time to a maintenance state, and determining maintenance requirement prediction information comprising maintenance priority and predicted maintenance date. The accuracy, objectivity and foresight of municipal road maintenance demand prediction are realized.

Inventors

ZHOU ZHENG
XU XIAOYA
LIU KANG
LI YUNPENG
ZHANG YAN
CHEN WEI
LI ZHIWEI
WANG KAI
LI MING
HUANG WANJUN
XIONG XIAOWEI
CHEN YUEXIN
LIAO JUNWEN

Assignees

四川集思数源信息技术有限公司

Dates

Publication Date: 20260508
Application Date: 20260408

Claims (10)

1. The municipal road maintenance demand prediction method based on machine learning is characterized by comprising the following steps of: acquiring service starting time of municipal roads and traffic frequency and full-load proportion of heavy vehicles in a plurality of sampling periods; Performing differential calculation on the service starting moment and sampling time points in each sampling period to obtain an attenuation coefficient sequence, performing time domain pulse transformation on the traffic frequency and the full-load proportion to generate a pulse sequence, and constructing a space-time correlation matrix by performing time sequence alignment on the attenuation coefficient sequence and the pulse sequence; Taking the attenuation coefficient in the space-time correlation matrix as a dynamic adjustment factor, and carrying out pulse coding on the pulse sequence by utilizing pulse coding modulation to obtain a bit stream sequence; Mapping the bit stream sequence as an observation sequence into a preset road raster pattern, calculating metric values of transition branches between different road states in the road raster pattern through a maximum likelihood criterion, and determining an optimal transition path through optimal path planning according to the metric values, wherein the road states comprise a health state, a sub-health state, a damage state and a maintenance state; determining a grid step length of transferring the road state at the current moment to the maintenance state from the optimal transfer path, determining the remaining time of the municipal road entering a maintenance period based on the grid step length, and determining maintenance requirement prediction information comprising maintenance priority and predicted maintenance date based on the remaining time and preset road weight.
2. The method of claim 1, wherein performing differential computation on the service start time and sampling time points in each sampling period to obtain an attenuation coefficient sequence, performing time domain pulse transformation on the traffic frequency and the full load proportion to generate a pulse sequence, and performing time sequence alignment on the attenuation coefficient sequence and the pulse sequence to construct a space-time correlation matrix, wherein the method comprises the steps of: calculating a time interval between the service starting moment and a sampling time point in each sampling period to obtain a service step sequence, and carrying out nonlinear mapping on the service step sequence by using a preset attenuation function to obtain an attenuation coefficient sequence; determining signal amplitude based on the full load proportion, determining signal density based on the traffic frequency, calculating interval time of pulse distribution by using the signal density, determining a plurality of trigger moments in the sampling period according to the interval time, and generating corresponding pulse signals at each trigger moment by taking the signal amplitude as pulse height to obtain a pulse sequence; And according to the sequence of sampling time points in the sampling period, carrying out position mapping by taking the attenuation coefficient in the attenuation coefficient sequence as a longitudinal component and taking the pulse signal in the pulse sequence as a transverse component, and constructing a space-time correlation matrix.
3. The method of claim 1, wherein pulse coding the pulse sequence with pulse code modulation using attenuation coefficients in the space-time correlation matrix as dynamic adjustment factors to obtain a bit stream sequence comprises: Taking the attenuation coefficient in the space-time correlation matrix as a dynamic adjustment factor, and matching the dynamic adjustment factor with a preset quantization mapping rule to obtain a plurality of quantization steps; Dividing an amplitude range corresponding to the pulse sequence into a plurality of quantization intervals based on the quantization step distance, and respectively classifying the pulse height of each pulse signal in the pulse sequence into corresponding quantization intervals to determine the quantization step corresponding to each pulse signal; and converting each quantization step into a corresponding binary code element by using a preset coding instruction to obtain a bit stream sequence.
4. The method according to claim 1, wherein mapping the bit stream sequence as an observation sequence into a preset road trellis diagram, and calculating metric values of transition branches between different road states in the road trellis diagram by a maximum likelihood criterion, comprises: Based on the evolution sequence of the road state, unidirectional connection is carried out on road state nodes between adjacent sampling time points in the road grid graph to obtain a plurality of transition branches, and each sampling time point in the road grid graph comprises four road state nodes corresponding to a health state, a sub-health state, a damage state and a maintenance state; Based on the sequence of the sampling time points, arranging the transfer branches corresponding to each sampling time point to obtain a plurality of candidate transfer paths; Intercepting a corresponding observation fragment from the bit stream sequence according to the sampling period to which each branch belongs, and comparing the similarity of each observation fragment with each branch in the corresponding sampling period based on a preset state feature code base to obtain the branch matching probability of each branch; And calculating the log likelihood probability of each branch under the constraint of the observation sequence based on the branch matching probability corresponding to each branch, and determining the log likelihood probability as a measurement value.
5. The method of claim 4, wherein calculating a log likelihood probability for each branch transition under the observation sequence constraint based on a branch match probability for each branch transition, and determining the log likelihood probability as a metric value, comprises: Determining a starting node to which each branch is connected in the road raster graph; In the first sampling period, determining a preset initial score as a path score, and summing the path score and branch matching probabilities corresponding to each transfer branch led out by the initial node to obtain the log likelihood probability of each transfer branch; In a subsequent sampling period, determining a path score according to the maximum log likelihood probability in all transfer branches pointed to the starting node at a sampling time point corresponding to a previous sampling period, and respectively summing the path score with a branch matching probability corresponding to each transfer branch led out by the starting node to obtain the log likelihood probability of each transfer branch; the log likelihood probability is determined as a metric value for the branch transition.
6. The method of claim 4, wherein determining an optimal transfer path from the metric value via an optimal ‌ path plan comprises: Determining a termination sampling time point of the observation sequence, and screening out a maximum metric value from metric values of the transfer branches corresponding to the termination sampling time point; And taking the transfer branch corresponding to the maximum measurement value as a backtracking starting point, carrying out state path backtracking in the road raster graph according to the reverse sequence direction of the sampling time point to obtain a corresponding node sequence, and determining the node sequence as an optimal transfer path.
7. The method according to claim 1, wherein determining a grid step for a road state transition from the optimal transition path to the maintenance state at a current time, determining a remaining time for the town road to enter a maintenance period based on the grid step, and determining maintenance demand prediction information including a maintenance priority and an expected maintenance date based on the remaining time and a preset road weight, comprises: Determining a node position corresponding to the current moment and an end position corresponding to the maintenance state in the optimal transfer path; taking the number of sampling points included between the node position and the end position as a grid step length, and calculating the product of the grid step length and a preset unit sampling time length to obtain the residual time of the municipal road entering a maintenance period; And based on a preset priority judgment matrix, carrying out cross matching on the emergency grade corresponding to the residual time and the importance grade corresponding to the road weight to determine maintenance priority, and carrying out addition operation on the residual time and the current time to obtain predicted maintenance date so as to determine maintenance requirement prediction information comprising the maintenance priority and the predicted maintenance date.
8. Machine learning-based municipal road maintenance demand prediction system, characterized by comprising: The acquisition module is used for acquiring the service starting time of the municipal road and the traffic frequency and full load proportion of the heavy-duty vehicle in a plurality of sampling periods; The calculation module is used for carrying out differential calculation on the service starting moment and sampling time points in each sampling period to obtain an attenuation coefficient sequence, carrying out time domain pulse transformation on the traffic frequency and the full-load proportion to generate a pulse sequence, and constructing a space-time correlation matrix by carrying out time sequence alignment on the attenuation coefficient sequence and the pulse sequence; The coding module is used for carrying out pulse coding on the pulse sequence by using the attenuation coefficient in the space-time correlation matrix as a dynamic adjustment factor and utilizing pulse coding modulation to obtain a bit stream sequence; The mapping module is used for mapping the bit stream sequence as an observation sequence into a preset road grid graph, calculating metric values of transition branches between different road states in the road grid graph through a maximum likelihood criterion, and determining an optimal transition path through optimal path planning according to the metric values, wherein the road states comprise a health state, a sub-health state, a damage state and a maintenance state; the determining module is used for determining a grid step length of transferring the road state at the current moment to the maintenance state from the optimal transfer path, determining the remaining time of the municipal road entering a maintenance period based on the grid step length, and determining maintenance requirement prediction information comprising maintenance priority and predicted maintenance date based on the remaining time and preset road weight.
9. An electronic device, comprising: A memory for storing a computer program; A processor for implementing the steps of the machine learning based town road maintenance demand prediction method of any one of claims 1 to 7 when executing the computer program.
10. A computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium, and when executed by a processor, the computer program is capable of implementing the machine learning-based town road maintenance demand prediction method of any one of claims 1 to 7.

Description

Machine learning-based municipal road maintenance demand prediction method and system Technical Field The application relates to the technical field of machine learning, in particular to a municipal road maintenance requirement prediction method and system based on machine learning. Background Along with the acceleration of the urban process, the municipal road is used as a basic skeleton of urban traffic, and the service state of the municipal road is directly related to urban operation efficiency and public travel safety. The development of the road loss trend prediction by using the artificial intelligence technology becomes a key link for constructing an intelligent municipal maintenance system and improving the urban fine management level. Currently, the prior art mainly utilizes regression analysis or deep neural network to predict maintenance time by processing road surface condition indexes and historical maintenance records. The method focuses on the statistical simulation of macroscopic state indexes, and combines historical traffic flow data to perform simple association analysis and trend extrapolation. However, the prior art has the problems of coarse modeling granularity and weak physical logic relevance when processing the dynamic interaction of burst heavy load and long-term attenuation of a structure. The digitization degree of the evolution logic between the random load event and the structural damage is insufficient, so that the reduction degree of the prediction result to the actual loss path of the road is not high, and the accurate maintenance requirement under the variable service environment is difficult to deal with. Therefore, the technical problems that accurate depiction of the dynamic evolution track of the road loss and reliable prediction of the medium-long maintenance requirement are difficult to realize in the prior art. Disclosure of Invention The application aims to provide a machine learning-based municipal road maintenance demand prediction method and system, which are used for solving the technical problems that in the prior art, accurate depiction of a road loss dynamic evolution track and reliable prediction of a medium-long-term maintenance demand are difficult to realize. In a first aspect, the present application provides a machine learning-based method for predicting maintenance requirements of a municipal road, comprising: acquiring service starting time of municipal roads and traffic frequency and full-load proportion of heavy vehicles in a plurality of sampling periods; carrying out differential calculation on the service starting moment and sampling time points in each sampling period to obtain an attenuation coefficient sequence, carrying out time domain pulse transformation on the traffic frequency and the full-load proportion to generate a pulse sequence, and constructing a space-time correlation matrix by carrying out time sequence alignment on the attenuation coefficient sequence and the pulse sequence; taking the attenuation coefficient in the space-time correlation matrix as a dynamic adjustment factor, and carrying out pulse coding on the pulse sequence by utilizing pulse coding modulation to obtain a bit stream sequence; Mapping the bit stream sequence as an observation sequence into a preset road raster pattern, calculating the metric value of a transition branch between different road states in the road raster pattern through a maximum likelihood criterion, and determining an optimal transition path through optimal path planning according to the metric value, wherein the road states comprise a health state, a sub-health state, a damage state and a maintenance state; Determining a grid step length of transferring the road state at the current moment to the maintenance state from the optimal transfer path, determining the residual time of the municipal road entering the maintenance period based on the grid step length, and determining maintenance requirement prediction information comprising maintenance priority and predicted maintenance date based on the residual time and preset road weight. Optionally, differential calculation is performed on the service starting time and sampling time points in each sampling period to obtain an attenuation coefficient sequence, time domain pulse transformation is performed on the traffic frequency and the full-load proportion to generate a pulse sequence, and time-space correlation matrix is constructed by aligning the attenuation coefficient sequence with the pulse sequence in time sequence, including: Calculating a duration interval between a service starting moment and a sampling time point in each sampling period to obtain a service step sequence, and carrying out nonlinear mapping on the service step sequence by using a preset attenuation function to obtain an attenuation coefficient sequence; Determining signal amplitude based on full load proportion, determining signal density based on traffic frequency, cal