Roadbed rolling track and compaction quality dynamic correlation analysis method

Abstract

The invention relates to the technical field of big data analysis and processing, and discloses a roadbed rolling track and compaction quality dynamic correlation analysis method, which comprises the following steps of: 1, acquiring position information, attitude information and vibration response information of a road roller in a roadbed rolling process, obtaining a hidden state sample set and a corresponding partial trajectory analysis unit observation sequence sample set; 2, executing a combined dynamic correlation analysis algorithm based on a hidden Markov model HMM and manifold learning; and step 3, processing a real-time trajectory analysis unit observation sequence constructed by real-time acquisition data by using a joint dynamic correlation analysis model, and generating a dynamic compaction quality grade distribution map covering a roadbed construction area by combining a compaction manifold embedding coordinate. According to the method, the continuity and the stability of compaction quality evaluation can be improved, and the capability of responding to the change of the construction state can be enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing and analysis, and particularly relates to a method for dynamically correlating roadbed rolling track and compaction quality. BACKGROUND

[0002] In the process of road engineering construction, the compaction quality of roadbed directly affects the bearing capacity, deformation performance and long-term service state of the pavement layer, so accurate evaluation and dynamic monitoring of the compaction process during the construction phase has always been the focus of the engineering field. In order to obtain the compaction quality of the roadbed, the traditional method mainly relies on compaction quality detection devices for on-site sampling inspection, and the dry density or compaction degree is measured at a number of sampling points after construction is completed, so as to judge the compaction quality grade of the corresponding point. This kind of method is widely used in engineering practice, and the measurement results can be used as the basis for final quality acceptance, but due to the limited number of detection points, it is often difficult to reflect the overall quality state of the entire construction area. During the rolling process, the rolling speed, vibration amplitude, vibration frequency and differences in the coverage path of each pass of the road roller will all cause the compaction quality to be unevenly distributed in space, and it is difficult to accurately capture this change by relying on a small number of detection points.

[0003] In order to make up for the shortcomings of the traditional method, the industry has begun to introduce sensor data on the road roller to digitally record the construction process. For example, the driving track is obtained through the positioning device installed on the machine, the attitude change is obtained through the inertial measurement device, and the vibration acceleration or vibration energy is recorded through the vibration response sensor. These data reflect the running state of the road roller at each track point and can be used to infer the compaction process at different positions. The existing technology attempts to statistically correlate these operating parameters with compaction quality detection data, for example, by setting thresholds or establishing empirical formulas to map combinations of speed, vibration amplitude and vibration frequency to compaction quality grades. This kind of method can achieve the visualization of compaction quality to some extent, but since it relies on artificial setting rules or simple linear relationships, it is often difficult to adapt to complex and dynamically changing construction conditions, and the sensitivity to abnormal situations and complex working conditions is also weak. When the working state of the construction machinery changes, such as speed fluctuations, changes in vibration system energy output or changes in roadbed moisture content, the model based on fixed thresholds is prone to misjudgment. SUMMARY

[0004] The purpose of the present application is to provide a subgrade rolling track and compaction quality dynamic correlation analysis method, by constructing a track analysis unit observation sequence, a hidden state sample set and a compaction feature vector set, the rolling machine operation parameters, the compaction quality detection data and the compaction behavior characteristics are unified into the analysis process, the hidden Markov model training, the compaction manifold embedding space construction and the joint iterative correction of the two are completed in turn, so that the joint dynamic correlation analysis model which can reflect the time evolution law of the rolling process and the spatial similar structure of the compaction behavior at the same time is obtained. In real-time application, the present application uses the real-time observation symbol sequence to infer the final hidden state number, and forms the compaction quality area based on the compaction manifold embedding coordinates, and further generates the dynamic compaction quality grade distribution map covering the construction area. Therefore, the present application not only can improve the continuity and stability of the compaction quality evaluation, but also can enhance the response ability to the construction state change, realize the fine, real-time and full-area dynamic monitoring of the compaction process, and provide a more reliable basis for the operation path planning and rolling pass control.

[0005] To solve the above technical problems, the present application provides a subgrade rolling track and compaction quality dynamic correlation analysis method, the method comprising: Step 1: collecting the rolling machine position information, attitude information and vibration response information in the subgrade rolling process, and dividing the track analysis unit in the subgrade construction area, constructing the track analysis unit observation sequence containing the rolling machine operation parameters; obtaining the compaction quality detection data, and establishing the corresponding relationship between the compaction quality grade and the hidden state, to obtain the hidden state sample set and the corresponding partial track analysis unit observation sequence sample set; Step 2: executing the joint dynamic correlation analysis algorithm based on the hidden Markov model HMM and manifold learning, comprising: Step 2.1: using the hidden state sample set and the partial track analysis unit observation sequence sample set to initially train the hidden Markov model, to obtain the hidden Markov model HMM initial model; Step 2.2: constructing the compaction manifold embedding space based on the compaction feature vector of all track analysis units, to obtain the compaction manifold embedding coordinates of each track analysis unit; Step 2.3: jointly and iteratively correcting the hidden Markov model HMM initial model and the compaction manifold embedding space, to obtain the joint dynamic correlation analysis model containing the converged final hidden state number; Step 3: using the joint dynamic correlation analysis model to process the real-time track analysis unit observation sequence constructed by the real-time collection data, and combining the compaction manifold embedding coordinates, to generate the dynamic compaction quality grade distribution map covering the subgrade construction area.

[0006] Further, the process of constructing the track analysis unit observation sequence in step 1 specifically includes: collecting position information by using the positioning device, collecting attitude information by using the attitude detection device, and collecting vibration response information by using the vibration response collection device; dividing the subgrade construction area on the plane into a plurality of track analysis units with a side length of one meter, each track analysis unit being provided with a unique number; classifying the position information at each collection time into the corresponding track analysis unit according to the plane coordinates; setting a rolling pass number for each complete rolling return process, regarding the sampling points belonging to the same track analysis unit and having the same rolling pass number as the observation segment of the same rolling pass on the corresponding track analysis unit; for each track analysis unit, arranging the observation segments corresponding to each rolling pass on the corresponding track analysis unit in sequence according to the collection time sequence to form a track analysis unit observation sequence with a length of not less than five; and each observation segment containing the average driving speed, average vibration amplitude, average vibration frequency and marker information of the corresponding rolling pass on the corresponding track analysis unit.

[0007] Further, in the track analysis unit with compaction quality detection data in step 1, the compaction quality corresponding to the corresponding track analysis unit is divided into three compaction quality grades of insufficient compaction grade, normal compaction grade and high compaction grade according to the numerical range of the compaction quality detection data; and a one-to-one correspondence relationship is established between the three compaction quality grades and the insufficient compaction hidden state, the normal compaction hidden state and the high compaction hidden state, respectively, to obtain a hidden state sample set and a partial track analysis unit observation sequence sample set corresponding to the hidden state sample set.

[0008] Further, the process of initial training of the hidden Markov model in step 2.1 specifically includes: selecting no less than fifty trajectory analysis units with compaction quality detection data as training samples; for each training sample, dividing each observation segment in the observation sequence of the trajectory analysis unit into one of the three speed interval encodings according to the average driving speed, one of the three vibration amplitude interval encodings according to the average vibration amplitude, and one of the three frequency interval encodings according to the average vibration frequency, and splicing the three interval encodings into a discrete observation symbol in a fixed order to obtain an observation symbol sequence for the observation segments in the same trajectory analysis unit observation sequence; in each training sample, within the same trajectory analysis unit observation sequence, the number of transitions between adjacent hidden states is counted in the order of the hidden states, and for each combination of a starting hidden state and a target hidden state, the number of transitions of the combination is divided by the total number of transitions of the corresponding starting hidden state to obtain a hidden state transition probability table containing three rows and three columns, and the sum of the three values in each row of the table is equal to one after rounding to three decimal places; in all training samples, the first hidden state of each trajectory analysis unit observation sequence is taken as a starting hidden state, the number of occurrences of each hidden state as a starting hidden state is counted, and the number of occurrences of each hidden state is divided by the total number of training samples to obtain a hidden state initial probability table containing three rows and one column; in each training sample, for each hidden state, the number of occurrences of each discrete observation symbol under the corresponding hidden state is counted, and the number of occurrences of each discrete observation symbol is divided by the sum of the total number of occurrences of all observation symbols under the corresponding hidden state to obtain three observation output probability tables containing a number of columns.

[0009] Further, the process of initial training of the hidden Markov model further includes: taking the hidden state initial probability table, the hidden state transition probability table, and the observation output probability table as the initial parameters of the hidden Markov model HMM, and taking the observation symbol sequences of all training samples as input; in each iteration, first perform forward calculation and backward calculation on each observation symbol sequence in time sequence, the forward calculation sequentially calculates the forward probability of the three hidden states at each time, and the backward calculation sequentially calculates the backward probability of the three hidden states from the last observation time; then update the probability values in the hidden state initial probability table, the hidden state transition probability table, and the observation output probability table based on the forward probability and the backward probability, and after each update, all probability values are retained to three decimal places; stop iteration when the change of any value in the hidden state transition probability table in adjacent two iterations is less than zero zero zero one or the number of iterations reaches fifty, and obtain the initial model of the hidden Markov model HMM.

[0010] Further, the process of constructing the compact manifold embedding space in step 2.2 specifically comprises: for each trajectory analysis unit, calculating the average value of the average travel speed, the average value of the average vibration amplitude, the average value of the average vibration frequency, and the cumulative residence time of all observation segments in the trajectory analysis unit observation sequence, and then combining the four statistics in a fixed order to form a compact feature vector in the compact feature vector set; for each feature in the compact feature vector set, calculating the average value and the standard deviation of each feature in all trajectory analysis units, respectively, and then performing linear scaling processing on each feature of each trajectory analysis unit using the calculated average value and the standard deviation, so that the average value of each feature of all trajectory analysis units is zero and the standard deviation is one.

[0011] Further, the process of constructing the compact manifold embedding space further comprises: for any two trajectory analysis units, calculating the Euclidean distance according to the scaled values of the four features, and storing the Euclidean distances between all trajectory analysis units in a distance matrix; for each trajectory analysis unit, selecting the five nearest trajectory analysis units in the distance matrix as the neighborhood trajectory analysis units, and constructing a neighborhood relationship set containing all trajectory analysis units; using the shortest path algorithm on the neighborhood relationship set, calculating the shortest path distance between each pair of trajectory analysis units through the neighborhood relationship set, and storing all shortest path distances in a shortest path distance matrix; performing dimensionality reduction processing on the shortest path distance matrix, specifically, performing row mean and column mean correction on the shortest path distance matrix to obtain a symmetric matrix, then performing eigenvalue decomposition on the symmetric matrix, selecting the two eigenvectors corresponding to the two largest eigenvalues, and taking the components of each trajectory analysis unit in the two eigenvectors as the first coordinate and the second coordinate of the corresponding trajectory analysis unit in the compact manifold embedding space, respectively, to obtain the compact manifold embedding coordinates of each trajectory analysis unit.

[0012] Further, in step 2.3, the process of joint iterative correction of the hidden Markov model HMM and the compact manifold embedding space specifically comprises: for all obtained trajectory analysis units, using the initial model of the hidden Markov model HMM to perform Viterbi decoding on the observation symbol sequence of each trajectory analysis unit, specifically, calculating the initial scores of the three hidden states at the first observation time, then calculating the scores of the three hidden states at the current time at the second and subsequent observation times, and recording the path number of the hidden state with the maximum score from the previous time at each observation time, selecting the hidden state number with the maximum score at the last observation time as the terminal hidden state number of the observation sequence, and then backtracking along the recorded hidden state path number to obtain the hidden state number at each time, and taking the hidden state number at the last observation time as the final hidden state number of the corresponding trajectory analysis unit.

[0013] Further, the joint iterative correction process further comprises: initializing an iteration count value to one, in each iteration, for each trajectory analysis unit, calculating the Euclidean distance between the trajectory analysis unit and other trajectory analysis units in the compaction manifold embedding space, selecting five trajectory analysis units with the smallest Euclidean distance as the neighborhood trajectory analysis units of the corresponding trajectory analysis unit in the compaction manifold embedding space, counting the number of occurrences of three hidden state numbers in the final hidden state numbers of the five neighborhood trajectory analysis units, if the number of occurrences of a certain hidden state number is not less than three and greater than the number of occurrences of the hidden state number corresponding to the current final hidden state number of the corresponding trajectory analysis unit in the neighborhood trajectory analysis units, replacing the final hidden state number of the corresponding trajectory analysis unit with the hidden state number with the highest number of occurrences; when one iteration ends, calculating the ratio of the number of trajectory analysis units with final hidden state number replacement to the total number of trajectory analysis units, if the calculated ratio is less than zero point zero one, or the iteration count value reaches ten, stopping iteration, obtaining the converged final hidden state number of each trajectory analysis unit, and using the hidden Markov model HMM initial model used at this time, the converged final hidden state number and the compaction manifold embedding coordinates as the joint dynamic correlation analysis model.

[0014] Further, step 3 specifically comprises: in the process of subgrade compaction construction, real-time acquisition of new subgrade compaction trajectory data, attitude data and vibration response data, and construction of real-time trajectory analysis unit observation sequence; for each real-time observation symbol sequence of the trajectory analysis unit, using the hidden Markov model HMM in the joint dynamic correlation analysis model to perform Viterbi decoding to obtain the real-time final hidden state number; and using the compaction manifold embedding space in the joint dynamic correlation analysis model, dividing several trajectory analysis units with compaction manifold embedding coordinates distance less than a set distance threshold and the same final hidden state number into the same compaction quality region; corresponding the insufficient compaction hidden state, the normal compaction hidden state and the high compaction hidden state to the unqualified compaction quality grade, the qualified compaction quality grade and the excellent compaction quality grade respectively, and performing color display of each compaction quality grade on the plane according to the spatial position of the trajectory analysis unit, to generate a dynamic compaction quality grade distribution map covering the entire subgrade construction area; and outputting the dynamic compaction quality grade distribution map as the subgrade compaction trajectory and compaction quality dynamic correlation analysis result, for guiding the adjustment of the path of the road roller and the control of the compaction pass.

[0015] The roadbed rolling track and compaction quality dynamic correlation analysis method of the application has the following beneficial effects: the application simultaneously incorporates the compactor operating parameters, compaction quality detection data and compaction behavior characteristics into a unified analysis system by constructing the hidden state sample set, track analysis unit observation sequence sample set and compaction feature vector set, and establishes a joint dynamic correlation analysis model of the hidden Markov model and the compaction manifold embedding space on this basis, so that the judgment of the compaction quality can simultaneously rely on the compaction process evolution law in the time dimension and the compaction feature similarity relationship in the space dimension. The application uses the discrete observation symbols to reflect the combined changes of the speed, vibration amplitude and vibration frequency in the rolling pass, so that the hidden Markov model can accurately capture the state transition trend of the compaction stage; through the standardized compaction feature vector, the construction of the adjacency relation set and the calculation of the shortest path distance, the compaction manifold embedding coordinates generated by the application can retain the structure of the compaction behavior in the high-dimensional feature space, so that similar compaction behaviors naturally gather in the space. The joint iterative correction process uses the compaction manifold embedding coordinates to adjust the spatial consistency of the hidden state results, so that the local abnormal state can be automatically corrected, thereby significantly improving the continuity and stability of the compaction state judgment. In addition, in the real-time application stage, the application can quickly generate real-time track analysis unit observation sequences based on the real-time collected compactor operating parameters, and infer the real-time final hidden state number through the joint dynamic correlation analysis model, and then form a compaction quality region in combination with the compaction manifold embedding coordinates and the set distance threshold, so that the dynamic compaction quality grade distribution map can reflect the latest state of the construction site. Compared with the traditional method of relying only on detection point data or inferring the compaction quality based on a single model, the application can simultaneously realize the time sequence analysis, spatial aggregation and dynamic update of the compaction process, improve the comprehensiveness, accuracy and real-time performance of the compaction quality evaluation, and provide a more reliable reference basis for the compactor operation path adjustment, rolling pass control and regional quality management. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 The hidden Markov model hidden state transition probability matrix visualization schematic diagram provided for the embodiments of the application; Figure 2 The hidden Markov model Viterbi decoding optimal path schematic diagram provided for the embodiments of the application. DETAILED DESCRIPTION

[0017] To make the objectives, technical solutions and advantages of the embodiments of the application clearer, the technical solutions in the embodiments of the application will be described below in conjunction with the drawings in the embodiments of the application. Obviously, the described embodiments are only a part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application.

[0018] A subgrade rolling track and compaction quality dynamic correlation analysis method, the method comprising: Step 1: Collect the position information, attitude information and vibration response information of the roller during the subgrade rolling process, divide the track analysis unit in the subgrade construction area, and construct the track analysis unit observation sequence containing the roller operation parameters; obtain the compaction quality detection data, and establish the corresponding relationship between the compaction quality level and the hidden state to obtain the hidden state sample set and the corresponding partial track analysis unit observation sequence sample set.

[0019] In an embodiment, step 1 first installs a positioning device, an attitude detection device and a vibration response acquisition device on the roller respectively. The positioning device can be a global satellite positioning system supporting receiving device, and the sampling period can be set to 0.1 seconds, so that when the roller travels at a speed of about 1 meter per second, the displacement between adjacent sampling time points is about 0.1 meters. In this way, when the subsequent track analysis unit is divided with a side length of 1 meter, each track analysis unit will usually contain multiple sampling points, which is beneficial to the statistics of the average travel speed and other roller operation parameters. The attitude detection device can be an inertial measurement device, which is used to collect the attitude information such as the pitch angle and roll angle of the roller in the same sampling period. The vibration response acquisition device can be installed near the vibrating wheel to collect the vertical direction vibration acceleration data or vibration energy data, and the sampling period can be consistent with the positioning device, which is convenient for alignment on the time axis.

[0020] Before the subgrade construction, the subgrade construction area is divided into several track analysis units in the plane coordinate system according to the designed width and length of the subgrade. Preferably, the track analysis unit can be set as a square unit with a side length of 1 meter, each track analysis unit corresponds to a unique row number and column number in the coordinate plane, and each track analysis unit is further assigned a unique number. After converting the roller position information into plane coordinates, for each collection time, the plane coordinates of the collection time are compared with the boundary range of the track analysis unit to determine the track analysis unit into which the collection point falls, and the roller position information, attitude information and vibration response information of the collection time are classified into the corresponding track analysis unit. By this way of dividing according to the track analysis unit, the continuous rolling track can be discretized into units with clear boundaries in space, which is convenient for statistical analysis within each track analysis unit and for direct alignment of subsequent compaction quality detection data to the corresponding track analysis unit.

[0021] In construction organization, the round trip of the roller along the same route can be regarded as a complete rolling round trip process, and a rolling pass number is set for each complete rolling round trip process. For example, the rolling pass number of the first round trip process is 1, the rolling pass number of the second round trip process is 2, and so on. For the same trajectory analysis unit, all sampling points located in the trajectory analysis unit and having the same rolling pass number are screened out, and these sampling points correspond to the rolling process of the same rolling pass on the trajectory analysis unit. The position information, attitude information and vibration response information of these sampling points are statistically analyzed to obtain the average driving speed, average vibration amplitude and average vibration frequency of the rolling pass on the trajectory analysis unit. For example, in a trajectory analysis unit, if 20 sampling points are collected when the rolling pass number is 3, the driving distance in the corresponding time interval can be obtained by using the sequence of the 20 sampling points in time, so as to obtain the average driving speed on the trajectory analysis unit. At the same time, the vibration response information of the 20 sampling points is averaged to obtain the average vibration amplitude and the average vibration frequency. In order to align with the compaction quality detection data subsequently, mark information can also be added in each observation segment to identify whether there is compaction quality detection data in the rolling pass and the trajectory analysis unit.

[0022] For each trajectory analysis unit, the observation segments of each rolling pass belonging to the trajectory analysis unit are arranged in the order of collection time to form a trajectory analysis unit observation sequence. Preferably, in order to ensure that the trajectory analysis unit observation sequence can reflect the whole process of the evolution of the compaction quality with the rolling pass, the length of the trajectory analysis unit observation sequence is not less than 5, that is, at least 5 rolling passes on the same trajectory analysis unit are required to complete and collect the corresponding observation segments. The advantage of this length setting is that it can cover multiple stages from initial non-rolling, gradual compaction to near stable compaction state, so that the subsequent state recognition based on the trajectory analysis unit observation sequence can more completely reflect the dynamic correlation between the changes of the roller operating parameters and the compaction quality.

[0023] In terms of compaction quality detection, detection points can be arranged at predetermined intervals in the roadbed construction area, such as one detection point every 10 meters along the direction of travel and a row of detection points every 2 meters along the transverse direction. At each detection point, a compaction quality detection device is used to obtain compaction quality detection data, such as a dry density or compaction degree value. After converting the spatial position of each detection point to a planar coordinate, the planar coordinate is matched to the corresponding trajectory analysis unit, so that each piece of compaction quality detection data is associated with a unique trajectory analysis unit. In some embodiments, if a detection point is located near the boundary between two adjacent trajectory analysis units, the compaction quality detection data can be assigned to the trajectory analysis unit that is closer in distance to the geometric center of the adjacent trajectory analysis unit, in order to avoid the same detection data being used repeatedly.

[0024] To establish the correspondence between the compaction quality level and the hidden state, the compaction quality can be graded according to the numerical range of the compaction quality detection data. For example, in one embodiment, when the compaction quality detection data is less than ninety percent of the design target value, the compaction quality of the corresponding trajectory analysis unit is classified as an insufficiently compacted level; when the compaction quality detection data is between ninety percent of the design target value and one hundred and ten percent of the design target value, the compaction quality of the corresponding trajectory analysis unit is classified as a normally compacted level; and when the compaction quality detection data is greater than one hundred and ten percent of the design target value, the compaction quality of the corresponding trajectory analysis unit is classified as a highly compacted level. Through this grading method, continuous detection values can be converted into discrete compaction quality levels, facilitating the establishment of a one-to-one correspondence with discrete hidden states. Subsequently, the insufficiently compacted level, the normally compacted level, and the highly compacted level are respectively associated with the insufficiently compacted hidden state, the normally compacted hidden state, and the highly compacted hidden state to establish a fixed correspondence, thereby obtaining a hidden state sample set, and the trajectory analysis unit observation sequence of the trajectory analysis unit with compaction quality detection data is taken as the corresponding partial trajectory analysis unit observation sequence sample set. This establishment of a correspondence relationship enables the hidden state of the trajectory analysis unit to be inferred subsequently based only on the trajectory analysis unit observation sequence, thereby indirectly estimating the compaction quality level.

[0025] In another alternative embodiment, the side length of the track analysis unit can be set to 0.5 meters to improve spatial resolution. In this case, the sampling period of the roller position information can be shortened to 0.05 seconds to ensure that each track analysis unit still has a sufficient number of sampling points for statistical average driving speed and vibration response information. For the acquisition of compaction quality detection data, a random sampling method can also be used, in which not less than 30 track analysis units are randomly selected for compaction quality detection within the construction range of each work shift at the end of the shift, and the detection point positions are matched to the corresponding track analysis units in the same way. Whether regular arrangement or random sampling is used, as long as each track analysis unit with compaction quality detection data can establish a corresponding compaction quality level and hidden state, a hidden state sample set and a partial track analysis unit observation sequence sample set can be formed to provide a training basis for subsequent steps.

[0026] Step 2: Perform joint dynamic correlation analysis algorithm based on hidden Markov model HMM and manifold learning, including: Step 2.1: Use the hidden state sample set and the partial track analysis unit observation sequence sample set to perform initial training of the hidden Markov model, to obtain a hidden Markov model HMM initial model; Step 2.2: Construct a compaction manifold embedding space based on the compaction feature vectors of all track analysis units, to obtain a compaction manifold embedding coordinate of each track analysis unit; Step 2.3: Perform joint iterative correction of the hidden Markov model HMM initial model and the compaction manifold embedding space, to obtain a joint dynamic correlation analysis model containing a converged final hidden state number.

[0027] In one embodiment, after the construction of the hidden state sample set and the partial track analysis unit observation sequence sample set in step 1 is completed, a joint dynamic correlation analysis algorithm based on hidden Markov model HMM and manifold learning is performed. The overall idea of this algorithm is: first, use the hidden state sample set and the partial track analysis unit observation sequence sample set to train a hidden Markov model HMM initial model that can represent the time series evolution law of the compaction process; then, based on the compaction feature vectors of all track analysis units, construct a compaction manifold embedding space that maintains the geometric structural relationship of compaction behavior; finally, use feedback iteration between the hidden Markov model HMM initial model and the compaction manifold embedding space, so that the evolution of the hidden state on the same time sequence and the compaction behavior between adjacent track analysis units tend to be consistent, thereby obtaining a joint dynamic correlation analysis model containing a converged final hidden state number.

[0028] In the initial training process of the Hidden Markov Model (HMM), no less than 50 trajectory analysis units with compaction quality detection data are selected as training samples. The sample number is set to be more than 50 because the hidden state sample set contains three hidden states of insufficient compaction, normal compaction and high compaction. If the sample number is too small, the occurrence number of each hidden state is insufficient, which leads to large deviation of the transition relationship and observation output relationship statistically calculated, and is not conducive to the stable description of the dynamic change of the compaction quality. For each training sample, the continuous observation segment in the observation sequence of the trajectory analysis unit needs to be converted into a discrete observation symbol. Specifically, according to engineering experience and common range of historical construction data, the average driving speed can be divided into three speed interval encodings of low speed interval, medium speed interval and high speed interval, the average vibration amplitude can be divided into three vibration amplitude interval encodings of low vibration amplitude interval, medium vibration amplitude interval and high vibration amplitude interval, and the average vibration frequency can be divided into three frequency interval encodings of low frequency interval, medium frequency interval and high frequency interval. For example, the average driving speed less than or equal to 0.8 meters per second can be divided into a low speed interval, greater than 0.8 meters per second and less than or equal to 1.2 meters per second can be divided into a medium speed interval, and greater than 1.2 meters per second can be divided into a high speed interval. Similarly, the vibration amplitude and vibration frequency can be set according to the equipment specifications and test data. The reason for dividing into three intervals is that the sensitive zone of the compaction effect to the speed and vibration parameters can usually be divided into three categories of low, moderate and high, and using three-level discrete coding can not only retain the coarse-grained change trend related to the compaction effect, but also avoid too many intervals leading to too sparse distribution of training samples in each interval. The speed interval encoding, vibration amplitude interval encoding and vibration frequency interval encoding of each observation segment are spliced in a fixed order to form a discrete observation symbol, so as to convert each trajectory analysis unit observation sequence into a sequence of discrete observation symbols. For example, in an observation sequence of a trajectory analysis unit with a length of 8, 8 discrete observation symbols can be obtained.

[0029] On the basis of the discrete observation symbol sequence, the hidden state at the corresponding time can be directly determined at the observation position with the compaction quality detection data, in combination with the correspondence between the compaction quality grade and the hidden state established in step 1. Within the observation sequence of the same trajectory analysis unit, the combinations of adjacent hidden states are counted in time sequence, and the transition times from each starting hidden state to the three target hidden states are accumulated. For example, in a certain training sample, if the transition from the insufficiently compacted hidden state to the normally compacted hidden state occurs 5 times, and the transition from the insufficiently compacted hidden state to the highly compacted hidden state occurs only once, it can be seen that in this sample, the insufficiently compacted hidden state is more likely to evolve to the normally compacted hidden state. The transition times in all training samples are combined, and the transition times indexed by the starting hidden state in each row are normalized, i.e. the transition times from each starting hidden state to each target hidden state are divided by the sum of the transition times of the starting hidden state, to obtain a hidden state transition probability table containing 3 rows and 3 columns. The sum of the 3 values in each row in the table is equal to 1 after rounding to 3 digits after the decimal point, so that for any current hidden state, the change of the next hidden state can always be explained in 3 possible transition directions.

[0030] In order to determine the initial probability of the hidden state of the hidden Markov model HMM, the distribution of the 1st hidden state of each trajectory analysis unit observation sequence needs to be counted. In the training sample, the number of occurrences of each hidden state as the 1st hidden state is counted, and then the number of occurrences of each hidden state is divided by the total number of training samples to obtain a hidden state initial probability table containing 3 rows and 1 column. The advantage of this setting is that it can reflect the overall distribution tendency of the subgrade at the beginning of rolling, for example, the insufficiently compacted hidden state at the beginning of most road sections, and the normally compacted hidden state at the beginning of the rework road section. Further, in each training sample, for each hidden state, the number of occurrences of each discrete observation symbol under the hidden state is counted, and each hidden state is normalized, i.e. the number of occurrences of each discrete observation symbol under the hidden state is divided by the sum of the total number of observation symbols under the hidden state, to obtain 3 observation output probability tables containing a number of columns. The observation output probability table depicts the probability distribution of the combination of the average travel speed, average vibration amplitude and average vibration frequency of the road roller under different hidden states, thereby establishing the correspondence between the combination of the road roller operating parameters and the compaction quality hidden state.

[0031] After constructing the initial probability table of hidden states, the probability table of hidden state transitions, and the probability table of observations and outputs, these are used as the initial parameters of the Hidden Markov Model (HMM). The observed symbol sequences of all training samples are used as input to perform an iterative training process. In each iteration, for each observed symbol sequence, forward computation is performed in chronological order, starting from the first observation time and calculating the forward probability of being in one of the three hidden states at each time step. Intuitively, the forward probability represents the likelihood of a certain hidden state being the current state, given the prefix of the current observed symbol sequence. Correspondingly, backward computation is performed starting from the last observation time step, calculating the backward probability of being in one of the three hidden states at each time step and subsequently generating the remaining observed symbol sequences. After both forward and backward probabilities are calculated, the expected number of occurrences of each hidden state at each time step and the expected number of occurrences of each hidden state transition combination at adjacent time steps can be estimated based on the probability product relationship. These expected values ​​are accumulated and normalized over all training samples to update the values ​​in the initial hidden state probability table, the hidden state transition probability table, and the observation output probability table. After each update, all probability values ​​are retained to three decimal places to avoid excessive accumulation of tail differences. Through repeated forward and backward calculations, the Hidden Markov Model (HMM) gradually converges to a parameter combination that better reflects the statistical regularity of the training data. Iteration stops when the change in any value in the hidden state transition probability table is less than 0.001 in two consecutive iterations, or when the number of iterations reaches 50. The initial hidden state probability table, the hidden state transition probability table, and the observation output probability table at this point are used as the initial model of the HMM. In another optional implementation, the allowable change threshold can be set to 0.0005 or the maximum number of iterations can be set to 100 to accommodate construction scenarios with larger data volumes or higher convergence requirements.

[0032] In one implementation, Figure 1 This section presents the hidden state transition probability matrix A in the initial Hidden Markov Model (HMM) obtained through step 2.1 training. This 3x3 square matrix characterizes the probability distribution of the evolution of the compaction mass hidden state from the current time step to the next time step during roadbed compaction. Specifically, the row index of the hidden state transition probability matrix A represents the initial hidden state at the current time step, and the column index represents the target hidden state at the next time step. Each element aij in the matrix represents the probability that when the compaction mass is in the i-th hidden state in the current compaction pass, it will transition to the j-th hidden state in the next compaction pass. In this embodiment, the three hidden states are, in order, the insufficiently compacted hidden state, the normally compacted hidden state, and the high-compaction hidden state, corresponding to the 1st row / column, 2nd row / column, and 3rd row / column of the matrix. Figure 1In the shown embodiment, the element a11=0.152 in the first row and the first column of the matrix represents the probability that a certain trajectory analysis unit remains in the insufficiently compacted hidden state in the next rolling pass when it is in the insufficiently compacted hidden state in the current rolling pass. This value is small, indicating that the insufficiently compacted state has a strong instability and tends to transfer to other states. The element a12=0.765 in the first row and the second column of the matrix represents the probability that the insufficiently compacted hidden state transfers to the normally compacted hidden state, which is the largest value in this row, indicating that the most likely evolution direction of the insufficiently compacted state after additional rolling is to reach the normally compacted state. The element a13=0.083 in the first row and the third column of the matrix represents the probability that the insufficiently compacted hidden state directly jumps to the highly compacted hidden state, which is only 8.3%, which is reasonable because the improvement of compacted quality is usually a gradual process, and it is rare to directly reach high compaction from insufficient compaction. The second row of the matrix represents the transfer law of the normally compacted hidden state. The element a21=0.042 represents the probability of degradation from the normally compacted state to the insufficiently compacted state, which is only 4.2%, indicating that once the normally compacted state is reached, due to the compaction memory effect of the roadbed material, even if the subsequent rolling parameters fluctuate slightly, it is difficult to significantly reduce the compaction quality. The element a22=0.687 represents the probability of self-maintenance of the normally compacted state, which is 68.7%, indicating that the normally compacted state has strong stability and has a high probability of remaining in the current state in subsequent rolling. The element a23=0.271 represents the probability of improving from the normally compacted state to the highly compacted state, which is 27.1%, indicating that on the basis of normal compaction, through additional rolling passes or optimization of rolling parameters, there is a high probability of further improving the compaction quality to the excellent level. The third row of the matrix shows the transfer characteristics of the highly compacted hidden state. The element a31=0.018 represents the probability of degradation from the highly compacted state to the insufficiently compacted state, which is only 1.8%, indicating that the highly compacted state has strong stability and it is almost impossible to have a significant quality decrease in subsequent rolling. The element a32=0.235 represents the probability of reducing from the highly compacted state to the normally compacted state, which is 23.5%, which is a relatively moderate probability, which may correspond to the slight loosening phenomenon caused by continuing to apply high-intensity vibration in the already over-compacted area. The element a33=0.747 represents the probability of self-maintenance of the highly compacted state, which is as high as 74.7%, which is the largest value among all diagonal elements, indicating that once the roadbed reaches the highly compacted state, its internal structure has formed a stable dense skeleton, and in subsequent rolling, it can maintain excellent compaction quality for a long time. In order to intuitively show the size difference of the transfer probabilities, Figure 1The matrix elements are visualized by gray filling, where the darker the gray, the larger the probability value. It can be observed that the gray of the three positions of the first row and the second column, the second row and the second column, and the third row and the third column is the darkest, corresponding to the probability values of 0.765, 0.687, and 0.747, respectively. These three elements represent the three most important state evolution paths of the transition from insufficient compaction to normal compaction, the self-maintenance of normal compaction, and the self-maintenance of high compaction, respectively. In contrast, the elements representing quality degradation in the matrix (such as a21=0.042 and a31=0.018) have very light gray, indicating that the probability of significant deterioration of compaction quality under normal construction conditions is extremely low. In Figure 1 Below, the four main transition directions and their corresponding transition probabilities are further labeled. The probability of "insufficient→normal" transition is 0.765, which is the highest among all transitions, indicating that increasing the number of rolling passes can effectively improve the compaction quality during construction. The probability of "normal→normal" transition is 0.687, indicating that the normal compaction state can be maintained stably. The probability of "high compaction→high compaction" transition is 0.747, showing that the high compaction state has high stability. The probability of "normal→high compaction" transition is 0.271, indicating that there is still a large space for quality improvement based on normal compaction. The construction of the hidden state transition probability matrix is based on the training process described in step 2.1, and is obtained by statistical analysis of not less than 50 trajectory analysis units with compaction quality detection data. The sum of each row element in the matrix is equal to 1 (under the condition of rounding to three decimal places and rounding), satisfying the normalization condition of probability distribution. This matrix, together with the observed output probability table and the hidden state initial probability table, will constitute the core parameters of the initial model of the hidden Markov model (HMM) in the subsequent steps, which is used to perform Viterbi decoding on the observed sequence of real-time trajectory analysis units, thereby inferring the compaction quality hidden state of each trajectory analysis unit at the current rolling stage.

[0033] In constructing the compaction manifold embedding space, it is necessary to first construct the compaction feature vector based on all trajectory analysis units. For each trajectory analysis unit, the average driving speed of all observation segments in its trajectory analysis unit observation sequence is averaged to obtain the average driving speed average value; the average vibration amplitude of all observation segments is averaged to obtain the average vibration amplitude average value; the average vibration frequency of all observation segments is averaged to obtain the average vibration frequency average value; and the cumulative residence time of the road roller in the trajectory analysis unit is counted. The four statistics are combined into a compaction feature vector in a fixed order. The reason for selecting these four statistics is that the average driving speed average value, the average vibration amplitude average value and the average vibration frequency average value reflect the typical operating state of the trajectory analysis unit during the long-term rolling process, and the cumulative residence time reflects the cumulative action intensity of the trajectory analysis unit. Combining the four can more comprehensively summarize the compaction behavior characteristics of the trajectory analysis unit.

[0034] In order to eliminate the differences in the dimensions and numerical ranges of different features, and avoid a certain feature value range being too large and dominating in subsequent distance calculation, it is necessary to standardize the compaction feature vector set. Specifically, for each feature, the average value and standard deviation of the feature in all trajectory analysis units are calculated, and the standard deviation can be obtained by squaring the difference between each trajectory analysis unit and the average value of the feature, and then taking the square root. Then, using the average value and the standard deviation, the corresponding feature of each trajectory analysis unit is linearly scaled, so that the average value of the scaled feature in all trajectory analysis units is close to 0, and the standard deviation is close to 1. After standardization, the four features are on the same order of magnitude in numerical scale, which is conducive to fairly measuring the contribution of each feature when calculating the Euclidean distance in the subsequent calculation.

[0035] After obtaining the set of normalized compaction feature vectors, for any two trajectory analysis units, the Euclidean distance is calculated according to the scaled values of the four features, and the Euclidean distances between all trajectory analysis units are stored in a distance matrix. The distance matrix reflects the similarity relationship between each trajectory analysis unit in the compaction feature space. In order to capture the local structure, on each trajectory analysis unit, the five different trajectory analysis units with the smallest distance from the trajectory analysis unit are selected from the distance matrix as the neighborhood trajectory analysis units of the trajectory analysis unit, and the Euclidean distances between the trajectory analysis unit and the five neighborhood trajectory analysis units are recorded to construct a neighborhood relationship set containing all trajectory analysis units. Selecting five neighborhood trajectory analysis units is the result of compromise. Too few neighborhood numbers will lead to disconnected neighborhood relationship sets, and too many will blur the local structure, so that slight differences are also considered as strong similarity. On the neighborhood relationship set, the shortest path algorithm is used. For each pair of trajectory analysis units, the shortest path distance through the neighborhood relationship set is calculated by finding the path in the neighborhood relationship set and accumulating the Euclidean distances on the path, and all shortest path distances are stored in a shortest path distance matrix. Compared with directly using the Euclidean distance, the shortest path distance not only considers the direct difference between two trajectory analysis units, but also considers the indirect difference through multiple intermediate trajectory analysis units, which can better reflect the overall geometric structure of the entire compaction feature space.

[0036] After obtaining the shortest path distance matrix, dimensionality reduction processing is performed on the matrix. Specifically, the mean value can be calculated for the shortest path distance matrix by row and by column respectively, and each element in the matrix is subtracted by the corresponding row mean value and column mean value, and then added by the overall mean value to obtain a symmetric matrix. Perform eigenvalue decomposition on the symmetric matrix to obtain a set of eigenvalues and corresponding eigenvectors. Select the two eigenvectors corresponding to the two largest eigenvalues, and the components of each trajectory analysis unit in the two eigenvectors are taken as the first coordinate and the second coordinate of the trajectory analysis unit in the compaction manifold embedding space, thereby obtaining the compaction manifold embedding coordinates of each trajectory analysis unit. The compaction manifold embedding space constructed in this way can preserve the relative position relationship between the trajectory analysis units in the original high-dimensional compaction feature space as much as possible in the two-dimensional space. Intuitively, trajectory analysis units with similar compaction behaviors will be close to each other in the compaction manifold embedding space, while trajectory analysis units with large differences in compaction behavior will be far away from each other, which is conducive to subsequent neighborhood statistics and hidden state correction in the space. In another optional implementation, three eigenvectors corresponding to the three largest eigenvalues can be selected to construct a three-dimensional compaction manifold embedding space, and the compaction behavior distribution can be displayed through three-dimensional interaction when more detailed visualization is needed.

[0037] After obtaining the initial HMM model and the compact manifold embedding space, the two are jointly iteratively corrected to obtain a joint dynamic correlation analysis model containing the final hidden state number. The joint iterative correction process first uses the initial HMM model to perform Viterbi decoding on the observation symbol sequence of each trajectory analysis unit obtained in step 1. Specifically, for each observation symbol sequence of the trajectory analysis unit, at the first observation time, the initial scores of the three hidden states are calculated respectively. The initial score can be understood as the degree of matching between the hidden state and the current observation symbol after considering the initial probability of the hidden state and the observation output probability. At the second and subsequent observation times, the scores of the three hidden states at the current time are calculated in turn. The score can be obtained by traversing the scores of the three hidden states at the previous time, the corresponding hidden state transition probability and the observation output probability corresponding to the current observation symbol, and recording the hidden state number of the previous time corresponding to each current hidden state score in the calculation process, so that the complete hidden state sequence can be recovered from the back to the front along the path with the maximum score. At the last observation time, the hidden state number with the maximum score is selected as the terminal hidden state number of the observation symbol sequence, and then the hidden state number is backtracked in reverse along the recorded path number to obtain the hidden state number at each observation time. The hidden state number at the last observation time is taken as the final hidden state number of the corresponding trajectory analysis unit, thereby obtaining the initial final hidden state number set for all trajectory analysis units.

[0038] After the initial final hidden state number is assigned to each trajectory analysis unit, the joint iterative correction process is started. The iteration count is initialized as 1. In each iteration, for each trajectory analysis unit, the Euclidean distances between the trajectory analysis unit and other trajectory analysis units in the compaction manifold embedding space are calculated, and the five trajectory analysis units with the smallest Euclidean distances are selected as the neighborhood trajectory analysis units of the trajectory analysis unit in the compaction manifold embedding space. Then the occurrence numbers of the three hidden state numbers of insufficient compaction hidden state, normal compaction hidden state and high compaction hidden state in the final hidden state numbers of the five neighborhood trajectory analysis units are counted. If the occurrence number of a certain hidden state number in the neighborhood trajectory analysis units is not less than 3 and is greater than the occurrence number of the current final hidden state number of the trajectory analysis unit in the neighborhood trajectory analysis units, the final hidden state number of the trajectory analysis unit is replaced by the hidden state number with the largest occurrence number in the neighborhood. The reason for such setting is that when a trajectory analysis unit is surrounded by multiple neighborhood trajectory analysis units with the same hidden state number in the compaction manifold embedding space, it indicates that the trajectory analysis unit is very close to the neighborhood trajectory analysis units in the compaction characteristics, and therefore adjusting the final hidden state number of the trajectory analysis unit to be consistent with the neighborhood mainstream hidden state is helpful to eliminate abnormal hidden state determination caused by local observation noise or individual abnormal passes, so that the final hidden state number is more continuous and smooth in spatial distribution. After the iteration is completed, the number of trajectory analysis units whose final hidden state numbers are replaced in the iteration is counted, and the ratio of the number to the total number of trajectory analysis units is calculated. If the ratio is less than 0.01, it indicates that the trajectory analysis units that need to be corrected in the current iteration are very few, and the overall hidden state distribution tends to be stable, and the iteration can be terminated; if the iteration count value has reached 10, the iteration is also terminated to limit the calculation overhead. Otherwise, the iteration count value is increased by 1, and the next round of iteration is continued. After the above iteration, when the termination condition is met, the converged final hidden state number of each trajectory analysis unit is obtained. At this time, the initial hidden Markov model HMM, the compaction manifold embedding space and the converged final hidden state number consistent with the compaction manifold embedding coordinates jointly constitute the joint dynamic correlation analysis model. In another optional implementation, the number of neighborhood trajectory analysis units can be adjusted to 7, the condition that the occurrence number is not less than 3 can be adjusted to not less than 4, or the ratio threshold can be adjusted from 0.01 to 0.02, so as to enhance the spatial consistency constraint in the scene where the road level is higher and the compaction quality requirement is stricter.

[0039] Step 3: Using the joint dynamic correlation analysis model, the real-time trajectory analysis unit observation sequence constructed by the real-time collected data is processed, and the dynamic compaction quality level distribution map covering the roadbed construction area is generated in combination with the compaction manifold embedding coordinates.

[0040] In an embodiment, after the training of the joint dynamic relational analysis model is completed, the joint dynamic relational analysis model is deployed in the road roller on-board computing device or the field edge computing device for real-time generation of a dynamic compaction quality level distribution map covering the subgrade construction area during the subgrade compaction construction process. To this end, new subgrade compaction track data, attitude data and vibration response data need to be continuously collected during the construction process, and real-time track analysis unit observation sequences are constructed in the same manner as before, and then the joint dynamic relational analysis model is used to process the real-time track analysis unit observation sequences, and the compaction quality area is divided and graphically displayed in combination with the compaction manifold embedding coordinates.

[0041] During the construction process, the positioning device, the attitude detection device and the vibration response acquisition device continue to work at a preset sampling period, for example, the sampling period is still 0.1 seconds. For each sampling, the position information, attitude information and vibration response information of the road roller are recorded, and the position information is converted into plane coordinates and is attributed to the corresponding track analysis unit. In order to keep consistent with the statistical characteristics in the joint dynamic relational analysis model, for each track analysis unit, the sampling points are aggregated by compaction pass number within the current construction time window (for example, within the last 5 minutes), the average driving speed, the average vibration amplitude and the average vibration frequency of each compaction pass on the track analysis unit are calculated, and the real-time track analysis unit observation sequence containing multiple observation segments is obtained. In order to ensure that the real-time track analysis unit observation sequence is sufficient to reflect the current compaction process, preferably, when the length of the real-time track analysis unit observation sequence of a certain track analysis unit is less than 3, the hidden state inference of the track analysis unit can be temporarily not performed, and after the track analysis unit accumulates 3 or more observation segments, the joint dynamic relational analysis model is processed. In this way, unstable hidden state judgment caused by too few observation segments can be avoided.

[0042] After the real-time track analysis unit observation sequence is constructed, each observation segment needs to be converted into a discrete observation symbol in the same way as in the offline training phase. Specifically, for each observation segment, according to the interval to which the average driving speed, the average vibration amplitude and the average vibration frequency belong, it is mapped into the corresponding speed interval code, vibration amplitude interval code and frequency interval code, for example, the average driving speed not greater than 0.8 meters per second is mapped into the low speed interval code, between 0.8 meters per second and 1.2 meters per second is mapped into the medium speed interval code, and greater than 1.2 meters per second is mapped into the high speed interval code; then the three interval codes are spliced into a discrete observation symbol in a fixed order, thereby forming a real-time observation symbol sequence. Since the initial model of the hidden Markov model HMM in the joint dynamic relational analysis model is trained under the same coding manner, the Viterbi decoding can be directly performed on the real-time observation symbol sequence.

[0043] For each real-time observation symbol sequence of the trajectory analysis unit, Viterbi decoding is performed by using the Hidden Markov Model (HMM) in the joint dynamic association analysis model. Specifically, at the first observation time, initial scores of the insufficiently compacted hidden state, the normally compacted hidden state and the highly compacted hidden state are respectively calculated, and the initial scores comprehensively consider the initial probability of the hidden state and the observation output probability, and are used to measure the matching degree of the combination of the hidden state and the current observation symbol. At the second and subsequent observation times, the current score of each hidden state is calculated in turn, and the current score is derived from the combination and superposition between the scores of the three hidden states at the previous time and the hidden state transition probability, and is multiplied by the observation output probability corresponding to the current observation symbol. In order to be able to restore the complete hidden state path, the hidden state number at the previous time that leads to the maximum current score is recorded at each observation time. At the last observation time, the hidden state number with the maximum score is selected as the terminal hidden state number of the real-time observation symbol sequence, and then the hidden state number is backtracked along the recorded hidden state path number to obtain the hidden state number at each observation time. Since the dynamic compactness quality grade distribution map focuses on the comprehensive compactness state of each trajectory analysis unit at the current time, the hidden state number at the last observation time can be taken as the real-time final hidden state number of the trajectory analysis unit.

[0044] In an embodiment, Figure 2 The complete process of performing Viterbi decoding algorithm on the observation symbol sequence of a certain trajectory analysis unit by using the Hidden Markov Model (HMM) is shown, so as to determine the optimal hidden state path of the trajectory analysis unit. The observation symbol sequence in the example has a length of 6, which corresponds to the observation fragments collected by the trajectory analysis unit in 6 continuous rolling passes, and is recorded as O1, O2, O3, O4, O5 and O6. The goal of Viterbi decoding is to find a path that best matches the observation symbol sequence from all possible hidden state sequences, so that the joint probability reaches the maximum.

[0045] Specifically, Figure 2 The calculation process of Viterbi decoding is shown by using a grid structure. In the figure, the horizontal direction represents the time dimension, and from left to right, t=1 to t=6 are six observation times, which correspond to six rolling passes. The vertical direction represents the hidden state dimension, and from top to bottom, the insufficiently compacted hidden state, the normally compacted hidden state and the highly compacted hidden state are three possible states. At the intersection of each time t and each hidden state i, a circular node is drawn, and the name of the hidden state is marked in the node. All nodes form a grid of 3 rows and 6 columns, i.e. 18 nodes.

[0046] At t=1, Viterbi algorithm first calculates the initial scores of the three hidden states by using the initial probability of the hidden state and the observation output probability corresponding to the first observation symbol O1. At t=2, the current scores of the three hidden states are calculated by using the initial scores of the three hidden states at t=1, the hidden state transition probability and the observation output probability corresponding to the second observation symbol O2. Figure 2In this example, the initial score of the under-compacted hidden state at t = 1 is δ1(1) = 0.32, the initial score of the normal-compacted hidden state is δ1(2) = 0.58, and the initial score of the over-compacted hidden state is δ1(3) = 0.10. These three scores indicate that, when the first set of operating parameters O1 is observed, the trajectory analysis unit is most likely in the normal-compacted state (with the highest score of 0.58), followed by the under-compacted state (with a score of 0.32), and least likely in the over-compacted state (with a score of only 0.10). This score distribution reflects the fact that the subgrade is usually in the normal-compacted or under-compacted state during the initial rolling stage, and rarely reaches the over-compacted state after the first rolling pass.

[0047] At t = 2, the algorithm needs to calculate the scores of the three hidden states at t = 2 based on the scores at t = 1, the hidden state transition probabilities, and the observation output probability of the current observation symbol O2. For any hidden state j at t = 2, the algorithm iterates through all three hidden states i at t = 1, and calculates the path score of the transition from state i to state j, which is equal to δ1(i) multiplied by the transition probability aij and multiplied by the observation output probability bj(O2). Among the three possible transition paths, the path with the highest score is selected as the optimal predecessor path leading to state j, and the predecessor state number corresponding to this path is recorded. In Figure 2 In this example, since all possible state transition paths are represented by thin gray lines, and the optimal path is marked by a thick black line, it can be clearly seen that, in the transition from t = 1 to t = 2, the optimal path transitions from the under-compacted state at t = 1 (with a score of 0.32) to the normal-compacted state at t = 2 (with a score of 0.72). This transition is consistent with the transition probability matrix shown in FIG. 6, in which the transition probability from the under-compacted state to the normal-compacted state is 0.765, which is the largest element in this row. Figure 2 The scores of the three states at t = 2 are δ2(1) = 0.15, δ2(2) = 0.72, and δ2(3) = 0.13, respectively. The score of the normal-compacted state is significantly higher than those of the other two states, indicating that, after two rolling passes, the compaction quality of the trajectory analysis unit has improved to a normal level.

[0048] At t = 3, the algorithm continues to perform the same calculation process. From Figure 2As can be seen, the optimal path continues from the normal compaction state at t=2 to the normal compaction state at t=3, with a corresponding score of δ3(2)=0.81. This score further increases, indicating that as the number of compaction passes increases, the model's confidence in judging that the trajectory analysis unit is in a normal compaction state continuously strengthens. At this moment, the score for the insufficiently compacted state has dropped to δ3(1)=0.08, and the score for the high-compacted state is δ3(3)=0.11, both of which are much lower than the score for the normal compaction state, indicating that the trajectory analysis unit stably maintains a normal compaction state during the third compaction pass.

[0049] At time t=4, the observation symbol O4 indicates a change in the combination of operating parameters of the roller. According to the calculation results, the score of the normal compaction state is δ4(2)=0.68, which is still the highest among the three states, but the value has decreased compared to 0.81 at time t=3. At the same time, the score of the high compaction state rises to δ4(3)=0.27, and the score of the insufficiently compacted state further decreases to δ4(1)=0.05. This trend indicates that the trajectory analysis unit is still in the normal compaction state during the fourth compaction, but the observed combination of operating parameters has begun to approach the parameter characteristics corresponding to the high compaction state, laying the groundwork for the transition to the high compaction state at the next time step. Figure 2 The optimal path from t=3 to t=4 remains in a normal compaction state.

[0050] At time t=5, the compaction quality underwent a significant evolution. The score for the high-compaction state jumped to δ5(3)=0.55, exceeding the score for the normal compaction state δ5(2)=0.43 for the first time, becoming the highest-scoring latent state at that time. The score for the insufficiently compacted state dropped to an extremely low δ5(1)=0.02, indicating that this state could be largely ruled out. Figure 2 In the process, the optimal path shifts from the normal compaction state at t=4 to the high compaction state at t=5. This shift corresponds to... Figure 2 The transition probability a23 = 0.271 indicates that, based on normal compaction, optimizing rolling parameters or increasing the number of rolling passes has a high probability of further improving the compaction quality to an excellent level. This transition path reflects the typical evolutionary law of gradual improvement in compaction quality during roadbed compaction.

[0051] At time t=6, the last observation time, the score for the high-compaction state further increased to δ6(3)=0.74, far higher than the score for the normal-compaction state δ6(2)=0.25 and the score for the insufficiently compacted state δ6(1)=0.01. The optimal path continued from the high-compaction state at t=5 to the high-compaction state at t=6, corresponding to Figure 2The transition probability a33=0.747 in the table indicates that the high-compaction state has a strong self-maintenance ability. In the last step of the Viterbi decoding, the algorithm selects the hidden state with the highest score at t=6, i.e., the high-compaction state, as the terminating hidden state.

[0052] After determining the terminating hidden state, the Viterbi algorithm backtracks along the recorded optimal predecessor path indices in reverse order. Starting from the high-compaction state at t=6, the optimal predecessor of this state at t=5 is found, resulting in the high-compaction state at t=5. Continuing the backtracking, the normal-compaction state at t=4, the normal-compaction state at t=3, the normal-compaction state at t=2, and the insufficiently-compaction state at t=1 are obtained. Arranging these six hidden states in chronological order, the complete optimal hidden state sequence is obtained: insufficiently-compaction→normal-compaction→normal-compaction→normal-compaction→high-compaction→high-compaction. This sequence is represented by the thick black line in the table, clearly showing the complete process of the trajectory analysis unit in which the compaction quality evolves from the initial insufficiently-compaction state to the high-compaction state. Figure 2

[0053] ​ In the legend at the bottom, the thick black solid line represents the Viterbi optimal path, i.e., the path with the highest probability among all possible hidden state sequences. The gray thin line represents other possible but non-optimal transition paths. At each time instant, there are theoretically 3x3=9 possible transition paths (from the 3 states at the previous time instant to the 3 states at the current time instant), but the Viterbi algorithm reduces the exponential path search complexity to linear complexity by retaining only one optimal predecessor path at each state node through dynamic programming strategy. In this embodiment, there are 5 state transitions from t=1 to t=6, each with 9 possibilities, resulting in a total of 9^5=59049 possible hidden state sequence combinations. The Viterbi algorithm can efficiently find the globally optimal hidden state sequence through local optimal selection and dynamic programming recording at each time instant, without the need to enumerate all possible combinations.

[0054] ​After obtaining the real-time final hidden state number, the trajectory analysis units need to be spatially clustered and divided into multiple compaction quality regions in the compaction manifold embedding space of the joint dynamic correlation analysis model. The compaction manifold embedding coordinates of each trajectory analysis unit have been pre-calculated in the joint dynamic correlation analysis model, and the compaction manifold embedding coordinates have the following properties: trajectory analysis units with similar compaction behaviors are close to each other in the compaction manifold embedding space, and trajectory analysis units with large differences in compaction behavior are far away from each other in the compaction manifold embedding space. Therefore, in real-time applications, these compaction manifold embedding coordinates can be directly used for compaction quality region division. Specifically, in a round of update period (for example, update every 10 seconds or every 30 seconds), for all trajectory analysis units that have obtained real-time final hidden state numbers, the Euclidean distance between each pair of trajectory analysis units is calculated according to the compaction manifold embedding coordinates, and for each trajectory analysis unit, other trajectory analysis units with a Euclidean distance less than a set distance threshold are found. The set distance threshold can be pre-adjusted according to the coordinate scale of the compaction manifold embedding space, for example, when the numerical range of the first coordinate and the second coordinate of the compaction manifold embedding coordinate is approximately between -3 and 3, the set distance threshold can be set to 0.8 or 1.0. A small distance threshold will result in more and finer compaction quality region division, but may result in adjacent trajectory analysis units being divided into different regions; a large distance threshold will result in more trajectory analysis units being combined into the same compaction quality region, but may mask some local differences. By comparing the compaction quality region division results under different distance thresholds in the trial operation stage, a set distance threshold that can reflect the construction quality differences without excessive splitting can be selected.

[0055] In the specific division of the compaction quality region, not only is the distance of the compaction manifold embedding coordinates required to be less than the set distance threshold, but the final hidden state numbers of the trajectory analysis units are also required to be the same. The reason for this is that the compaction manifold embedding coordinates mainly reflect the overall similarity of long-term compaction behavior, while the real-time final hidden state number reflects the state of the current compaction phase. If clustering is performed only according to the compaction manifold embedding coordinates, trajectory analysis units in different compaction phases may be divided into the same region; if division is performed only according to the real-time final hidden state number, the cumulative differences in compaction behavior may be ignored. By requiring both the compaction manifold embedding coordinate distance to be less than the set distance threshold and the final hidden state number to be the same, it can be ensured that the trajectory analysis units in the same compaction quality region have high consistency in both historical behavior and current state, making them more suitable as basic units for construction control.

[0056] After the formation of the compaction quality regions, it is necessary to map the hidden state information into an easily understandable compaction quality level. In the joint dynamic association analysis model, the insufficiently compacted hidden state, the normally compacted hidden state and the highly compacted hidden state have been respectively corresponded to the unqualified compaction quality level, the qualified compaction quality level and the excellent compaction quality level. Therefore, for each compaction quality region, the compaction quality level of the region can be directly determined according to the final hidden state number of the track analysis units in the region. For example, when the final hidden state number of all track analysis units in a compaction quality region is the number corresponding to the insufficiently compacted hidden state, the compaction quality region is marked as the unqualified compaction quality level; when it is the number corresponding to the normally compacted hidden state, it is marked as the qualified compaction quality level; when it is the number corresponding to the highly compacted hidden state, it is marked as the excellent compaction quality level. In actual application, individual track analysis units in a compaction quality region may have a final hidden state number inconsistent with the majority of track analysis units due to data loss or occasional noise. The compaction quality level can be determined according to the final hidden state number of the majority of track analysis units in the region, so as to maintain the consistency within the region.

[0057] When generating the dynamic compaction quality level distribution map, it is necessary to map the compaction quality level information to the planar position of the roadbed construction region. Since each track analysis unit has a unique spatial position in the roadbed planar coordinate system, a grid consistent with the division of the track analysis units can be constructed on the display terminal, and each track analysis unit is represented as a small rectangle or polygon on the screen. According to the compaction quality region to which the track analysis unit belongs and its compaction quality level, the small rectangle or polygon is filled with the corresponding color, for example, the unqualified compaction quality level is set to red, the qualified compaction quality level is set to yellow, and the excellent compaction quality level is set to green. For track analysis units that have not completed compaction or lack real-time track analysis unit observation sequences, gray can be used for representation. In this way, a dynamic compaction quality level distribution map covering the entire roadbed construction region can be generated on the display terminal, and construction personnel can intuitively see which regions have reached the excellent compaction quality level, which regions are only in the qualified compaction quality level, and which regions are still in the unqualified compaction quality level.

[0058] In order to reflect the dynamic characteristics, the dynamic compaction quality grade distribution map can be updated at preset time intervals. For example, every 10 seconds or every 30 seconds, a batch processing is performed on the current collected real-time trajectory analysis unit observation sequence, the real-time final hidden state number of each trajectory analysis unit is recalculated, the compaction quality area is re-divided, and the color distribution on the display terminal is updated. Since the displacement of the road roller within 10 seconds or 30 seconds is usually within the range of 10 meters to 30 meters, each update introduces a batch of new observation segments, so that the dynamic compaction quality grade distribution map can better reflect the current construction progress. In some embodiments, a certain number of historical layers can be reserved for different time slices, for example, 10 dynamic compaction quality grade distribution maps generated in the last 5 minutes, so that the construction management personnel can compare the compaction quality change trend at different times and identify the quality fluctuations caused by factors such as water content change and equipment state change.

[0059] When the dynamic compaction quality grade distribution map is output as the roadbed rolling track and dynamic correlation analysis result of compaction quality, statistical information related thereto can also be output, such as the total area of unqualified compaction quality grade area, the total area of qualified compaction quality grade area, the total area of excellent compaction quality grade area, and the proportion of each compaction quality grade area in the entire construction area. The construction management personnel can guide the road roller operation path adjustment and rolling pass control according to these statistical information and graphical distribution. For example, for the unqualified compaction quality grade area displayed in red in the dynamic compaction quality grade distribution map, the road roller can be arranged to pass through this area for additional rolling in the next round of construction; for the qualified compaction quality grade area displayed in yellow, if the construction specification requires to reach the excellent compaction quality grade, one or two additional rolling can be reasonably arranged according to the area and equipment capacity; for the excellent compaction quality grade area displayed in green, unnecessary repeated rolling can be reduced, thereby reducing fuel consumption and equipment wear. In another optional embodiment, the real-time position and travel track of the road roller can also be superimposed on the dynamic compaction quality grade distribution map, so that the driver can directly observe the compaction quality grade distribution of the front section of the road through the display terminal in the cab, and spontaneously adjust the path according to the current position of the road roller, thereby improving the operation efficiency.

[0060] The above has introduced the present application in detail. The principles and embodiments of the present application are described by applying specific examples in this paper, and the above examples are only used to help understand the method of the present application and its core idea. It should be noted that for ordinary skilled personnel in the technical field, some improvements and modifications can be made to the present application without departing from the principles of the present application, and these improvements and modifications also fall within the protection scope of the claims of the present application.

Claims

1. A method for dynamically correlating subgrade rolling track and compaction quality, characterized in that, The method comprises: Step 1: collecting position information, attitude information and vibration response information of the road roller in the roadbed rolling process, dividing track analysis units in the roadbed construction area, and constructing a track analysis unit observation sequence containing the running parameters of the road roller; obtaining compaction quality detection data, and establishing a corresponding relationship between the compaction quality grade and the hidden state to obtain a hidden state sample set and a corresponding partial track analysis unit observation sequence sample set; Step 2: performing a joint dynamic correlation analysis algorithm based on a hidden Markov model (HMM) and manifold learning, comprising: Step 2.1: using the hidden state sample set and the partial track analysis unit observation sequence sample set to perform initial training on the hidden Markov model to obtain a hidden Markov model (HMM) initial model; Step 2.2: constructing a compaction manifold embedding space based on the compaction feature vectors of all track analysis units to obtain a compaction manifold embedding coordinate of each track analysis unit; Step 2.3: jointly and iteratively correcting the hidden Markov model (HMM) initial model and the compaction manifold embedding space to obtain a joint dynamic correlation analysis model containing a converged final hidden state number; Step 3: using the joint dynamic correlation analysis model to process a real-time track analysis unit observation sequence constructed from real-time collection data, and combining the compaction manifold embedding coordinate to generate a dynamic compaction quality grade distribution map covering the roadbed construction area.

2. The method of claim 1, wherein, The process of constructing the track analysis unit observation sequence in step 1 specifically comprises: collecting position information using a positioning device, collecting attitude information using an attitude detection device, and collecting vibration response information using a vibration response collection device; dividing the roadbed construction area into a plurality of track analysis units with a side length of one meter on a plane, and setting a unique number for each track analysis unit; assigning each collection time position information to the corresponding track analysis unit according to the plane coordinates; setting a rolling pass number for each complete rolling return process, and regarding sampling points belonging to the same track analysis unit and having the same rolling pass number as observation segments of the same rolling pass on the corresponding track analysis unit; for each track analysis unit, arranging the observation segments corresponding to each rolling pass on the corresponding track analysis unit in sequence according to the collection time sequence to form a track analysis unit observation sequence with a length of not less than five; each observation segment contains the average driving speed, average vibration amplitude, average vibration frequency and marker information of the corresponding rolling pass on the corresponding track analysis unit.

3. The method according to claim 1 or 2, characterized in that, In step 1, in the track analysis units with compaction quality detection data, the compaction quality corresponding to the corresponding track analysis unit is divided into three compaction quality grades of insufficient compaction, normal compaction and high compaction according to the numerical range of the compaction quality detection data; and a one-to-one correspondence is established between the three compaction quality grades and the insufficient compaction hidden state, the normal compaction hidden state and the high compaction hidden state to obtain a hidden state sample set and a partial track analysis unit observation sequence sample set corresponding to the hidden state sample set.

4. The method of claim 1, wherein, The process of initial training of the hidden Markov model in step 2.1 specifically includes: selecting no less than fifty trajectory analysis units with compaction quality detection data as training samples; for each training sample, dividing each observation segment in the observation sequence of the trajectory analysis unit into one of the three speed interval encodings according to the average driving speed, one of the three vibration amplitude interval encodings according to the average vibration amplitude, and one of the three frequency interval encodings according to the average vibration frequency, and splicing the three interval encodings into a discrete observation symbol in a fixed order to obtain an observation symbol sequence for the observation segments in the same trajectory analysis unit observation sequence; in each training sample, within the same trajectory analysis unit observation sequence, the number of transitions between adjacent hidden states is counted in the order of the hidden states, and for each combination of a starting hidden state and a target hidden state, the number of transitions of the combination is divided by the total number of transitions of the corresponding starting hidden state to obtain a hidden state transition probability table containing three rows and three columns, and the sum of the three values in each row of the table is equal to one after rounding to three decimal places; in all training samples, the first hidden state of each trajectory analysis unit observation sequence is taken as a starting hidden state, the number of occurrences of each hidden state as a starting hidden state is counted, and the number of occurrences of each hidden state is divided by the total number of training samples to obtain a hidden state initial probability table containing three rows and one column; in each training sample, for each hidden state, the number of occurrences of each discrete observation symbol under the corresponding hidden state is counted, and the number of occurrences of each discrete observation symbol is divided by the sum of the total number of occurrences of all observation symbols under the corresponding hidden state to obtain three observation output probability tables containing a number of columns.

5. The method of claim 4, wherein, The process of initial training of the hidden Markov model further includes: taking the hidden state initial probability table, the hidden state transition probability table, and the observation output probability table as the initial parameters of the hidden Markov model HMM, and taking the observation symbol sequences of all training samples as input; in each iteration, first perform forward calculation and backward calculation on each observation symbol sequence in time sequence, the forward calculation sequentially calculates the forward probability of the three hidden states at each time, and the backward calculation sequentially calculates the backward probability of the three hidden states from the last observation time; then update the probability values in the hidden state initial probability table, the hidden state transition probability table, and the observation output probability table based on the forward probability and the backward probability, and after each update, all probability values are retained to three decimal places; stop iteration when the change of any value in the hidden state transition probability table in adjacent two iterations is less than zero zero zero one or the number of iterations reaches fifty, and obtain the initial model of the hidden Markov model HMM.

6. The method of claim 1, wherein, The process of constructing the compact manifold embedding space in step 2.2 specifically includes: for each trajectory analysis unit, calculating the average value of the average travel speed, the average value of the average vibration amplitude, the average value of the average vibration frequency, and the cumulative residence time of all observation segments in the trajectory analysis unit observation sequence, and combining the four statistics in a fixed order to form a compact feature vector in the compact feature vector set; for each feature in the compact feature vector set, calculating the average value and standard deviation of each feature in all trajectory analysis units, and then performing linear scaling processing on each feature of each trajectory analysis unit using the calculated average value and standard deviation, so that the average value of each feature of all trajectory analysis units is zero and the standard deviation is one.

7. The method of claim 6, wherein, The process of constructing the compact manifold embedding space further includes: for any two trajectory analysis units, calculating the Euclidean distance according to the scaled values of the four features, and storing the Euclidean distances between all trajectory analysis units in a distance matrix; for each trajectory analysis unit, selecting the five different trajectory analysis units with the smallest distance in the distance matrix as the neighborhood trajectory analysis units, and constructing a neighborhood relationship set containing all trajectory analysis units; using the shortest path algorithm on the neighborhood relationship set, calculating the shortest path distance through the neighborhood relationship set for each pair of trajectory analysis units, and storing all shortest path distances in a shortest path distance matrix; performing dimensionality reduction processing on the shortest path distance matrix, specifically, performing row mean and column mean correction on the shortest path distance matrix to obtain a symmetric matrix, then performing eigenvalue decomposition on the symmetric matrix, selecting the two eigenvectors corresponding to the two largest eigenvalues, and taking the components of each trajectory analysis unit in the two eigenvectors as the first coordinate and the second coordinate of the corresponding trajectory analysis unit in the compact manifold embedding space, respectively, to obtain the compact manifold embedding coordinates of each trajectory analysis unit.

8. The method of claim 1, wherein, In step 2.3, the process of joint iterative correction of the hidden Markov model HMM and the compact manifold embedding space specifically includes: for all obtained trajectory analysis units, using the initial model of the hidden Markov model HMM to perform Viterbi decoding on the observation symbol sequence of each trajectory analysis unit, specifically, calculating the initial scores of three hidden states at the first observation time, then calculating the scores of three hidden states at the current time at the second and subsequent observation times, and recording the path number of the hidden state with the maximum score from the previous time at each observation time, selecting the hidden state number with the maximum score at the last observation time as the termination hidden state number of the observation sequence, and then backtracking along the recorded hidden state path number to obtain the hidden state number at each time, and taking the hidden state number at the last observation time as the final hidden state number of the corresponding trajectory analysis unit.

9. The method of claim 8, wherein, The joint iterative correction process further comprises: initializing an iteration count value as one, in each iteration, for each trajectory analysis unit, calculating the Euclidean distance between the trajectory analysis unit and other trajectory analysis units in the compact manifold embedding space, selecting five trajectory analysis units with the smallest Euclidean distance as the neighborhood trajectory analysis units of the corresponding trajectory analysis unit in the compact manifold embedding space, counting the number of occurrences of three hidden state numbers in the final hidden state numbers of the five neighborhood trajectory analysis units, and if the number of occurrences of a certain hidden state number is not less than three and greater than the number of occurrences of the hidden state number corresponding to the current final hidden state number of the corresponding trajectory analysis unit in the neighborhood trajectory analysis units, replacing the final hidden state number of the corresponding trajectory analysis unit with the hidden state number with the highest number of occurrences; when one iteration ends, calculating the ratio of the number of trajectory analysis units with final hidden state number replacement to the total number of trajectory analysis units, if the calculated ratio is less than zero point zero one, or the iteration count value reaches ten, stopping iteration, obtaining the converged final hidden state number of each trajectory analysis unit, and using the hidden Markov model HMM initial model used at this time, the converged final hidden state number and the compact manifold embedding coordinates as the joint dynamic correlation analysis model.

10. The method of claim 1, wherein, Step 3 specifically comprises: in the process of subgrade rolling construction, real-time acquisition of new subgrade rolling trajectory data, attitude data and vibration response data, and construction of real-time trajectory analysis unit observation sequence; for each real-time observation symbol sequence of the trajectory analysis unit, using the hidden Markov model HMM in the joint dynamic correlation analysis model to perform Viterbi decoding to obtain a real-time final hidden state number; and using the compact manifold embedding space in the joint dynamic correlation analysis model, dividing several trajectory analysis units with a distance between compact manifold embedding coordinates less than a set distance threshold and the same final hidden state number into the same compactness quality region; corresponding the insufficiently compacted hidden state, the normally compacted hidden state and the high compacted hidden state to the unqualified compactness quality level, the qualified compactness quality level and the excellent compactness quality level respectively, and performing color display of each compactness quality level on a plane according to the spatial position of the trajectory analysis unit to generate a dynamic compactness quality level distribution map covering the entire subgrade construction area; and outputting the dynamic compactness quality level distribution map as the joint dynamic correlation analysis result of the subgrade rolling trajectory and the compactness quality, for guiding the adjustment of the path of the road roller and the control of the rolling pass.

Images