Machine learning based municipal road maintenance need prediction method and system

Abstract

The application provides a municipal road maintenance demand prediction method and system based on machine learning, relates to the technical field of machine learning, and obtains the service starting time of a municipal road and the passing frequency and full load ratio of heavy vehicles in multiple sampling periods; the service starting time and the sampling time points in each sampling period are differentially calculated to construct a space-time correlation matrix; the attenuation coefficient in the space-time correlation matrix is taken as a dynamic adjustment factor to obtain a bit stream sequence; the bit stream sequence is mapped to a preset road grid map as an observation sequence to determine an optimal transfer path through optimal path planning; the grid step length of the road state at the current time to the maintenance state is determined from the optimal transfer path to determine maintenance demand prediction information including a maintenance priority and a predicted maintenance date. The accuracy, objectivity and forward-looking nature of municipal road maintenance demand prediction are achieved.

Description

Technical Field

[0001] This application relates to the field of machine learning technology, and in particular to a machine learning-based method and system for predicting municipal road maintenance needs. Background Technology

[0002] With the acceleration of urbanization, municipal roads, as the basic framework of urban transportation, directly affect urban operational efficiency and public travel safety. Utilizing artificial intelligence technology to predict road wear trends has become a key link in building a smart municipal maintenance system and improving the level of refined urban management.

[0003] Currently, existing technologies mainly utilize regression analysis or deep neural networks to predict maintenance timing by processing pavement condition indices and historical maintenance records. These methods primarily focus on statistical simulation of macroscopic condition indicators and combine them with historical traffic flow data for simple correlation analysis and trend extrapolation.

[0004] However, existing technologies suffer from coarse modeling granularity and weak physical-logical correlation when dealing with the dynamic interaction between sudden heavy loads and long-term structural degradation. This insufficient digitization of the evolutionary logic between random load events and structural damage results in low accuracy of predictions reflecting the actual road wear path, making it difficult to meet the precise maintenance needs under varying service environments. Therefore, existing technologies face the technical challenge of accurately depicting the dynamic evolution trajectory of road wear and reliably predicting medium- and long-term maintenance requirements. Summary of the Invention

[0005] The purpose of this application is to provide a method and system for predicting municipal road maintenance needs based on machine learning, in order to solve the technical problem that it is difficult to accurately characterize the dynamic evolution trajectory of road wear and tear and reliably predict medium- and long-term maintenance needs in the existing technology.

[0006] Firstly, this application provides a machine learning-based method for predicting municipal road maintenance needs, including:

[0007] Obtain the service start time of municipal roads and the frequency of heavy-duty vehicle traffic and the proportion of full load in multiple sampling periods;

[0008] Differential calculations are performed on the service start time and sampling time points within each sampling period to obtain the attenuation coefficient sequence. Time-domain pulse transformation is performed on the traffic frequency and full load ratio to generate a pulse sequence. By aligning the attenuation coefficient sequence with the pulse sequence in time, a spatiotemporal correlation matrix is ​​constructed.

[0009] Using the attenuation coefficient in the spatiotemporal correlation matrix as a dynamic adjustment factor, pulse code modulation is used to pulse code the pulse sequence to obtain a bit stream sequence.

[0010] The bitstream sequence is mapped as an observation sequence to a preset road grid map. The metric value of the transition branch between different road states in the road grid map is calculated by the maximum likelihood criterion. Based on the metric value, the optimal transition path is determined by optimal path planning. The road states include healthy state, sub-healthy state, damaged state and maintenance state.

[0011] The grid step size for transitioning the road status to the maintenance status at the current moment is determined from the optimal transition path. Based on the grid step size, the remaining time for the municipal road to enter the maintenance cycle is determined. Based on the remaining time and the preset road weight, maintenance demand prediction information including maintenance priority and expected maintenance date is determined.

[0012] Optionally, differential calculations are performed on the service start time and sampling time points within each sampling period to obtain an attenuation coefficient sequence. A time-domain pulse transformation is then performed on the traffic frequency and full-load ratio to generate a pulse sequence. By aligning the attenuation coefficient sequence with the pulse sequence in time, a spatiotemporal correlation matrix is ​​constructed, including:

[0013] The service start time and the time interval between sampling time points in each sampling period are calculated to obtain the service step sequence. The service step sequence is then nonlinearly mapped using a preset attenuation function to obtain the attenuation coefficient sequence.

[0014] The signal amplitude is determined based on the full load ratio, and the signal density is determined based on the passage frequency. The interval time of the pulse distribution is calculated using the signal density, and multiple trigger times are determined within the sampling period according to the interval time. The signal amplitude is used as the pulse height to generate the corresponding pulse signal at each trigger time, thus obtaining the pulse sequence.

[0015] Based on the chronological order of sampling time points within the sampling period, a spatiotemporal correlation matrix is ​​constructed by mapping the position using the attenuation coefficient in the attenuation coefficient sequence as the vertical component and the pulse signal in the pulse sequence as the horizontal component.

[0016] Optionally, using the attenuation coefficient in the spatiotemporal correlation matrix as a dynamic adjustment factor, pulse code modulation is used to pulse code the pulse sequence to obtain a bitstream sequence, including:

[0017] The attenuation coefficient in the spatiotemporal correlation matrix is ​​used as a dynamic adjustment factor, and the dynamic adjustment factor is matched with the preset quantization mapping rule to obtain multiple quantization steps.

[0018] Based on the quantization step, the amplitude range corresponding to the pulse sequence is divided into multiple quantization intervals, and the pulse height of each pulse signal in the pulse sequence is assigned to the corresponding quantization interval to determine the quantization level corresponding to each pulse signal.

[0019] Each quantization level is converted into a corresponding binary symbol using preset encoding instructions to obtain a bitstream sequence.

[0020] Optionally, the bitstream sequence is mapped as an observation sequence to a preset road grid map, and the metric value of the transition branch between different road states in the road grid map is calculated using the maximum likelihood criterion, including:

[0021] Based on the road state evolution order, the road state nodes between adjacent sampling time points in the road grid are unidirectionally connected to obtain multiple transition branches. Each sampling time point in the road grid includes four road state nodes corresponding to the healthy state, sub-healthy state, damaged state, and maintenance state.

[0022] Based on the order of sampling time points, the transfer branches corresponding to each sampling time point are arranged to obtain multiple candidate transfer paths;

[0023] Based on the sampling period of each transfer branch, the corresponding observation segment is extracted from the bit stream sequence, and based on the preset state feature code library, the similarity of each observation segment is compared with each transfer branch in the corresponding sampling period to obtain the branch matching probability of each transfer branch.

[0024] Based on the branch matching probability corresponding to each transition branch, the log-likelihood probability of each transition branch under the constraint of the observation sequence is calculated, and the log-likelihood probability is determined as the metric.

[0025] Optionally, based on the branch matching probability corresponding to each transition branch, the log-likelihood probability of each transition branch under the observation sequence constraint is calculated, and the log-likelihood probability is determined as a metric, including:

[0026] Determine the starting node in the road grid map where each transfer branch is connected;

[0027] In the first sampling period, the preset initial score is determined as the path score, and the path score is summed with the branch matching probability corresponding to each transition branch led out from the starting node to obtain the log likelihood probability of each transition branch.

[0028] In subsequent sampling periods, the path score is determined based on the maximum log-likelihood probability among all transition branches pointing to the starting node at the sampling time point corresponding to the previous sampling period. The path score is then summed with the branch matching probability corresponding to each transition branch derived from the starting node to obtain the log-likelihood probability of each transition branch.

[0029] The log-likelihood probability is determined as a measure of the transition branch.

[0030] Optionally, based on the metric, the optimal transfer path is determined through optimal path planning, including:

[0031] Determine the termination sampling time point of the observation sequence, and select the maximum metric value from the metric values ​​of the transition branch corresponding to the termination sampling time point;

[0032] Using the transition branch corresponding to the maximum metric value as the backtracking starting point, the state path is backtracked in the reverse order of the sampling time points in the road grid map to obtain the corresponding node sequence, and the node sequence is determined as the optimal transition path.

[0033] Optionally, the grid step size for transitioning the road state to the maintenance state at the current moment is determined from the optimal transition path. Based on the grid step size, the remaining time for the municipal road to enter the maintenance cycle is determined. Based on the remaining time and preset road weights, maintenance demand forecast information, including maintenance priority and expected maintenance date, is determined, including:

[0034] Determine the node position at the current moment and the endpoint position corresponding to the evolution to the maintenance state in the optimal transition path;

[0035] The number of sampling points between the node location and the endpoint location is used as the grid step size. The product of the grid step size and the preset unit sampling time is calculated to obtain the remaining time for the municipal road to enter the maintenance cycle.

[0036] Based on a preset priority determination matrix, the emergency level corresponding to the remaining time is cross-matched with the importance level corresponding to the road weight to determine the maintenance priority. The remaining time and the current time are then added together to obtain the expected maintenance date, thus determining the maintenance demand prediction information including the maintenance priority and the expected maintenance date.

[0037] Secondly, this application provides a machine learning-based system for predicting municipal road maintenance needs, including:

[0038] The acquisition module is used to acquire the service start time of municipal roads and the frequency of heavy-duty vehicles and the proportion of full load in multiple sampling periods;

[0039] The calculation module is used to perform differential calculations on the service start time and sampling time points within each sampling period to obtain the attenuation coefficient sequence, and to perform time-domain pulse transformation on the passage frequency and full load ratio to generate a pulse sequence. By aligning the attenuation coefficient sequence with the pulse sequence in time, a spatiotemporal correlation matrix is ​​constructed.

[0040] The encoding module is used to pulse-encode the pulse sequence using pulse code modulation with the attenuation coefficient in the spatiotemporal correlation matrix as a dynamic adjustment factor, so as to obtain a bit stream sequence.

[0041] The mapping module is used to map the bit stream sequence as an observation sequence to a preset road grid map. It calculates the metric value of the transition branch between different road states in the road grid map through the maximum likelihood criterion. Based on the metric value, it determines the optimal transition path through optimal path planning. The road states include healthy state, sub-healthy state, damaged state and maintenance state.

[0042] The determination module is used to determine the grid step size for the road state to transition to the maintenance state from the optimal transfer path at the current moment, determine the remaining time for the municipal road to enter the maintenance cycle based on the grid step size, and determine maintenance demand prediction information including maintenance priority and expected maintenance date based on the remaining time and preset road weights.

[0043] Thirdly, this application provides an electronic device, comprising:

[0044] Memory, used to store computer programs;

[0045] A processor is used to execute computer programs to implement the steps of the machine learning-based municipal road maintenance demand prediction method as described in the first aspect above.

[0046] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the machine learning-based municipal road maintenance demand prediction method described in the first aspect above.

[0047] The machine learning-based method for predicting municipal road maintenance needs provided in this application obtains the service start time, traffic frequency, and full load ratio of municipal roads, ensuring that the input sources of the prediction model have physical reality significance. It achieves deep fusion of internal damage baselines and external impact variables, laying a data structure foundation for accurately reconstructing the actual road wear path. It enhances the noise resistance and parsing efficiency of feature information, overcoming the limitations of coarse-grained modeling in traditional methods. It solves the problem of unreliable prediction of medium- and long-term maintenance needs, enabling refined and forward-looking control of maintenance requirements.

[0048] Furthermore, this application establishes multiple quantization steps; divides the amplitude range of the pulse sequence into multiple quantization intervals based on the quantization steps, and assigns the pulse height of each pulse signal to the corresponding quantization interval to lock the quantization level; and uses preset encoding instructions to convert the quantization level into binary code elements, thereby generating a complete bit stream sequence. This improves the accuracy of digital features in depicting the evolution trajectory of road damage, providing highly accurate digital evidence to meet the precise maintenance needs under varying service environments. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 A flowchart illustrating the machine learning-based municipal road maintenance demand prediction method provided in this application embodiment;

[0051] Figure 2 A flowchart illustrating a method for obtaining a bitstream sequence provided in an embodiment of this application;

[0052] Figure 3 A flowchart illustrating the method for determining a metric value provided in an embodiment of this application;

[0053] Figure 4 A schematic diagram of the structure of a machine learning-based municipal road maintenance demand prediction system provided in an embodiment of this application;

[0054] Figure 5 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0055] To address the shortcomings of existing technologies in municipal road maintenance prediction, which focus on macroscopic statistical analysis and struggle to characterize the dynamic interaction between sudden heavy loads and long-term structural attenuation, resulting in coarse modeling granularity, this application characterizes the inherent, continuous weakening of the structure using an attenuation coefficient sequence determined at the start of service. This sequence is then time-aligned with a pulse sequence reflecting instantaneous load impacts to construct a spatiotemporal correlation matrix. Secondly, pulse code modulation (PCM) technology is used to dynamically adjust the quantization step using the attenuation coefficient, solidifying the complex physical damage mechanism into a highly digitized bitstream sequence, thus solving the problem of weak physical-logical correlation. Finally, drawing on the decoding concept of the maximum likelihood criterion, this sequence is used as an observation feature to perform path search in a road grid map containing four road state nodes. Optimal path planning is then used to reconstruct the optimal transfer path for actual road wear. This represents a technological leap from macroscopic statistical prediction to highly accurate depiction of dynamic evolution trajectories.

[0056] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] The core of this application is to provide a machine learning-based method for predicting municipal road maintenance needs, and a flowchart of one specific implementation is shown below. Figure 1 As shown, the method includes:

[0058] Step 101: Obtain the service start time of municipal roads and the frequency of heavy-duty vehicles and the proportion of full load in multiple sampling periods.

[0059] In this step, municipal roads refer to urban public transport road infrastructure that requires maintenance demand forecasting. The service start time refers to the initial point in time after the completion of construction and handover of the municipal road, or after the most recent overall overhaul, which can be represented as... The sampling period refers to a fixed time unit used to aggregate traffic data and calculate structural attenuation coefficients, which can be 1 day or 1 week, etc. Heavy-duty vehicles refer to large freight vehicles or special-purpose vehicles that exert load impact on the road structure. Traffic frequency refers to the total number of times heavy-duty vehicles pass through the target road segment within a single sampling period. Full load ratio refers to the proportion of heavy-duty vehicles passing through the target road segment whose actual load reaches the approved load capacity, or the ratio of the average load of heavy-duty vehicles to the standard load.

[0060] In this embodiment of the application, the service start time of the municipal road is first obtained by consulting the road asset management records. Simultaneously, raw traffic information is acquired using dynamic weighing equipment, traffic devices, or sensor networks deployed at key nodes of municipal roads. Then, the traffic information is divided into multiple consecutive sampling periods using a preset time window. , The frequency of passage is obtained by cumulatively counting the heavy-load vehicles identified in each sampling period. Then, the percentage of vehicles fully loaded in each sampling period was calculated by comparing the dynamic weighing data with the vehicle's rated load capacity. Ultimately, it resulted in a combination of multiple and The original dataset that constitutes this.

[0061] For example, taking Road B in Area A as an example, we can retrieve archival data to determine its service start time. The sampling period is set to May 1, 2022. The frequency of passage in the first sampling period was obtained by using intelligent equipment at the entrance of Road B over a 24-hour period. The calculation was performed for 300 cycles, and the corresponding full-load ratio was obtained by combining data from a dynamic weighing platform buried underground. It is 0.75.

[0062] Step 102: Perform differential calculations on the service start time and sampling time points within each sampling period to obtain the attenuation coefficient sequence, and perform time-domain pulse transformation on the traffic frequency and full load ratio to generate a pulse sequence. By aligning the attenuation coefficient sequence with the pulse sequence in time, construct a spatiotemporal correlation matrix.

[0063] In this step, the sampling time point refers to the start or end time coordinate corresponding to each sampling period. The attenuation coefficient sequence refers to a set of values ​​that describe the nonlinear degradation of road strength with increasing service life, and can be expressed as: A pulse sequence refers to the conversion of the original traffic load into a discrete pulse dataset with amplitude and density characteristics, which can be represented as... The spatiotemporal correlation matrix is ​​a two-dimensional data structure obtained by mapping and matching the longitudinal component reflecting structural performance degradation with the transverse component reflecting external load impact along the time dimension. It can be represented as follows: .

[0064] Step 201: Calculate the time interval between the service start time and the sampling time point in each sampling period to obtain the service step length sequence, and use the preset attenuation function to perform nonlinear mapping on the service step length sequence to obtain the attenuation coefficient sequence.

[0065] In this step, the service step sequence refers to a multi-dimensional vector constructed based on the time difference between the sampling time point and the service start time, represented as follows: The preset structural attenuation function refers to a pre-established mathematical mapping model used to simulate the degradation logic of pavement physical properties over time. The attenuation coefficient sequence refers to a set of values ​​reflecting the change in road structural strength over time, obtained by processing the service step sequence using the attenuation function; it is represented as... .

[0066] In this embodiment of the application, the service start time is first extracted. With multiple sampling time points currently being processed as well as The absolute time interval between the two is calculated through subtraction to obtain the service step sequence. Taking road B in region A as an example, the service start time... If the first sampling time point is May 1, 2022... If the date is May 2, 2022, then the first infantry commander will be in service. Given 1 day, the service step length sequence vector, composed of multiple step length values, is represented as follows: .

[0067] Next, a nonlinear mapping is performed on the service step sequence using a preset structural decay function. The specific mathematical expression is as follows: ,in Indicates the first The attenuation coefficient corresponding to each sampling time point Represents the natural constant. This indicates the preset attenuation factor related parameters. Indicates the first step in the service step sequence Each element. Taking 0.0001 as an example, the attenuation coefficient at each time step is calculated. The final integrated output attenuation coefficient sequence is expressed as: .

[0068] Step 202: Determine the signal amplitude based on the full load ratio and the signal density based on the passage frequency. Calculate the pulse distribution interval using the signal density and determine multiple trigger times within the sampling period based on the interval time. Use the signal amplitude as the pulse height to generate the corresponding pulse signal at each trigger time to obtain the pulse sequence.

[0069] In this step, the signal amplitude refers to the intensity value of the pulse signal in the vertical direction, expressed as: Signal density refers to the frequency of pulse signal occurrences within a unit sampling period. The pulse distribution interval refers to the time span between two adjacent pulse signals, expressed as... The preset distribution pattern refers to the mathematical probability distribution model used to arrange multiple trigger moments within the sampling period, which can be a Poisson distribution or a uniform distribution. The trigger moment refers to the time position at which a specific pulse signal is generated within the sampling period. A pulse signal is a discrete waveform with a specific amplitude and duration. A pulse sequence is a set of pulse signals arranged in chronological order within the sampling period, represented as... .

[0070] In this embodiment of the application, firstly, based on the obtained full load ratio Determine signal amplitude At the same time, based on the frequency of passage Determine the signal density using the sampling period. Total duration divided by frequency of passage Calculate the pulse distribution interval: ,by 24 hours and Taking 240 times as an example, the interval time is calculated. The interval is 6 minutes. Then, based on this interval and the preset distribution pattern, the trigger times are determined within the sampling period. Finally, the signal amplitude is... The pulse height is used to generate a corresponding pulse signal at each trigger moment. (Based on full load ratio) Taking 0.85 as an example, the corresponding signal amplitude is... The value is 0.85, and the final set of pulse signals generated on the time axis is represented as a pulse sequence. .

[0071] Step 203: According to the chronological order of sampling time points within the sampling period, position mapping is performed using the attenuation coefficient in the attenuation coefficient sequence as the vertical component and the pulse signal in the pulse sequence as the horizontal component to construct a spatiotemporal correlation matrix.

[0072] In this step, the longitudinal component refers to the structural performance parameters that serve as column references or row guides in the matrix. The lateral component refers to the traffic load parameters distributed horizontally in the matrix. Dimensional combination refers to the process of converting multiple sets of one-dimensional sequences into a two-dimensional matrix structure through coordinate correlation. The spatiotemporal correlation matrix is ​​a two-dimensional matrix structure obtained by spatially fusing attenuation information reflecting the inherent strength of the road with pulse information reflecting external impacts, represented as... .

[0073] In this embodiment, the chronological order of sampling time points within each sampling period is first retrieved as an index. At each sampling time point, the attenuation coefficient sequence is... attenuation coefficient in As the longitudinal component, the pulse sequence corresponding to this moment is also used. The pulse signal in the sample is used as the transverse component. Due to the passage frequency within different sampling periods... The differences may lead to inconsistent pulse subsequence lengths. To construct a regular two-dimensional matrix, this application sets a preset maximum passage frequency. For sampling periods where the actual passage frequency is less than the maximum passage frequency, zero-padding is performed at the end of the horizontal component to ensure the spatiotemporal correlation matrix. Each row has the same dimensions. The value is 0.9 and the corresponding pulse subsequence is and For example, the constructed two-dimensional data matrix structure is represented as follows:

[0074] ;

[0075] The first column represents the longitudinal structural strength constraints, while the remaining columns represent the transverse time-series load intensities. This method aligns structural attenuation with traffic pulses in a temporal sequence, ultimately constructing a spatiotemporal correlation matrix. .

[0076] Step 103: Using the attenuation coefficient in the spatiotemporal correlation matrix as a dynamic adjustment factor, pulse code modulation is used to pulse code the pulse sequence to obtain the bit stream sequence.

[0077] In this step, the attenuation coefficient refers to the structural performance index at a specific moment in the spatiotemporal correlation matrix, which can be expressed as: The dynamic adjustment factor refers to the adjustment variable that corrects the quantization accuracy or coding rules based on the real-time attenuation level of the road. It can be expressed as... Pulse code modulation (PCM) is a process that converts the continuous amplitude of a discrete pulse signal into a finite-bit binary digital sequence. A bitstream sequence is a digital sequence composed of multiple binary bits that reflects the characteristics of road surface loss; it can be represented as... .

[0078] like Figure 2 As shown, Figure 2 This is a flowchart illustrating a method for obtaining a bitstream sequence provided in an embodiment of this application.

[0079] Step 301: Use the attenuation coefficient in the spatiotemporal correlation matrix as a dynamic adjustment factor, and match the dynamic adjustment factor with the preset quantization mapping rule to obtain multiple quantization steps.

[0080] In this step, the preset quantization mapping rule refers to the corresponding logic that nonlinearly matches the physical attenuation degree of the road structure with the signal quantization accuracy. The quantization step refers to the smallest numerical interval in the pulse amplitude space used to divide different levels, expressed as... .

[0081] In this embodiment of the application, the spatiotemporal correlation matrix is ​​first extracted. Attenuation coefficient in the first column Determined as a dynamic regulating factor Taking area A, road B as an example, to obtain the... Attenuation coefficient corresponding to each sampling time The value is 0.8. Next, the quantization step is calculated using a preset quantization mapping rule. The specific calculation formula is shown in formula (1):

[0082]

[0083] in This indicates the preset initial quantization step size. This indicates the preset correction factor. This represents the attenuation coefficient at the current sampling time. The sequence vector composed of multiple quantization steps for different structural states is obtained in this way as follows: .

[0084] Step 302: Based on the quantization step, divide the amplitude range corresponding to the pulse sequence into multiple quantization intervals, and assign the pulse height of each pulse signal in the pulse sequence to the corresponding quantization interval to determine the quantization level corresponding to each pulse signal.

[0085] In this step, the amplitude range refers to the pulse height. The possible numerical range can be the set of real numbers between 0 and 1. A quantization interval refers to multiple continuous sub-regions obtained by dividing the amplitude range based on the quantization step. A quantization grade refers to the integer index value assigned to the pulse height after it falls within a specific quantization interval, denoted as... .

[0086] In the embodiments of this application, firstly based on the obtained quantization step The preset amplitude range is divided into equal intervals to obtain multiple quantization intervals. Then, the pulse sequence is extracted. The pulse height corresponding to each pulse signal The height of each pulse is recorded separately. It is assigned to the corresponding quantization interval. Based on the height of the first pulse signal... Taking 0.75 as an example, if the current quantization step... The value is 0.05. By performing a rounding operation, the quantization interval index to which the pulse signal belongs is determined, thereby determining the corresponding quantization level. The value is 15. The final result is a quantization hierarchy sequence composed of multiple quantization values, represented as follows: .

[0087] Step 303: Use preset encoding instructions to convert each quantization level into the corresponding binary code to obtain a bit stream sequence.

[0088] In this step, the preset encoding instruction refers to the logical control instruction that specifies the conversion relationship between quantization levels and digital bit sequences. A binary symbol is the smallest unit of digital information composed of 0s and 1s.

[0089] In this embodiment of the application, the quantized hierarchical sequence is first quantized using preset encoding instructions. Each element in the code is converted into its corresponding binary code. (Quantization classification) Taking 15 as an example, an 8-bit binary encoding conversion is performed according to the preset encoding instructions, resulting in a binary code sequence of 00001111. Then, the converted binary code elements are concatenated according to the triggering time sequence of each pulse signal. The final output bitstream sequence reflecting the road damage characteristics is represented as follows: For example, to obtain the first The bitstream segment vector corresponding to each sampling period is represented as follows: .

[0090] Step 104: Map the bit stream sequence as an observation sequence to a preset road grid map, calculate the metric value of the transition branch between different road states in the road grid map using the maximum likelihood criterion, and determine the optimal transition path based on the metric value through optimal path planning. The road states include healthy state, sub-healthy state, damaged state, and maintenance state.

[0091] In this step, the observation sequence refers to the data bit stream that serves as the logical input. The preset road grid map refers to the spatiotemporal topology map built on the sampling time axis, including health status. Sub-health status Damage status and maintenance status Four road state nodes. The maximum likelihood criterion is a rule for determining the state path with the highest probability of occurrence given a known observation sequence. A transition branch is a unidirectional path connecting different state nodes between two adjacent sampling time points. The metric is the log-likelihood score of the degree of matching between the transition branch and the observation segment, which can be expressed as... Optimal path planning refers to the process of determining the evolutionary trajectory with the highest total score through a global optimization algorithm. The optimal transition path refers to the sequence of state nodes that reflects the true evolution of the road, obtained through backtracking.

[0092] Step 401: Based on the road state evolution order, the road state nodes between adjacent sampling time points in the road grid map are unidirectionally connected to obtain multiple transition branches. Each sampling time point in the road grid map includes four road state nodes corresponding to the healthy state, sub-healthy state, damaged state, and maintenance state.

[0093] In this step, the road condition evolution sequence refers to the logical direction in which the health condition of municipal roads gradually deteriorates with service time and load intensity. A one-way connection refers to an irreversible state-direction relationship established between adjacent sampling time points, conforming to the laws of physical evolution.

[0094] In this embodiment of the application, firstly, based on the road state evolution sequence, each sampling time point is represented in the road grid map. Configure the corresponding health status respectively Sub-health status Damage status and maintenance status The four road state nodes are then used. Next, following the time axis, the road state nodes between adjacent sampling time points are connected unidirectionally, resulting in multiple transition branches. Taking road B in region A as an example, at the sampling time point... Health status node At this location, a pointer can be established. Health status node The self-evolutionary branch, and the direction to Sub-health state nodes The degraded branch. By traversing all state nodes, a state transition topology is constructed.

[0095] Step 402: Based on the order of sampling time points, arrange the transfer branches corresponding to each sampling time point to obtain multiple candidate transfer paths.

[0096] In this step, a candidate transition path refers to a complete evolutionary sequence that a road may undergo from its initial service state to its current state, consisting of multiple consecutive transition branches connected end-to-end. This is represented as... .

[0097] In this embodiment, the transition branches corresponding to each sampling time point are first arranged and combined based on the chronological order of the sampling time points. Path search technology is then used to traverse all feasible connected paths from the initial time node to the final time node in the road grid, thereby obtaining multiple candidate transition paths. Taking road B as an example, the path vector composed of multiple state nodes can be represented as follows: The first subscript letter indicates the state type, and the second subscript letter indicates the sampling time index.

[0098] Step 403: Based on the sampling period to which each transition branch belongs, extract the corresponding observation segment from the bit stream sequence, and based on the preset state feature code library, compare the similarity of each observation segment with each transition branch in the corresponding sampling period to obtain the branch matching probability of each transition branch.

[0099] In this step, the observation segment refers to the binary subsequence corresponding to a specific sampling period obtained after segmenting the bitstream sequence, denoted as: The pre-stored state feature code library refers to a set of binary code streams pre-stored for matching standard features of different road state transition types. Similarity comparison refers to the processing method for calculating the degree of overlap between observed segments and standard feature codes. Branch matching probability refers to the degree of agreement between the current transition branch and the actual observed data at the feature level, expressed as... .

[0100] In this embodiment of the application, firstly, based on the sampling period to which each transfer branch belongs, the bitstream sequence is... Extract the corresponding observation segment from the middle Next, based on a pre-defined state feature code library, standard feature codes corresponding to the current transition type are extracted. Then, Hamming distance or cross-correlation algorithms are used to analyze each observation segment. Similarity comparisons were performed with each transition branch within the corresponding sampling period. (Based on the observed segment) For example, if the corresponding transition code in the state feature code library is The similarity calculation result is 0.75, thus determining the branch matching probability of this transition branch. It is 0.75.

[0101] Step 404: Based on the branch matching probability corresponding to each transition branch, calculate the log-likelihood probability of each transition branch under the constraints of the observation sequence, and determine the log-likelihood probability as a metric.

[0102] In this step, the log-likelihood probability refers to the numerical index used for accumulation operations obtained by performing a logarithmic transformation on the probability values, denoted as: .

[0103] like Figure 3 As shown, Figure 3 This is a flowchart illustrating the method for determining a metric value provided in an embodiment of this application.

[0104] Step 4041: Determine the starting node that each transfer branch connects to in the road grid map.

[0105] In this step, the starting node refers to the time-point state node in the road grid diagram that serves as the logical starting point for each transition branch, and can be represented as: .

[0106] In this embodiment of the application, the sampling time points in the road grid map are first traversed. All road state nodes. Then, based on the unidirectional connection relationship determined by the road state evolution order, the starting node connected to each transition branch in the road grid diagram is determined. Taking Road B in Region A as an example, if the current analysis is for the sampling time point... arrive The evolution between them is located in The health status node at any given time is the starting node of multiple transition branches within that sampling period. The set vector of starting nodes can be represented as... .in Nodes representing health status Represents a sub-healthy state. Represents a node in a damaged state. This represents a node that maintains the state.

[0107] Step 4042: In the first sampling period, the preset initial score is determined as the path score, and the path score is summed with the branch matching probability corresponding to each transition branch led out from the starting node to obtain the log-likelihood probability of each transition branch.

[0108] In this step, the first sampling period refers to the period immediately preceding the start of service. The first sampling interval after that is denoted as The preset initial score refers to the logarithmic score of the state probability pre-set based on the road surface condition at the time of delivery of the municipal road, which can be expressed as: The path score refers to the cumulative probability score up to the current node position, which can be expressed as: .

[0109] In this embodiment, it is first determined whether the current period is the first sampling period. If so, then the preset initial score will be applied. The path score is directly determined as the path score for each state node. Taking road B as an example, if the road surface condition is excellent at the time of delivery, the preset initial score vector is represented as follows: .in This represents the logarithmic probability of being in a healthy state. A preset maximum constant, such as 100, indicates an extremely low probability of being in an unhealthy state. Then, the path scores are... With the starting node The log-likelihood probability of each transition branch is obtained by summing the branch matching probabilities corresponding to each transition branch. Taking a self-transition branch derived from a healthy state node as an example, if the corresponding branch matching probability is... in If the summation is 0.1, then it is calculated by summation. The log-likelihood probability of obtaining this branch is -0.1.

[0110] Step 4043: In subsequent sampling periods, determine the path score based on the maximum log-likelihood probability among all transition branches pointing to the starting node at the sampling time point corresponding to the previous sampling period. Sum the path score with the branch matching probability corresponding to each transition branch derived from the starting node to obtain the log-likelihood probability of each transition branch.

[0111] In this step, the subsequent sampling period refers to any sampling interval that follows the first sampling period in the time series, denoted as... The maximum log-likelihood probability refers to the probability of finding the same starting node across all nodes. The log-likelihood score that is the largest among the transition branches.

[0112] In this embodiment of the application, for those in subsequent sampling periods The transition branch first obtains the sampling time point corresponding to the previous sampling period. Point to the starting node All transition branches. Then, the maximum log-likelihood probability is determined from the metric values ​​of these branches using a maximum value screening algorithm, and this probability is used as the current starting node. Path score .by Taking a sub-healthy state node at time t as an example, if the set vector of the log-likelihood probabilities of all predecessor transition branches pointing to that node is represented as... The maximum value is selected as the path score for the current node. Then, the path is scored separately. With the starting node The branch matching probabilities corresponding to each derived transition branch are summed. The specific iterative calculation formula is as follows: ,in Indicates the current sampling time point The log-likelihood probability corresponding to the transition branch This represents the branch matching probability in the current sampling period.

[0113] Based on path score for in The probability is 0.5, and the current branch matches. for in Taking 0.2 as an example, the log-likelihood probability of this transition branch is finally obtained as -0.7 through summation. By repeatedly determining the starting node and accumulating scores for each period, a sequence of metrics composed of multiple log-likelihood probabilities is finally generated, represented as follows: .

[0114] Step 411: Determine the termination sampling time point of the observation sequence, and select the maximum metric value from the metric values ​​of the transition branch corresponding to the termination sampling time point.

[0115] In this step, the termination sampling time point refers to the coordinate of the last sampling time corresponding to the observation sequence. The maximum metric value is the value with the highest cumulative log-likelihood score among all feasible paths at the termination sampling time point.

[0116] In this embodiment, the termination sampling time point of the observation sequence is first determined. Then, the accumulated metrics are extracted from the transition branches derived from all road state nodes corresponding to this termination sampling time point. The maximum value is selected from all metrics corresponding to the termination sampling time point using a maximum value filtering algorithm. Taking road B as an example, if at the termination time... There exist multiple sets of metrics corresponding to different states. The one with the largest value is selected as the endpoint criterion for the globally optimal path.

[0117] Step 412: Using the transition branch corresponding to the maximum metric value as the backtracking starting point, backtrack the state path in the reverse order of the sampling time points in the road grid map to obtain the corresponding node sequence, and determine the node sequence as the optimal transition path.

[0118] In this step, the backtracking starting point refers to the endpoint position corresponding to the maximum metric value in the road grid map. The node sequence refers to a set of state nodes obtained by tracing back in reverse chronological order. The optimal transition path refers to the evolution trajectory determined through backtracking that best matches the actual physical wear and tear of the road.

[0119] In this embodiment, the transition branch corresponding to the maximum metric value is first used as the backtracking starting point. Then, state path backtracking is performed in the road grid map in reverse order of the sampling time points. Specifically, by finding the optimal predecessor node pointer for each node, the process backtracks step-by-step to the service start time. Taking road B as an example, the node sequence generated during the backtracking process can be represented as a vector. Finally, this node sequence was determined as the optimal transfer path.

[0120] In this embodiment, monitoring of road B in region A is taken as an example. First, a road grid map containing... arrive A node matrix with four states is used, and unidirectional connections are established based on the state evolution order. A set of candidate transition paths is obtained through permutations and combinations. Next, the bitstream sequence from March 2026. The first week's observation segment was extracted from the data. It then compares the similarity of the data with the standard features in the preset state feature code library to calculate the branch matching probability of transitioning from a healthy state to a healthy state. It is 0.9, and then the formula is used. Get the measurement value It is -0.105.

[0121] Based on this, the system accumulates the metrics over the entire timeline. Until the final sampling point of March 25, 2026, the system selects the maximum metric with the highest accumulated score from all state nodes reaching that point; the corresponding state is then designated as a sub-healthy state. Starting from this node, we trace back in reverse order along the transition branches in the raster graph to the initial service time, resulting in a node sequence as follows: This sequence is determined as the optimal transfer path, thereby accurately identifying that the current true health level of road B is in a sub-healthy state. This provides a core path benchmark for determining subsequent grid step sizes and predicting maintenance dates.

[0122] It should be noted that the aforementioned optimal transition path reflects the actual state evolution trajectory of the municipal road from the start of its service life to the current observation time. To predict future maintenance needs, this application, after determining the node position at the current time, extrapolates and predicts based on the state degradation trend reflected in the optimal transition path and a preset structural decay function. Specifically, by simulating the state transition under predicted load conditions in each future sampling period, a predicted path is continuously generated in the road grid map until the predicted path reaches the maintenance state. The nodes are used to determine the endpoint position corresponding to the evolution to the maintenance state.

[0123] Step 105: Determine the grid step size for the road status to transition to the maintenance status from the optimal transfer path at the current moment. Based on the grid step size, determine the remaining time for the municipal road to enter the maintenance cycle. Based on the remaining time and the preset road weight, determine the maintenance demand prediction information, including maintenance priority and expected maintenance date.

[0124] In this step, the current time refers to the specific time coordinate for performing the maintenance prediction. The grid step size refers to the number of sampling points traversed from the current position to the end position of the maintenance state on the optimal transfer path, which can be expressed as... The maintenance cycle refers to the time interval during which a road can maintain safe operation until maintenance work is necessary. Remaining time refers to the predicted actual physical time remaining before the road enters maintenance mode, which can be expressed as... The preset road weight refers to an evaluation coefficient that represents the administrative level or traffic importance of a road, and can be expressed as: Maintenance priority refers to the hierarchical indicators that determine the order in which road maintenance work is performed. Estimated maintenance date refers to the specific year, month, and day of maintenance, calculated by accumulating time. Maintenance demand forecast information refers to comprehensive forecast data that includes both priority and date.

[0125] Step 501: Determine the node position at the current moment and the endpoint position corresponding to the evolution to the maintenance state in the optimal transition path.

[0126] In this step, the node position refers to the index value of the element in the optimal transition path vector that logically corresponds to the current time step. The endpoint position refers to the first state attribute found along the time axis in the optimal transition path that is a maintenance state. The node index coordinates.

[0127] In this embodiment of the application, the current time is first obtained. The optimal transition path sequence vector, previously determined through path backtracking, is then represented as follows: Perform a matching search to determine the current time. The corresponding node location index. Taking area A, road B as an example, if the optimal transfer path vector is... At present Corresponding to the third node The node position is represented as And the value is 3. Then, starting from the node position... Begin searching backwards to determine the first state. The endpoint position index. In the example above, the endpoint position is represented as... And the value is 5. Finally, the logical span from the current state of the road to the maintenance stage was determined through location.

[0128] Step 502: Use the number of sampling points between the node location and the endpoint location as the grid step size, calculate the product of the grid step size and the preset unit sampling time, and obtain the remaining time for the municipal road to enter the maintenance cycle.

[0129] In this step, the preset unit sampling duration refers to the physical time interval between adjacent sampling time points defined when constructing the road raster map, expressed as: .

[0130] In this embodiment of the application, the node position is first extracted. and the finish line Perform a subtraction operation to calculate the grid step size. Taking road B as an example, if the node location... The endpoint is 3. If the value is 5, then the grid step size is calculated. The value is 2. Then, the grid step size is used. With the preset unit sampling time Perform a multiplication operation. The specific calculation formula is as follows: ,in This indicates the remaining time before the generated municipal road enters its maintenance cycle. Indicates the grid step size. This indicates the preset unit sampling duration. If the preset unit sampling duration... The remaining time is calculated by substituting 30 days into the formula. It lasts for 60 days.

[0131] Step 503: Based on the preset priority determination matrix, cross-match the emergency level corresponding to the remaining time with the importance level corresponding to the road weight to determine the maintenance priority, and add the remaining time and the current time to obtain the expected maintenance date, so as to determine the maintenance demand prediction information including the maintenance priority and the expected maintenance date.

[0132] In this step, the preset priority determination matrix refers to a logical decision table that cross-references the urgency of the time dimension with the importance of the road in the spatial dimension, represented as follows: As shown in Table 1:

[0133] Table 1: Priority Decision Matrix

[0134]

[0135] Emergency level refers to the level based on the remaining time. The urgency of maintaining the numerical size partition is expressed as: Importance level refers to the level based on preset road weights. The defined road grade is represented as The estimated maintenance date refers to the specific maintenance date calculated by accumulating time, and is expressed as... Maintenance requirement forecasting information refers to information that aggregates maintenance priorities. With the expected maintenance date The decision reference dataset.

[0136] In this embodiment of the application, the remaining time is first established. and emergency level The mapping relationship between them. Specifically, when the remaining time... When the number of days is less than or equal to 90, the emergency level is... If the value is set to 1, vectors can be used in logical operations. Indicates the remaining time. Emergency level: greater than 90 days and less than or equal to 180 days The value is set to 2. When the remaining time... When it exceeds 180 days, the emergency level The value is determined to be 3.

[0137] Simultaneously establish road weights Importance level The mapping relationship between them. Specifically, when the preset road weights... When the importance level is greater than or equal to 0.8, the importance level is... The value is set to 1. When the road weight... When the importance level is greater than or equal to 0.5 and less than 0.8, the importance level is... The value is set to 2. When the road weight... When it is less than 0.5, the importance level The value is determined to be 3.

[0138] Then, the preset priority decision matrix is ​​used. Perform cross-matching. Taking road B as an example, obtain its preset road weights. Its importance level is determined based on the mapping relationship, which is 0.9. The value is 1. This is because the remaining time was previously calculated. The timeframe is 60 days, which meets the condition of being less than or equal to 90 days; therefore, its emergency level is determined. The value is 1. A search of Table 1 reveals that when the importance level... Level 1 and Emergency When the value is 1, the maintenance priority obtained from cross-matching The value is 1.

[0139] Then utilize the current moment With remaining time Perform addition. The specific formula is: ,in Indicates the expected maintenance date. Indicates the current moment. Indicates the remaining time. If the current time... The remaining 60 days are until March 25, 2026. The estimated maintenance date is calculated by overlaying this date. The final integration and maintenance priority is set for May 24, 2026. and the expected maintenance date It outputs complete maintenance requirement prediction information. This decision matrix... This approach achieves a comprehensive balance between the time urgency and spatial importance of municipal road maintenance needs, ensuring the scientific allocation of maintenance resources.

[0140] This application's embodiments obtain the service start time, traffic frequency, and full load ratio of municipal roads, ensuring that the input source of the prediction model has physical reality significance. It achieves deep fusion of internal damage baselines and external impact variables, laying the data structure foundation for accurately reconstructing the actual road wear path. It enhances the noise resistance and parsing efficiency of feature information, overcoming the limitations of coarse modeling granularity in traditional methods. It solves the problem of unreliable prediction of medium- and long-term maintenance needs, achieving refined and forward-looking control of maintenance requirements.

[0141] Figure 4 A schematic diagram illustrating a specific implementation of the machine learning-based municipal road maintenance demand prediction system provided in this application, with reference to... Figure 4 The system may include:

[0142] The acquisition module 21 is used to acquire the service start time of municipal roads and the frequency of heavy-duty vehicles and the proportion of full load in multiple sampling periods;

[0143] The calculation module 22 is used to perform differential calculations on the service start time and the sampling time points in each sampling period to obtain the attenuation coefficient sequence, and to perform time-domain pulse transformation on the passage frequency and full load ratio to generate a pulse sequence. By aligning the attenuation coefficient sequence with the pulse sequence in time, a spatiotemporal correlation matrix is ​​constructed.

[0144] Encoding module 23 is used to pulse code the pulse sequence using pulse code modulation with the attenuation coefficient in the spatiotemporal correlation matrix as a dynamic adjustment factor to obtain a bit stream sequence;

[0145] Mapping module 24 is used to map the bit stream sequence as an observation sequence to a preset road grid map, calculate the metric value of the transition branch between different road states in the road grid map through the maximum likelihood criterion, and determine the optimal transition path through optimal path planning based on the metric value. The road states include healthy state, sub-healthy state, damaged state and maintenance state.

[0146] The determination module 25 is used to determine the grid step size for the road state to transition to the maintenance state at the current moment from the optimal transfer path, determine the remaining time for the municipal road to enter the maintenance cycle based on the grid step size, and determine maintenance demand prediction information including maintenance priority and expected maintenance date based on the remaining time and preset road weights.

[0147] The machine learning-based municipal road maintenance demand prediction system of this application embodiment is used to implement the aforementioned machine learning-based municipal road maintenance demand prediction method. Therefore, the specific implementation of the machine learning-based municipal road maintenance demand prediction system can be found in the embodiment section of the machine learning-based municipal road maintenance demand prediction method above. The specific implementation can be referred to the description of the corresponding embodiment, and will not be repeated here.

[0148] Figure 5 A schematic diagram of the hardware structure of the electronic device provided in an embodiment of this application is shown.

[0149] This application also provides an electronic device, including: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the machine learning-based municipal road maintenance demand prediction method described above.

[0150] The electronic device may include a processor 510 and a memory 520 storing computer program instructions.

[0151] Specifically, the processor 510 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0152] Memory 520 may include mass storage for data or instructions. For example, and not limitingly, memory 520 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 520 may include removable or non-removable (or fixed) media. Where appropriate, memory 520 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 520 is non-volatile solid-state memory.

[0153] Memory may include read-only memory (ROM), random access memory (RAM), disk storage media devices, optical storage media devices, flash memory devices, and electrical, optical, or other physical / tangible memory storage devices. Therefore, typically, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to the first aspect of this disclosure.

[0154] The processor 510 reads and executes computer program instructions stored in the memory 520 to implement any of the machine learning-based municipal road maintenance demand prediction methods in the above embodiments.

[0155] In one example, the electronic device may also include a communication interface 530 and a bus 540. Wherein, such as Figure 5 As shown, the processor 510, memory 520, and communication interface 530 are connected through bus 540 and complete communication with each other.

[0156] The communication interface 530 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0157] Bus 540 includes hardware, software, or both, that couples components of an online data traffic metering device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 540 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, any suitable bus or interconnect is contemplated herein.

[0158] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described machine learning-based municipal road maintenance demand prediction methods.

[0159] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0160] Embodiments of the present invention also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the embodiments of the machine learning-based municipal road maintenance demand prediction method described above.

[0161] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0162] The above provides a detailed description of the machine learning-based municipal road maintenance demand prediction method and system provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A machine learning-based method for predicting municipal road maintenance needs, characterized in that, include: Obtain the service start time of municipal roads and the frequency of heavy-duty vehicle traffic and the proportion of full load in multiple sampling periods; Differential calculations are performed on the service start time and the sampling time points within each sampling period to obtain the attenuation coefficient sequence. Time-domain pulse transformation is performed on the passage frequency and the full load ratio to generate a pulse sequence. By aligning the attenuation coefficient sequence with the pulse sequence in time, a spatiotemporal correlation matrix is ​​constructed. Using the attenuation coefficient in the spatiotemporal correlation matrix as a dynamic adjustment factor, pulse code modulation is used to pulse code the pulse sequence to obtain a bit stream sequence. The bitstream sequence is mapped as an observation sequence to a preset road grid map. The metric value of the transition branch between different road states in the road grid map is calculated by the maximum likelihood criterion. Based on the metric value, the optimal transition path is determined by optimal path planning. The road states include healthy state, sub-healthy state, damaged state and maintenance state. The grid step size for transitioning the current road state to the maintenance state is determined from the optimal transfer path. Based on the grid step size, the remaining time for the municipal road to enter the maintenance cycle is determined. Based on the remaining time and the preset road weight, maintenance demand prediction information including maintenance priority and expected maintenance date is determined. The process involves performing differential calculations on the service start time and sampling time points within each sampling period to obtain an attenuation coefficient sequence, and performing a time-domain pulse transformation on the passage frequency and the full load ratio to generate a pulse sequence. A spatiotemporal correlation matrix is ​​constructed by temporally aligning the attenuation coefficient sequence with the pulse sequence, including: The service start time and the time interval between the sampling time points in each sampling period are calculated to obtain the service step sequence. The service step sequence is then nonlinearly mapped using a preset attenuation function to obtain the attenuation coefficient sequence. The signal amplitude is determined based on the full load ratio, and the signal density is determined based on the passage frequency. The interval time of the pulse distribution is calculated using the signal density, and multiple trigger times are determined within the sampling period according to the interval time. The signal amplitude is used as the pulse height to generate a corresponding pulse signal at each trigger time, thus obtaining a pulse sequence. According to the chronological order of sampling time points within the sampling period, position mapping is performed using the attenuation coefficient in the attenuation coefficient sequence as the vertical component and the pulse signal in the pulse sequence as the horizontal component to construct a spatiotemporal correlation matrix; The step of using the attenuation coefficient in the spatiotemporal correlation matrix as a dynamic adjustment factor to pulse code the pulse sequence using pulse code modulation to obtain a bitstream sequence includes: The attenuation coefficient in the spatiotemporal correlation matrix is ​​used as a dynamic adjustment factor, and the dynamic adjustment factor is matched with a preset quantization mapping rule to obtain multiple quantization steps. Based on the quantization step, the amplitude range corresponding to the pulse sequence is divided into multiple quantization intervals, and the pulse height of each pulse signal in the pulse sequence is assigned to the corresponding quantization interval to determine the quantization level corresponding to each pulse signal. Each quantization level is converted into a corresponding binary symbol using preset encoding instructions to obtain a bitstream sequence; The step of mapping the bitstream sequence as an observation sequence to a preset road grid map, and calculating the metric value of the transition branch between different road states in the road grid map using the maximum likelihood criterion, includes: Based on the road state evolution order, the road state nodes between adjacent sampling time points in the road grid are unidirectionally connected to obtain multiple transition branches. Each sampling time point in the road grid includes four road state nodes corresponding to the healthy state, sub-healthy state, damaged state, and maintenance state. Based on the chronological order of the sampling time points, the transfer branches corresponding to each sampling time point are arranged to obtain multiple candidate transfer paths; According to the sampling period to which each transfer branch belongs, the corresponding observation segment is extracted from the bit stream sequence, and based on the preset state feature code library, the similarity of each observation segment is compared with each transfer branch in the corresponding sampling period to obtain the branch matching probability of each transfer branch. Based on the branch matching probability corresponding to each transition branch, the log-likelihood probability of each transition branch under the constraint of the observation sequence is calculated, and the log-likelihood probability is determined as a metric.

2. The method according to claim 1, characterized in that, Based on the branch matching probability corresponding to each transition branch, the log-likelihood probability of each transition branch under the constraints of the observation sequence is calculated, and the log-likelihood probability is determined as a metric, including: Determine the starting node to which each transfer branch connects in the road grid map; Within the first sampling period, the preset initial score is determined as the path score, and the path score is summed with the branch matching probability corresponding to each transition branch derived from the starting node to obtain the log-likelihood probability of each transition branch. In subsequent sampling periods, the path score is determined based on the maximum log-likelihood probability among all transition branches pointing to the starting node at the sampling time point corresponding to the previous sampling period. The path score is then summed with the branch matching probability corresponding to each transition branch derived from the starting node to obtain the log-likelihood probability of each transition branch. The log-likelihood probability is determined as a metric for the transition branch.

3. The method according to claim 1, characterized in that, Based on the aforementioned metric, the optimal transfer path is determined through optimal path planning, including: Determine the termination sampling time point of the observation sequence, and select the maximum metric value from the metric values ​​of the transition branch corresponding to the termination sampling time point; Using the transition branch corresponding to the maximum metric value as the backtracking starting point, the state path is backtracked in the reverse direction of the sampling time point in the road grid map to obtain the corresponding node sequence, and the node sequence is determined as the optimal transition path.

4. The method according to claim 1, characterized in that, The grid step size for transitioning the current road state to the maintenance state is determined from the optimal transfer path. Based on the grid step size, the remaining time for the municipal road to enter the maintenance cycle is determined. Based on the remaining time and preset road weights, maintenance demand prediction information, including maintenance priority and expected maintenance date, is determined, including: In the optimal transfer path, determine the node position corresponding to the current moment and the endpoint position corresponding to the maintenance state; The number of sampling points between the node position and the endpoint position is used as the grid step size. The product of the grid step size and the preset unit sampling time is calculated to obtain the remaining time for the municipal road to enter the maintenance cycle. Based on a preset priority determination matrix, the emergency level corresponding to the remaining time is cross-matched with the importance level corresponding to the road weight to determine the maintenance priority. The remaining time and the current time are then added together to obtain the expected maintenance date, thereby determining maintenance demand prediction information that includes the maintenance priority and the expected maintenance date.

5. A municipal road maintenance demand prediction system based on machine learning, characterized in that, include: The acquisition module is used to acquire the service start time of municipal roads and the frequency of heavy-duty vehicles and the proportion of full load in multiple sampling periods; The calculation module is used to perform differential calculation on the service start time and the sampling time points in each sampling period to obtain the attenuation coefficient sequence, and to perform time-domain pulse transformation on the passage frequency and the full load ratio to generate a pulse sequence. By aligning the attenuation coefficient sequence with the pulse sequence in time, a spatiotemporal correlation matrix is ​​constructed. The calculation module is specifically used to calculate the time interval between the service start time and the sampling time points within each sampling period to obtain a service step sequence, and to perform nonlinear mapping on the service step sequence using a preset attenuation function to obtain an attenuation coefficient sequence; to determine the signal amplitude based on the full load ratio, and to determine the signal density based on the passage frequency, to calculate the interval time of the pulse distribution using the signal density, and to determine multiple trigger times within the sampling period according to the interval time, to generate a corresponding pulse signal at each trigger time using the signal amplitude as the pulse height, to obtain a pulse sequence; and to construct a spatiotemporal correlation matrix by performing position mapping with the attenuation coefficient in the attenuation coefficient sequence as the vertical component and the pulse signal in the pulse sequence as the horizontal component according to the chronological order of the sampling time points within the sampling period. The encoding module is used to pulse code the pulse sequence using the attenuation coefficient in the spatiotemporal correlation matrix as a dynamic adjustment factor to obtain a bit stream sequence; The encoding module is specifically used to use the attenuation coefficient in the spatiotemporal correlation matrix as a dynamic adjustment factor, and to match the dynamic adjustment factor with a preset quantization mapping rule to obtain multiple quantization steps; Based on the quantization step, the amplitude range corresponding to the pulse sequence is divided into multiple quantization intervals, and the pulse height of each pulse signal in the pulse sequence is assigned to the corresponding quantization interval to determine the quantization level corresponding to each pulse signal. Each quantization level is converted into a corresponding binary symbol using preset encoding instructions to obtain a bitstream sequence; The mapping module is used to map the bitstream sequence as an observation sequence to a preset road grid map. It calculates the metric value of the transition branches between different road states in the road grid map using the maximum likelihood criterion. Based on the metric value, it determines the optimal transition path through optimal path planning. The road states include healthy state, sub-healthy state, damaged state, and maintenance state. Specifically, the mapping module is used to unidirectionally connect road state nodes between adjacent sampling time points in the road grid map based on the road state evolution order, obtaining multiple transition branches. Each sampling time point in the road grid map includes four road state nodes corresponding to the healthy state, sub-healthy state, damaged state, and maintenance state. Based on the chronological order of the sampling time points, the transition branches corresponding to each sampling time point are arranged to obtain multiple candidate transition paths. According to the sampling period to which each transition branch belongs, the corresponding observation segment is extracted from the bitstream sequence, and based on a preset state feature code library, each observation segment is compared with each transition branch within the corresponding sampling period to obtain the branch matching probability of each transition branch. Based on the branch matching probability corresponding to each transition branch, calculate the log-likelihood probability of each transition branch under the constraint of the observation sequence, and determine the log-likelihood probability as a metric. The determination module is used to determine the grid step size for the road state to transition to the maintenance state from the optimal transfer path at the current moment, determine the remaining time for the municipal road to enter the maintenance cycle based on the grid step size, and determine maintenance demand prediction information including maintenance priority and expected maintenance date based on the remaining time and preset road weights.

6. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the machine learning-based municipal road maintenance demand prediction method as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the implementation of the machine learning-based municipal road maintenance demand prediction method as described in any one of claims 1 to 4.

Images