An asphalt pavement apparent disease evolution deduction method, system, device and medium

Abstract

The present application provides an asphalt pavement apparent disease evolution deduction method, system, device and medium, which belongs to the field of road engineering, transportation infrastructure operation and maintenance and intelligent detection technology, and comprises the following steps: constructing a unified space-time reference system, aligning and normalizing multi-source data in space-time; forming section-level disease quantification characteristics based on the recognition results of the inspection images and the structural indexes; event coding the maintenance measures of the road surface and pre-processing the section-level disease quantification characteristics corresponding to the time of the maintenance measures; constructing the correlation, time lag effect and spatial consistency between the disease types, disease characteristics and structural indexes to generate the collaborative evolution characteristics of the coupling relationship between diseases; constructing a deep time series prediction model to obtain the disease development trend at a specified time scale in the future, and forming the section-level disease evolution prediction result. The present application can depict the disease development law at different road sections and different time scales, and predict the future disease expansion.

Description

Technical Field

[0001] This invention belongs to the field of road engineering, transportation infrastructure operation and maintenance and intelligent detection technology, and in particular relates to a method, system, equipment and medium for predicting the evolution of apparent defects in asphalt pavement. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Road surface defects exhibit significant temporal, evolutionary, and coupled characteristics. For example, transverse cracks develop under the influence of temperature, longitudinal cracks and ruts exhibit a coupling mechanism in the wheel track, and network cracks will significantly increase the probability of potholes after they are connected.

[0004] Existing research has the following shortcomings: multi-source data lacks a unified spatiotemporal benchmark (image, traffic, climate, and structural detection data are difficult to align); disease quantification indicators lack a unified system, and geometric, topological, and dynamic characteristics are difficult to compare across periods; exogenous interventions (maintenance, overlay) and structural mutations are not explicitly modeled, leading to chaotic evolution processes; prediction models mostly use single-modal inputs, ignoring the coupling relationship between diseases; and there is a lack of a visualization and deduction system for maintenance decision-making. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, this invention proposes a method, system, equipment, and medium for predicting the evolution of apparent pavement defects. By performing spatiotemporal unified processing of inspection images, structural indicators, traffic loads, and meteorological data, and constructing segment-level defect quantification features, intervention event models, and defect co-evolution characteristics, a deep time-series prediction model is used to predict future trends of pavement defects. This invention can characterize defect development patterns in different road segments and at different time scales, and predict future defect expansion, providing a basis for precise road maintenance.

[0006] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0007] In a first aspect, the present invention discloses a method for predicting the evolution of apparent defects in asphalt pavements, comprising:

[0008] A unified spatiotemporal reference system is constructed to align and normalize multi-source data in spatiotemporal terms; wherein, the multi-source data includes inspection images, structural indicators, traffic loads and meteorological data;

[0009] Based on the recognition results of the inspection images and combined with the structural indicators, segment-level disease quantitative features are formed;

[0010] The maintenance measures for the road surface are coded as events, and the segment-level defects at the time corresponding to the maintenance measures are preprocessed to obtain the defect type and defect characteristics.

[0011] Construct the co-evolutionary characteristics of the coupling relationship between disease types, disease characteristics, and structural indicators, including correlation, time lag effect, and spatial consistency;

[0012] The deep time-series prediction model, constructed by inputting the quantitative characteristics of the segment-level disease, co-evolution characteristics, traffic load, and meteorological data, obtains the disease development trend at a specified future time scale, thus forming the segment-level disease evolution prediction result.

[0013] Secondly, this invention discloses a system for predicting the evolution of apparent defects in asphalt pavements, comprising:

[0014] The data acquisition module, by constructing a unified spatiotemporal reference system, performs spatiotemporal alignment and normalization of multi-source data; wherein, the multi-source data includes inspection images, structural indicators, traffic loads, and meteorological data;

[0015] The feature extraction module is used to form segment-level disease quantitative features based on the recognition results of the inspection images and the structural indicators;

[0016] The intervention modeling module is used to encode the maintenance measures of the road surface as events and to perform feature preprocessing on the quantitative features of the section-level defects at the time corresponding to the maintenance measures to obtain the defect type and defect features.

[0017] The co-evolution module is used to construct co-evolutionary features that generate coupling relationships between disease types, disease characteristics, and structural indicators, including correlations, time lag effects, and spatial consistency.

[0018] The disease prediction module is used to construct a deep time series prediction model by inputting the segment-level disease quantitative characteristics, co-evolution characteristics, traffic load and meteorological data to obtain the disease development trend at a specified time scale in the future, and form segment-level disease evolution prediction results.

[0019] Thirdly, the present invention discloses an electronic device, including a memory and a processor, and computer instructions stored in the memory and running on the processor, wherein the computer instructions, when run by the processor, complete the steps of the above-mentioned method for extrapolating the evolution of apparent defects in asphalt pavement.

[0020] Fourthly, the present invention discloses a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps of the above-mentioned method for extrapolating the evolution of apparent defects in asphalt pavement.

[0021] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0022] (1) This invention realizes unified spatiotemporal alignment and fusion analysis of multi-source heterogeneous data. By constructing a unified spatial reference system based on the linear mileage of roads and a unified time axis with a fixed time step, the multi-source data such as inspection images, structural detection data, traffic load and meteorology are spatiotemporally normalized and gridded, which solves the problem of inconsistent data benchmarks and difficulty in collaborative analysis in traditional methods, and provides a reliable and consistent data foundation for disease evolution modeling.

[0023] (2) This invention establishes a multi-level and multi-dimensional disease feature system, which comprehensively improves the ability to characterize diseases, breaks through the limitations of traditional single indicators (such as area and length), and extracts comprehensive feature vectors including disease geometric shape, directional distribution, topological connectivity and time series change rate. It can describe the current state and development trend of diseases more precisely and comprehensively, and enhance the model's ability to represent complex disease morphology and its evolution law.

[0024] (3) This invention introduces exogenous intervention event modeling to ensure the continuity and authenticity of the disease evolution sequence. By encoding and adjusting the characteristics of maintenance measures such as sealing, repairing, and covering, the invention explicitly models the inhibitory or resetting effect of maintenance behavior on disease development, avoiding the temporal break and noise interference caused by human intervention, and enabling the model to learn the disease development trend that is closer to the natural evolution law.

[0025] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0026] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0027] Figure 1 This is a flowchart of the method for extrapolating the evolution of apparent defects in asphalt pavement as described in Embodiment 1 of the present invention.

[0028] Figure 2 This is a schematic diagram of the feature construction relationship in the evolutionary deduction described in Embodiment 1 of the present invention. Detailed Implementation

[0029] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0030] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0031] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0032] Example 1

[0033] In one or more embodiments, a method for extrapolating the evolution of apparent defects in asphalt pavements is disclosed. This method integrates domain knowledge with a unified spatiotemporal grid of multi-source observation data, structured features of apparent defects, and a deep extrapolation model to obtain interpretable and predictable defect evolution patterns. It addresses the spatiotemporal development characteristics of typical defects such as cracks, potholes, and rutting, constructing a unified spatiotemporal benchmark, a structured defect feature system, an exogenous intervention model, and a co-evolutionary feature set. Combined with a deep time-series prediction model, it achieves accurate extrapolation of future defect development trends. This invention can quantitatively describe the defect evolution process across road sections and time scales, providing a scientific basis for road asset management and maintenance decisions, such as... Figures 1-2 As shown, it includes the following steps:

[0034] Step S1: Construct a unified spatiotemporal reference system to align and normalize multi-source data in spatiotemporal terms; wherein, the multi-source data includes inspection images, structural indicators, traffic loads, and meteorological data.

[0035] The system integrates inspection image recognition results, structural indicators such as rut ​​depth, traffic load, and meteorological data. It projects spatial locations according to the linear mileage of roads and generates a segment-time two-dimensional grid based on a fixed time step, achieving spatiotemporal alignment and normalization of multi-source data. Specifically, this includes:

[0036] Step S1-1: Establish a spatial linear reference system.

[0037] Obtain the centerline geometric information and corresponding mileage marker data of the road to be analyzed, and establish a one-dimensional linear mileage coordinate axis along the road travel direction with the road starting point as the reference zero point.

[0038] The locations of inspection images, disease identification results, structural detection points, traffic section locations, and meteorological observation points under the geographic coordinate system are uniformly mapped onto the linear mileage coordinate axis, thereby achieving a spatially consistent expression of data from different sources in the road direction.

[0039] Through the above processing, all disease-related observation data are indexed by "road mileage location" as a unique spatial index, avoiding data misalignment caused by coordinate system differences.

[0040] Step S1-2: Divide the road into sections.

[0041] The road is discretized along the linear mileage coordinate axis according to the preset segment length, forming multiple continuous and non-overlapping road segments, with each segment serving as a spatial analysis unit.

[0042] Preferably, the preset section length is 200 meters. This length is consistent with the road segment division unit in the current road maintenance management system and ensures the statistical stability of the number of defects within each section. This ensures the statistical stability of the number of defects within the section while meeting the practical application requirements of the road maintenance management unit.

[0043] Each disease observation result is assigned to the corresponding road section based on its mileage location, realizing the spatial aggregation of disease information at the section level.

[0044] Steps S1-3: Time axis unification and resampling.

[0045] The system acquires the raw data collection time information of inspection images, structural indicators, traffic load data, and meteorological data, and sets a uniform time step as the analysis cycle.

[0046] Data from different time frequencies and collection times are resampled using a fixed time step to ensure a correspondence between various data types at the same point in time. The fixed time step is set to one month to align with the regular road surface inspection cycle and monthly average meteorological data.

[0047] For data collected at a frequency higher than the time step, statistical aggregation is used for integration; for data collected at a frequency lower than the time step or with missing data, interpolation or smoothing is used to fill in the missing data and the missing status information is recorded.

[0048] Steps S1-4: Construct a segment-time two-dimensional grid. Combine the unified resampled time nodes with the divided road segments using Cartesian products to form a structured segment-time two-dimensional grid, achieving alignment and integration of all data in the spatiotemporal dimension.

[0049] Specifically, the road segments obtained in step S1-2 are combined with the unified time nodes in step S1-3 to construct a segment-time two-dimensional grid with road segments and time steps as dimensions.

[0050] Each grid cell in the two-dimensional grid is used to carry the disease characteristics, structural indicators, traffic load information and meteorological information of the corresponding section at the corresponding time.

[0051] This two-dimensional grid structure enables the synchronous expression of disease information in space and time, providing a unified data framework for subsequent disease feature extraction and temporal modeling.

[0052] Steps S1-5: Data Consistency and Quality Control Processing. Consistency checks and quality control processes are performed on various types of data in the segment-time two-dimensional grid, including:

[0053] When there is insufficient effective observation data for a certain segment within a certain time step, the grid cell is marked or downweighted.

[0054] When the image recognition area of ​​the disease and the structural index change in opposite directions within the same segment and time step, the image recognition result is taken as the standard according to the preset rules, and the grid cell is recorded as a conflict sample.

[0055] The grid data, after consistency verification and quality control, will serve as a reliable input for subsequent disease feature extraction, intervention modeling, and evolutionary deduction, ensuring that the data foundation for model training and prediction is consistent and credible.

[0056] To more clearly illustrate the process of constructing a unified spatiotemporal reference system in step S1, this embodiment takes a section of a highway from K100+000 to K120+000 as an example to give a specific example of data spatiotemporal normalization and gridding organization.

[0057] (1) Establishment and mapping of spatial linear reference systems;

[0058] Using the road starting point K100+000 as the reference zero point, a one-dimensional linear mileage coordinate axis is established along the road travel direction, and multi-source data is uniformly mapped to this coordinate axis:

[0059] Inspection images and disease identification results: During a certain inspection, a transverse crack was identified at K105+230 and a pothole was identified at K112+450, which were mapped to mileage points 105230 and 112450 respectively.

[0060] Structural inspection data: The falling weight deflectometer was used at chainages such as K108+000 and K110+000, and the original chainage mapping was used directly;

[0061] Traffic load data: The traffic survey station is located at K115+500, and the axle load data is linked to this mileage point;

[0062] Meteorological observation data: observation data from meteorological stations near the road section, marked as a uniform meteorological background value covering the entire road section.

[0063] (2) Section division;

[0064] The road is discretized along the mileage axis in 200-meter sections, forming continuous, non-overlapping segments. For example:

[0065] Section S1: K100+000~K100+200;

[0066] Section S2: K100+200~K100+400;

[0067] ...

[0068] Section S100: K119+800~K120+000;

[0069] The aforementioned defects, structural inspections, and traffic load data are grouped into corresponding sections according to their mileage locations. For example, the transverse crack at K105+230 is assigned to section S26 (corresponding to K105+200~K105+400), and the pothole at K112+450 is assigned to section S62 (corresponding to K112+400~K112+600).

[0070] (3) Time axis unification and resampling;

[0071] A fixed time step of one month was set, with the analysis period based on the calendar month. Resampling processing for various data types was performed as follows:

[0072] Inspection images: This road section is inspected quarterly, with images collected in January, April, July, and October 2023. After resampling, January data was used as January features, April data as April features, and data for February, March, and other months without data were filled in using linear interpolation before and after the data was collected, and the missing data were marked.

[0073] Structural indicators: tested annually (June 2023), using a forward-filling method, assigning the test value to each time step from June to May of the following year;

[0074] Traffic load: The traffic survey station provides daily axle load data, which is summarized monthly as monthly cumulative equivalent axle loads;

[0075] Meteorological data: Daily temperature and precipitation data are calculated monthly averages and cumulative values ​​to form monthly meteorological characteristics.

[0076] (4) Construction of segment-time two-dimensional grid;

[0077] The 100 road sections are combined with time nodes on the timeline using Cartesian products. Taking June 2023 as an example, the following grid cells are generated, as shown in Table 1.

[0078] Table 1. Examples of Mesh Cells

[0079]

[0080] (5) Quality control procedures;

[0081] Perform a consistency check on the above two-dimensional mesh:

[0082] Section S63 had a significantly higher structural index (deflection 0.42mm) in June 2023 than adjacent sections. After verification, this was consistent with pothole defects and was therefore retained.

[0083] Section S15 had no inspection data in February 2023 and a traffic load missing rate of over 30%. This grid cell was marked as "insufficient data" and was downweighted in subsequent model training.

[0084] Through the above processing, the multi-source data that was originally scattered in images, detection reports, traffic sections, and meteorological stations was uniformly organized into a structured segment-time two-dimensional grid, providing a spatiotemporally consistent data foundation for subsequent disease feature extraction, intervention modeling, and evolution prediction.

[0085] Step S2: Based on the identification results of the inspection images, combine structural indicators to form section-level disease quantitative characteristics.

[0086] The geometric morphology, directionality, density, and topological relationships of defects such as cracks and pits are extracted from image recognition results. These are then combined with structural indicators to form segment-level quantitative features of defects, used to describe the severity and development status of the defects. Specifically:

[0087] Step S2-1: Segment aggregation of disease identification results;

[0088] Based on the segment-time two-dimensional grid constructed in step S1, the defects such as cracks and potholes identified in the inspection images are collected into the corresponding road segments and time nodes according to their corresponding road mileage locations and collection times.

[0089] Multiple disease identification results within the same section and at the same time point are aggregated and processed to form a section-level raw disease data set.

[0090] Step S2-2: Extract the geometric features of the disease.

[0091] For the aggregated original data set of disease at the section level, extract geometric features that characterize the scale and morphology of the disease, including but not limited to: the number, length, area, width of the disease, and the proportion of the disease to the effective area of ​​the section.

[0092] Among them, crack-type diseases are statistically analyzed based on their length and width according to linear characteristics, while pit-type diseases are statistically analyzed based on their area according to their surface characteristics. The effective area of ​​the section is used as the normalization benchmark to achieve comparability of disease severity between different sections.

[0093] Specifically, a segment-level disease area severity index can be defined. for:

[0094]

[0095] in, This represents the area of ​​the disease corresponding to segment S at time t. Indicates the effective area of ​​the section.

[0096] Step S2-3: Extract the directional and distribution characteristics of the disease.

[0097] The driving direction information of the road section is obtained, and the extension direction of the crack is compared with the driving direction. The distribution ratio of cracks in the longitudinal and transverse directions is calculated to obtain directional characteristics.

[0098] Through the above processing, the main development direction of different defects in the road structure can be distinguished to reflect the differences in the influence of traffic load or environmental factors on defects.

[0099] Meanwhile, the spatial uniformity of disease distribution within a section can be quantified using the coefficient of variation method to describe the concentrated or discrete development state of the disease. Specifically, if road section s is divided into n equal-length sub-units, the disease density of each sub-unit i is... (The proportion of diseased area to the area of ​​the sub-unit), then:

[0100]

[0101]

[0102]

[0103] in, This represents the average density of disease. The standard deviation of disease density, The value is the coefficient of variation. The smaller the value, the more uniform the distribution of the disease; the larger the value, the more concentrated the distribution of the disease.

[0104] Step S2-4: Extract the topological features of the disease.

[0105] Connectivity analysis was performed on cracks within the section to extract topological features reflecting whether cracks have evolved from isolated to interconnected states. These topological features characterize the degree to which cracks evolve from isolated morphologies to a network structure, thus depicting the structural differences in the disease at different development stages.

[0106] This embodiment introduces connectivity and penetration features to make the disease characteristics not only reflect the current scale, but also the potential expansion trend.

[0107] Step S2-5: Extract the temporal variation characteristics of the disease.

[0108] Specifically, to characterize the evolution trend of diseases over time, a disease growth rate index can be constructed based on the changes in disease characteristics at adjacent time points as a feature of disease temporal change.

[0109]

[0110] in, Indicates a section At any moment The disease characteristic values, For time step, To prevent stable terms with a denominator of zero, This is a disease growth rate indicator. Its changing characteristics describe the growth, expansion, or slowdown trends of diseases, providing time-dimensional information support for subsequent disease evolution projections.

[0111] Step S2-6: Form segment-level disease quantitative features. The geometric features, directional features, spatial distribution uniformity features, topological features, and temporal variation features of the diseases mentioned above are combined to form unified segment-level disease quantitative features. These segment-level disease quantitative features serve as input data for subsequent exogenous intervention modeling, disease co-evolution analysis, and disease evolution prediction models.

[0112] Step S3: Event-based coding of pavement maintenance measures and feature preprocessing of the segment-level defect quantification characteristics at the corresponding time of each maintenance measure to obtain defect types and characteristics. Specifically, event-based coding is performed on maintenance measures such as joint sealing, local repair, and milling overlay, and the defect characteristics at the corresponding time are reset, weighted, or smoothed to eliminate the abrupt impact caused by intervention events and obtain a defect sequence that conforms to the laws of natural evolution. Exogenous intervention event modeling of pavement maintenance measures specifically includes:

[0113] Step S3-1: Collection and type identification of maintenance intervention events.

[0114] Obtain historical road maintenance records, which include the maintenance time, the section where the maintenance occurred, and the maintenance method. The maintenance method includes at least one or more types such as joint sealing, partial repair, milling, and overlay. Map the maintenance records to the section-time two-dimensional grid constructed in step S1 according to their corresponding road sections and occurrence times, forming section-level, time-seriesd intervention event records.

[0115] Step S3-2, Intervention Event Sequence Coding:

[0116] For the mapped maintenance intervention events, intervention indication information is constructed in the segment-time two-dimensional grid to identify whether maintenance behavior has occurred at the corresponding time node.

[0117] For the sections where maintenance occurred, the start time of the maintenance event was recorded, and event continuity information reflecting the time span after maintenance was constructed to distinguish the disease evolution status at different stages before and after maintenance.

[0118] The above coding allows maintenance behaviors to be explicitly present in a structured form in the time series, avoiding misidentification as natural fluctuations in disease.

[0119] Step S3-3: Modeling the impact of intervention events on disease characteristics:

[0120] Based on the characteristics of different maintenance methods, corresponding adjustments are made to the disease characteristics, including: for maintenance methods that can significantly reduce the severity of disease, the disease characteristics are reset or reduced at the time point after the maintenance occurs; for maintenance methods that have a delaying effect on disease development, the trend of disease characteristic changes is smoothed or attenuated.

[0121] The above treatment methods are used to reflect the actual impact of maintenance practices on the state of disease, making the changes in disease characteristics more consistent with engineering practice.

[0122] Steps S3-4: Maintain the continuity of disease evolution after intervention.

[0123] After completing the intervention and impact treatment, the continuity of the disease characteristic sequence before and after maintenance is constrained to avoid abnormal jumps in the disease time sequence caused by maintenance.

[0124] Specifically, the continuity constraint can be implemented using a time window smoothing method. When using the time window smoothing method, the maintenance time point is used as the basis for the constraint. Select window as center The disease characteristic sequence within the data is smoothed using an exponentially weighted moving average, so that the characteristic mutations caused by maintenance are transformed into continuous gradual changes, while preserving the overall trend of maintenance adjustments.

[0125] This continuous processing ensures that the disease evolution sequence maintains the authenticity of the maintenance effect while still possessing a stable structure that can be used for time series analysis and prediction.

[0126] Step S3-5: Output the corrected segment-level disease quantitative characteristics of each type of disease as intervention event characteristics, including disease type and its corresponding disease characteristics;

[0127] The disease characteristics, after being modeled and processed by the intervention event, along with the intervention type and time information, are output as enhanced segment-level disease quantitative features. These enhanced segment-level disease quantitative features are used in subsequent steps of constructing disease co-evolution features and inferring disease evolution.

[0128] Step S4: Construct the correlation, time delay effect and spatial consistency between disease types, disease characteristics and structural indicators to generate co-evolutionary features that characterize the coupling relationship between diseases such as cracks and pits.

[0129] Step S4-1: Align and combine the characteristics of multiple disease types.

[0130] Based on the segment-time two-dimensional grid constructed in step S1, the enhanced segment-level disease quantification features obtained in step S3 are extracted. Disease types include at least crack diseases, pothole diseases, and rutting diseases. The segment-level disease quantification features of different disease types at the same segment and the same time node are aligned to form a multi-disease joint feature set, which is used to characterize the coexistence of multiple diseases under the same spatiotemporal conditions.

[0131] Step S4-2: Construct the correlation features between diseases.

[0132] Based on the multi-disease joint feature set generated in step S4-1, the historical feature sequence of each disease type at the segment scale is extracted, and the consistency features of changes among different disease types at the segment scale are calculated.

[0133] The consistency of change characteristic can be obtained by calculating the Pearson correlation coefficient between characteristic sequences of different disease types. Specifically, for each segment, historical characteristic sequences of each disease type are extracted, and the correlation coefficient between any two disease type sequences is calculated to construct a correlation feature vector reflecting the degree of synchronous change among diseases. The consistency of change characteristic is used to describe the degree of synchronicity of different diseases over time, reflecting the synergistic or inhibitory relationship between diseases.

[0134] This correlation feature enables the model to identify the interconnected trends in the evolution of defects such as cracks, potholes, and ruts.

[0135] Step S4-3: Construct disease evolution time lag characteristics.

[0136] Based on the multi-disease joint feature set generated in step S4-1, a time lag analysis mechanism is introduced for the segment-level historical feature sequences of different disease types. By comparing the changes of different diseases at different time intervals, time lag features reflecting the leading and lagging characteristics of diseases are constructed.

[0137] The time lag feature is obtained by calculating the time lag cross-correlation between feature sequences of different disease types. Specifically, for each segment, historical feature sequences of two disease types are extracted, their cross-correlation coefficients within a preset time lag range are calculated, and the time lag at which the absolute value of the cross-correlation coefficient is the largest is taken as the lead / lag time of the two diseases, and this maximum cross-correlation coefficient is recorded.

[0138] In this embodiment, the preset time delay range for the time delay cross-correlation analysis is set to 0 to 12 months, with a step size of 1 month, to accommodate the common evolution time intervals between cracks and potholes, and between ruts and cracks.

[0139] The optimal time delay value and maximum cross-correlation coefficient of each pair of disease types are combined in a preset order to form a time delay feature vector. The time delay feature vector is constructed in the following preset order: optimal time delay of crack-pothole, maximum cross-correlation coefficient of crack-pothole, optimal time delay of crack-rut, maximum cross-correlation coefficient of crack-rut, optimal time delay of pothole-rut, and maximum cross-correlation coefficient of pothole-rut. The time delay features are used to characterize evolution paths such as crack development leading to potholes, and rutting intensification inducing longitudinal cracks.

[0140] Step S4-4: Construct spatial consistency and directional features.

[0141] Based on the multi-disease joint feature set generated in step S4-1, spatial distribution consistency analysis is performed on the disease characteristics of adjacent sections at the same time node. By comparing the direction and magnitude of disease severity changes in adjacent sections, spatial consistency features reflecting the expansion or concentration of diseases along the road direction are constructed. Spatial consistency features are used to characterize the longitudinal spread trend of diseases along the road, improving the ability to express the spatial continuity of disease evolution.

[0142] Step S4-5: Generate a co-evolutionary feature set.

[0143] The correlation features, time lag features, and spatial consistency features among diseases are combined to form a segment-level disease co-evolution feature set. This feature set serves as an independent feature input for subsequent disease evolution inference steps based on deep time-series models.

[0144] Step S5: Input the segment-level disease quantitative characteristics, co-evolution characteristics, traffic load and meteorological data into the constructed deep time series prediction model to obtain the disease development trend at a specified time scale in the future, and form the segment-level disease evolution prediction result.

[0145] Step S5-1: Align and combine multiple types of disease features, including segment-level disease quantitative features, co-evolution features, traffic load and meteorological data.

[0146] Based on the segment-time two-dimensional grid, each segment is used as the basic prediction unit, and disease evolution time series samples are constructed in a sliding time window manner.

[0147] Using each road segment as the basic prediction unit, a sliding time window approach is employed to construct time-series samples of disease evolution. Let the current time be t, and the historical time step be T, then each time-series sample can be represented as:

[0148]

[0149] in, This represents the input time series sample of segment s at time t; This is the comprehensive feature vector of segment s at time t, including: geometric features of the disease, co-evolutionary features of the disease, and external features (traffic load, meteorological data, etc.). Each time series sample includes at least the disease features, co-evolutionary features of the disease, and external features such as traffic load and meteorological environment corresponding to the current time node and multiple time nodes before it.

[0150] The time series samples constructed in the above manner only use historical observable information, avoiding the introduction of future information, thereby ensuring the causal consistency of the prediction process.

[0151] Step S5-2: Organize and standardize input features.

[0152] After constructing a complete time-series sample, features from different sources and with different physical meanings need to be uniformly organized and scaled to adapt to the input requirements of the deep model and ensure training stability. This step first organizes the multidimensional features from disease quantification, co-evolution, and the external environment into a normalized tensor structure according to time steps, and then performs standardization to eliminate dimensional differences.

[0153] Eigenvector construction: For any segment s, the eigenvector at time t It consists of the following three parts joined together vertically:

[0154]

[0155] in, The segment-level disease quantitative features are composed of the geometric, directional, topological, and temporal variation features of the disease extracted in step S2. The segment-level disease co-evolution characteristics are composed of the disease correlation, time lag, and spatial consistency features constructed in step S4; These are exogenous characteristics, including traffic load index, cumulative equivalent axle load, temperature, precipitation, and freeze-thaw cycles.

[0156] Feature standardization: Due to significant differences in the physical units and numerical ranges of various features (e.g., length in meters, area in square meters, temperature in degrees Celsius, traffic volume in standard axle loads), directly inputting them into the model would lead to the optimization process being dominated by large numerical features. Therefore, for each feature dimension... Perform Z-score standardization:

[0157]

[0158] In the formula, and Calculated based on the training set, and represent the mean and standard deviation of feature j, respectively. Each dimension of the feature after standardization.

[0159] This processing ensures that each feature follows a distribution with a mean of 0 and a standard deviation of 1, improving the model's convergence speed and generalization ability. For features with obvious skewed distributions, a logarithmic transformation or Box-Cox transformation can be performed before standardization.

[0160] Features with different dimensions and different value ranges are standardized or normalized to reduce the impact of feature scale differences on model training.

[0161] Step S5-3: Construct a deep time series prediction model.

[0162] The constructed time-series samples are organized by features, and the quantitative features of diseases, co-evolutionary features and exogenous features are combined into multi-dimensional time-series input data in chronological order.

[0163] Establish deep temporal prediction models such as Long Short-Term Memory Network (LSTM), Convolutional Temporal Network (TCN) or Transformer, input multidimensional temporal input data, output the disease development trend at a specified future time scale, and form segment-level disease evolution prediction results.

[0164] Step S5-4: Predict the evolution trend of the disease.

[0165] The constructed time-series samples are organized by features, and the quantitative features of diseases, co-evolutionary features and exogenous features are combined into multi-dimensional time-series input data in chronological order.

[0166] The model output is the predicted value of disease characteristics at the next time step Δt:

[0167]

[0168] In the formula, These are predicted values ​​for disease characteristics. This is a prediction model.

[0169] The prediction loss function uses a combination of mean squared error (MSE) and mean absolute error (MAE):

[0170]

[0171] in, For predicting the loss function, Mean square error, The mean absolute error, , These are the weighting coefficients. and The values ​​are 0.7 and 0.3 respectively. These values ​​are determined based on a grid search on the validation set, specifically according to the following:

[0172] (1) Engineering application requirements: Road maintenance decisions have a low tolerance for prediction errors of severe defects. Missing or seriously underestimating defects may lead to missing the optimal maintenance window and causing accelerated structural deterioration. The mean square error applies a square penalty to larger errors, which can effectively constrain the extreme bias of the model for high-risk samples.

[0173] (2) Error distribution characteristics: In the disease evolution data, the proportion of rapidly expanding samples caused by factors such as sudden overload and extreme weather is relatively small, but the impact is significant. By using a combination of mean squared error and mean absolute error, we can prioritize the control of extreme deviations while avoiding overfitting the model to a few abnormal samples and maintaining a stable fitting ability to the conventional evolution trend.

[0174] In practical applications, both can also be used as adjustable hyperparameters, which can be adaptively adjusted according to different road grades and data quality.

[0175] Step S5-5: Post-processing and visualization of prediction results.

[0176] The predicted values ​​output by the model are destandardized to restore the original dimensions:

[0177]

[0178] In the formula, and Calculated based on the training set, and represent the mean and standard deviation of feature j, respectively. The standardized predicted values ​​output by the model. These are the predicted values ​​in their original dimensions.

[0179] The preferred results are mapped back to the time grid of the segment to generate visualization results such as disease evolution heatmaps and trend curves, supporting maintenance decision analysis.

[0180] This invention solves the problems of inconsistent spatial benchmarks and inconsistent temporal frequencies among multi-source data by constructing a unified spatiotemporal reference system; it compensates for the shortcomings of traditional single-scale and multi-dimensional disease features by extracting multi-scale and multi-dimensional disease features, which cannot characterize the complex morphology of diseases; it eliminates the damage to the stability of data sequences caused by maintenance behavior by introducing exogenous intervention event modeling; it characterizes the coupling relationship between multiple types of diseases by constructing disease co-evolution features; and finally, it achieves accurate prediction of the evolution trend of pavement diseases by combining a deep time series prediction model.

[0181] Example 2

[0182] In one or more embodiments, a system for predicting the evolution of apparent defects in asphalt pavement is disclosed, specifically including:

[0183] The data acquisition module, by constructing a unified spatiotemporal reference system, performs spatiotemporal alignment and normalization of multi-source data; wherein, the multi-source data includes inspection images, structural indicators, traffic loads, and meteorological data;

[0184] The feature extraction module is used to form segment-level disease quantitative features based on the recognition results of the inspection images and the structural indicators;

[0185] The intervention modeling module is used to encode the maintenance measures of the road surface as events and to perform feature preprocessing on the quantitative features of the section-level defects at the time corresponding to the maintenance measures to obtain the defect type and defect features.

[0186] The co-evolution module is used to construct co-evolutionary features that generate coupling relationships between disease types, disease characteristics, and structural indicators, including correlations, time lag effects, and spatial consistency.

[0187] The disease prediction module is used to construct a deep time series prediction model by inputting the segment-level disease quantitative characteristics, co-evolution characteristics, traffic load and meteorological data to obtain the disease development trend at a specified time scale in the future, and form segment-level disease evolution prediction results.

[0188] Example 3

[0189] This embodiment provides an electronic device, including a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the computer instructions are executed by the processor, they complete the steps of the above-mentioned method for deducing the evolution of apparent defects in asphalt pavement.

[0190] Example 4

[0191] This embodiment provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps of the above-described method for extrapolating the evolution of apparent defects in asphalt pavement.

[0192] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0193] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0194] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment, whereby a series of operational steps are performed to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0195] The descriptions of each embodiment in the above embodiments have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0196] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for asphalt pavement apparent distress evolution extrapolation, characterized in that, include: A unified spatiotemporal reference system is constructed to align and normalize multi-source data in spatiotemporal terms; wherein, the multi-source data includes inspection images, structural indicators, traffic loads and meteorological data; Based on the recognition results of the inspection images and combined with the structural indicators, segment-level disease quantitative features are formed; The maintenance measures for the road surface are coded as events, and the segment-level defects at the time corresponding to the maintenance measures are preprocessed to obtain the defect type and defect characteristics. Construct the co-evolutionary characteristics of the coupling relationship between disease types, disease characteristics, and structural indicators, including correlation, time lag effect, and spatial consistency; The deep time series prediction model, constructed by inputting the segment-level disease quantitative characteristics, co-evolution characteristics, traffic load and meteorological data, obtains the disease development trend at a specified time scale in the future, forming the segment-level disease evolution prediction result. The construction of the co-evolutionary features is specifically as follows: Align the segment-level disease quantitative features of different disease types in the same area and at the same time node to form a set of joint features of multiple diseases; Based on the aforementioned multi-disease joint feature set, historical feature sequences of each disease type at the segment scale are extracted, and the consistency features of changes among different disease types at the segment scale are calculated to obtain the correlation features between diseases. Based on the aforementioned set of combined features of multiple diseases, a time lag analysis mechanism is introduced for the segment-level historical feature sequences of different disease types. By comparing the changing relationships of different diseases at different time intervals, a time lag feature reflecting the leading and lagging characteristics of diseases is constructed. Based on the aforementioned set of multiple disease joint features, spatial distribution consistency analysis is performed on the disease features of adjacent sections at the same time node. By comparing the direction and magnitude of the changes in disease severity in adjacent sections, spatial consistency features reflecting the expansion or concentration of diseases along the road direction are constructed. By combining the correlation characteristics, time lag characteristics, and spatial consistency characteristics among diseases, a segment-level disease co-evolution characteristic is formed.

2. The method for predicting the evolution of apparent defects in asphalt pavement as described in claim 1, characterized in that, The identification results based on the inspection images, combined with the structural indicators, form segment-level disease quantitative features, specifically as follows: The section where disease identification results are collected; Geometric features characterizing the scale and morphology of diseases are extracted from the aggregated original data set of segment-level diseases. The driving direction information of the road section is obtained, and the extension direction of the crack is compared and analyzed with the driving direction. The distribution ratio of cracks in the longitudinal and transverse directions is calculated to obtain directional characteristics. Connectivity analysis was performed on the cracks within the section to extract topological features that reflect whether the cracks have developed from a localized state to a continuous state. A disease growth rate index is constructed based on the changes in disease characteristics at adjacent time points as a characteristic of disease temporal change; By combining geometric features, directional features, topological features, and disease temporal variation features, a unified segment-level disease quantitative feature is formed.

3. The method for predicting the evolution of apparent defects in asphalt pavement as described in claim 1, characterized in that, The event-based coding of road maintenance measures includes: Obtain historical road maintenance records, which include the maintenance time, the section where the maintenance occurred, and the maintenance method. The maintenance records are mapped to the constructed section-time two-dimensional grid according to their corresponding road sections and occurrence time, forming section-level, time-series intervention event records; For the mapped maintenance intervention events, intervention indication information is constructed in a segment-time two-dimensional grid; For the sections where maintenance occurs, record the start time of the maintenance event and construct event continuity information that reflects the time span after maintenance.

4. The method for predicting the evolution of apparent defects in asphalt pavement as described in claim 3, characterized in that, The modeling intervention events affect disease characteristics. Based on the characteristics of different maintenance methods, corresponding adjustments are made to the disease characteristics, including: for maintenance methods that can significantly reduce the severity of disease, the disease characteristics are reset or reduced at the time point after the maintenance occurs; for maintenance methods that delay the development of disease, the trend of disease characteristic changes is smoothed or attenuated.

5. The method for predicting the evolution of apparent defects in asphalt pavement as described in claim 1, characterized in that, Align and combine multiple types of disease characteristics, including segment-level disease quantitative characteristics, co-evolution characteristics, traffic load and meteorological data; Based on the segment-time two-dimensional grid, each segment is used as the basic prediction unit, and disease evolution time series samples are constructed in the form of a sliding time window. Specifically, each road segment is used as the basic prediction unit, and disease evolution time series samples are constructed in the form of a sliding time window.

6. The method for predicting the evolution of apparent defects in asphalt pavement as described in claim 5, characterized in that, The constructed time-series samples are organized by features, and the quantitative features of diseases, co-evolution features and exogenous features are combined into multi-dimensional time-series input data in chronological order; The deep temporal prediction model uses a long short-term memory network, a convolutional temporal network, or a Transformer network to output the predicted value of the disease characteristics at the next time step Δt. The predicted values ​​output by the model are destandardized to restore the original dimensions and obtain the segment-level disease evolution prediction results.

7. A system for predicting the evolution of apparent defects in asphalt pavement, characterized in that, include: The data acquisition module, by constructing a unified spatiotemporal reference system, performs spatiotemporal alignment and normalization of multi-source data; wherein, the multi-source data includes inspection images, structural indicators, traffic loads, and meteorological data; The feature extraction module is used to form segment-level disease quantitative features based on the recognition results of the inspection images and the structural indicators; The intervention modeling module is used to encode the maintenance measures of the road surface as events and to perform feature preprocessing on the quantitative features of the section-level defects at the time corresponding to the maintenance measures to obtain the defect type and defect features. The co-evolution module is used to construct co-evolutionary features that generate coupling relationships between disease types, disease characteristics, and structural indicators, including correlations, time lag effects, and spatial consistency. The construction of the co-evolutionary features is specifically as follows: Align the segment-level disease quantitative features of different disease types in the same area and at the same time node to form a set of joint features of multiple diseases; Based on the aforementioned multi-disease joint feature set, historical feature sequences of each disease type at the segment scale are extracted, and the consistency features of changes among different disease types at the segment scale are calculated to obtain the correlation features between diseases. Based on the aforementioned set of combined features of multiple diseases, a time lag analysis mechanism is introduced for the segment-level historical feature sequences of different disease types. By comparing the changing relationships of different diseases at different time intervals, a time lag feature reflecting the leading and lagging characteristics of diseases is constructed. Based on the aforementioned set of multiple disease joint features, spatial distribution consistency analysis is performed on the disease features of adjacent sections at the same time node. By comparing the direction and magnitude of the changes in disease severity in adjacent sections, spatial consistency features reflecting the expansion or concentration of diseases along the road direction are constructed. By combining the correlation characteristics, time lag characteristics, and spatial consistency characteristics among diseases, a segment-level disease co-evolution characteristic is formed; The disease prediction module is used to construct a deep time series prediction model by inputting the segment-level disease quantitative characteristics, co-evolution characteristics, traffic load and meteorological data to obtain the disease development trend at a specified time scale in the future, and form segment-level disease evolution prediction results.

8. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, complete the method for extrapolating the evolution of apparent defects in asphalt pavement as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, complete the method for deducing the evolution of apparent defects in asphalt pavement as described in any one of claims 1-6.

Images