User request real-time interaction method oriented to massive road surface monitoring data sharing platform

Abstract

The invention discloses a user request real-time interaction method for a mass road surface monitoring data sharing platform, and belongs to the field of roads. The recognition efficiency of a traditional response recognition method cannot meet the real-time processing requirement of platform data. A section of continuous data subset is selected from the road surface monitoring historical data set and divided into multiple sections of data sets, and each section of data set comprises a preset number of pieces of data; identifying whether each data group contains a dynamic response fragment, if so, determining that the data group is a label 1, and if not, determining that the data group is a label 0, and taking the data group as a training sample; model training: training a plurality of prediction models by using a training sample, and screening one candidate prediction model from the plurality of trained prediction models; and after the sharing platform receives data sent by a user and downloads the data subset of a certain time period, the data subset of the certain time period is subjected to data segmentation and then input into the candidate prediction model, and a data group with the result of 1 is sent to the user. The method is used for identifying the dynamic response segment.

Description

Technical Field

[0001] The present invention belongs to the field of roads. Background Art

[0002] With the continuous development of economic technology, road construction has been continuously promoted and improved, and the scale of road infrastructure has become increasingly large. It is necessary to deploy a large number of stress and strain sensing devices on traffic infrastructure such as roads, bridges, and tunnels to monitor the service performance of roads in real time, and through big data analysis, construct a basic theoretical system for infrastructure performance evaluation and design with characteristics of climate, environment, hydrology, and geology, providing basic data and R & D support for engineering structure safety, maintenance scientific decision-making, etc. Currently, to obtain the transient response information of the road structure under the action of high-speed vehicle loads, high-frequency sampling of dynamic responses is required, and the acquisition frequency of stress and strain sensors can usually be as high as 2000 Hz. This means that in the long-term observation process, the amount of data collected by the sensor per month will be as high as about 6 TB. Facing the massive road surface monitoring data, how to perform efficient data management and sharing is one of the challenges faced by the current road monitoring field.

[0003] The establishment of a data sharing platform is crucial for the management and maintenance of road infrastructure. By sharing and analyzing massive road surface monitoring data, government agencies, researchers, and industry experts can better understand the service performance, expected life of the road surface, and formulate effective maintenance strategies. Among them, taking the two typical data sharing platforms of LTPP and FAA as examples, they not only provide valuable data support for the construction and maintenance of roads and airport runways, but also promote the development of related research on the sharing of traffic infrastructure performance monitoring data globally. For the long-term performance scientific observation network of traffic infrastructure, each observation site has also built a corresponding monitoring data sharing platform to provide data sharing services such as online data query and data application download for platform users.

[0004] When users use the data services of the massive road surface monitoring data sharing platform, users initially screen the platform data through relevant attribute information such as the corresponding time period of the data and the type of data acquisition sensor. Among the massive monitoring data, the road structure dynamic response data plays an important role in the evaluation of road service status, disease prediction, and maintenance decision-making, and is also an important indicator for users to screen data during the process of data access and data application download. Therefore, in the process of user requests, obtaining the dynamic response situation in the data selected by the user in real time and performing real-time interaction of this information with the user is an important way to improve the efficiency of the platform data service.

[0005] However, the recognition efficiency of traditional response recognition methods cannot meet the requirements of real-time processing of platform data, and existing machine learning recognition methods are also not well applicable to unlabeled high-dimensional monitoring data. Therefore, how to ensure the accuracy of recognizing dynamic responses while improving the recognition speed to meet the requirements of real-time interaction on the platform has become a key issue in realizing real-time interaction of user requests. Summary of the Invention

[0006] The purpose of the present invention is to solve the problem that the recognition efficiency of traditional response recognition methods cannot meet the requirements of real-time processing of platform data, and a method for real-time interaction of user requests for a massive road surface monitoring data sharing platform is proposed.

[0007] A method for real-time interaction of user requests for a massive road surface monitoring data sharing platform, the method includes the following:

[0008] Step 1, data segmentation: Select a continuous data subset from the historical road surface monitoring data set, and segment the subset into multiple data groups, each data group including a preset number of data;

[0009] Data annotation: Use the method of manual annotation to judge whether each data group contains a dynamic response segment. If so, assign the corresponding data group the label 1. If not, assign the corresponding data group the label 0, and use all the labeled data groups as training samples;

[0010] Step 2, model training: Use the training samples to train multiple recognition models to obtain multiple trained recognition models, and select 1 candidate recognition model from the multiple trained recognition models;

[0011] Step 3, after the sharing platform receives the data sent by the user and downloads the data subset of a certain time period, segment the data subset of the certain time period. The multiple data groups formed after segmentation are input into the candidate recognition model, and the result of each data group being 1 or 0 is output, and the information of the data group with the result of 1 is fed back to the user.

[0012] Preferably, in step 2, the specific process of selecting 1 candidate prediction model from the multiple trained prediction models is:

[0013] Select a trained recognition model with good recognition effect and high recognition efficiency from the multiple trained recognition models as the candidate recognition model.

[0014] Preferably, between data annotation and model training, it further includes: performing dimensionality reduction processing on the training samples, and using the training samples after dimensionality reduction processing for model training.

[0015] Preferably, the specific process of dimensionality reduction processing is as follows: calculate various statistical indicators of each data group, calculate the grey correlation degree between each statistical indicator and the corresponding label, sort the various grey correlation degrees of each data group obtained from largest to smallest, select the statistical indicators corresponding to the top pre-number of grey correlation degrees from the sorted various grey correlation degrees of each data group, and use the selected statistical indicators and corresponding labels of each data group as the training samples after dimensionality reduction processing.

[0016] Preferably, the various statistical indicators include mean value, maximum value and peak value.

[0017] Preferably, in step 1, before using all the data groups after dimensionality reduction processing as training samples, it further includes:

[0018] Adjust the ratio of the data groups with label 0 and label 1 to 1:1, and use the adjusted data groups as training samples.

[0019] Preferably, in step 3, after data segmentation of the data in a certain time period, it further includes: performing dimensionality reduction processing on the segmented multiple data groups, and inputting the multiple data groups after dimensionality reduction into the candidate recognition model.

[0020] The beneficial effects of the present invention are:

[0021] The present invention performs data segmentation, data annotation, data dimensionality reduction and model training on historical data to generate a candidate recognition model. When the data sharing platform receives the application data selected by the user, the application data includes a data subset of a certain time period to be downloaded. Perform data segmentation and dimensionality reduction on the data subset of the certain time period, input the multiple data groups formed after dimensionality reduction into the candidate recognition model, output the result of 1 or 0 for each data group, and feed back the relevant information of the data groups with the result of 1 to the user in real time. Among them, the data groups with the result of 1 contain dynamic response characteristics. Therefore, this application can feed back the data with dynamic response characteristics to the user in real time and automatically filter out the data without dynamic response characteristics. Therefore, the user can judge the demand for the data according to the dynamic response situation included in the selected data and decide whether to apply for downloading the data, thus ensuring the data application efficiency of the data sharing platform.

[0022] The present invention performs grey correlation analysis on the statistical indicators of the segmented multiple data groups and the manually annotated label 1 or 0, which helps to quickly perform dimensionality reduction of the data groups, thereby reducing the data processing pressure and ensuring the efficiency of model training and recognition. Therefore, the present invention has a high efficiency in identifying dynamic response characteristics. Description of the Drawings

[0023] Figure 1 It is a flowchart of the interaction between the data sharing platform and the user;

[0024] Figure 2It is a specific flowchart of the real-time interaction method for user requests of the massive road surface monitoring data sharing platform. Detailed implementation manners

[0025] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0026] It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments may be combined with each other.

[0027] Next, the present invention will be further described in conjunction with the accompanying drawings and specific embodiments, but it is not a limitation of the present invention.

[0028] Embodiment:

[0029] A real-time interaction method for user requests of the massive road surface monitoring data sharing platform, the method includes the following contents:

[0030] Step 1, data segmentation: Select a continuous data subset from the historical road surface monitoring data set, and segment the subset into multiple data groups, each data group includes a preset number of data;

[0031] Data annotation: Use the method of manual annotation to judge whether each data group contains a dynamic response segment. If so, assign the corresponding data group the label 1. If not, assign the corresponding data group the label 0. All the data groups with assigned labels are used as training samples;

[0032] Step 2, model training: Use the training samples to train multiple recognition models to obtain multiple trained recognition models, and select 1 candidate recognition model from the multiple trained recognition models;

[0033] Step 3, after the shared platform receives the data sent by the user to download the data subset of a certain time period, segment the data subset of the certain time period. The multiple data groups formed after segmentation are input into the candidate recognition model, and the result of each data group being 1 or 0 is output, and the information of the data group with the result of 1 is fed back to the user.

[0034] Specifically, if the data segment requested by the user to access is D, since the start and end times of the data segment are determined by the user himself, the length of the data segment cannot be determined. Every 300 pieces of data in the data segment are segmented into a file d i, and finally the data with less than 300 pieces is a file. The corresponding data in the existing data download records are segmented and manually labeled, and the files containing the dynamic response fragments are labeled "1", and the files without the dynamic response fragments are labeled "0". The machine learning model is trained using the manually labeled data. When the user enters the start and end time of the data segment, the data segment is segmented in real time and the data segment is classified to determine the dynamic response in the data segment.

[0035] The recognition results can also be displayed visually to achieve dynamic responsive user interaction.

[0036] Based on the output results, the dynamic response in the data group is displayed to the user, assisting the user in applying for data download.

[0037] Sensors can be used to collect data in real time and store it in a shared platform; a dynamic response segment refers to data generated when a vehicle passes by a sensor buried in the road surface, and a segment not containing dynamic response refers to data generated when no vehicle passes by a sensor buried in the road surface.

[0038] Further defined, in step 2, the specific process of selecting one candidate prediction model from the plurality of trained prediction models is as follows:

[0039] A trained recognition model with good recognition effect and high recognition efficiency is selected from the multiple trained recognition models as a candidate recognition model.

[0040] Specifically, the multiple recognition models in step 2 include three traditional models and five machine learning models. The recognition accuracy and recognition time of different models are compared, and the recognition model with the best recognition effect and fast recognition speed is selected as the candidate recognition model. The best recognition effect refers to comparing the recognition results with the preset results. The model with the largest number of results consistent with the preset results is called the best effect.

[0041] It is further defined that in step 3, after data segmentation is performed on the data of the certain time period, the step further includes: performing dimensionality reduction processing on the segmented multiple data groups, and inputting the reduced dimensionality multiple data groups into the candidate recognition model.

[0042] It is further defined that the specific process of dimensionality reduction processing is: calculating multiple statistical indicators for each data group, calculating the grey correlation between each statistical indicator and the corresponding label, sorting the multiple grey correlations obtained for each data group from large to small, selecting statistical indicators corresponding to a predetermined number of grey correlations from the multiple grey correlations of each data group after sorting, and using the selected statistical indicators and corresponding labels for each data group as training samples after dimensionality reduction processing.

[0043] Specifically, the data group after dimensionality reduction only contains the effective features for dynamic response identification of the original data group.

[0044] Further defined, the multiple statistical indicators include mean value, maximum value and peak value.

[0045] Further defined, in step 1, before using all the data groups after dimensionality reduction as training samples, it further includes:

[0046] Adjust the ratio of the data groups with label 0 and label 1 to 1:1, and use the adjusted data groups as training samples. Further defined, in step 3, after splitting the data of the certain time period, it further includes: performing dimensionality reduction on the multiple split data groups, and inputting the dimensionality-reduced multiple data groups into the candidate recognition model.

[0047] Specifically, the core idea of this embodiment is to perform real-time dynamic response identification on the data selected by the user through steps such as data segment splitting, data feature dimensionality reduction, and model recognition, and visually display the recognition results to achieve real-time information interaction.

[0048] This embodiment first performs equal-length splitting on the data subset according to the existing user request data subset, and performs manual annotation of the dynamic response on the split data groups. Since the data groups containing dynamic response data account for about 1 / 200 of all the split data groups, to ensure the recognition accuracy of the dynamic response identifier (used to set the recognition model internally), during the model training process, the ratio of positive examples (data groups containing dynamic response data) to negative examples (data groups not containing dynamic response data) is adjusted to 1:1, and the adjustment method is deletion, deleting multiple data groups without dynamic response data, so that the number of data groups containing dynamic response data is the same as the number of data groups without dynamic response data.

[0049] Although the present invention has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the present invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed, as long as they do not deviate from the spirit and scope of the present invention as defined by the appended claims. It should be understood that different dependent claims and the features described herein can be combined in a manner different from that described in the original claims. It should also be understood that the features described in connection with a single embodiment can be used in other described embodiments.

Claims

1. A real-time interaction method for user requests of a massive road surface monitoring data sharing platform, characterized in that, The method includes the following steps: Step 1, data segmentation: Select a continuous data subset from the historical road surface monitoring dataset, and segment the subset into multiple data groups, each data group including a preset number of data; Data annotation: Use the method of manual annotation to determine whether each data group contains a dynamic response segment. If so, assign the corresponding data group the label 1. If not, assign the corresponding data group the label 0. All the labeled data groups are used as training samples; Step 2, model training: Use the training samples to train multiple recognition models to obtain multiple trained recognition models, and select 1 candidate recognition model from the multiple trained recognition models; Step 3, after the shared platform receives the data sent by the user and downloads the data subset for a certain period of time, perform data segmentation on the data subset for the certain period of time. The multiple data groups formed after segmentation are input into the candidate recognition model, and the result of each data group being 1 or 0 is output, and the information of the data groups with the result of 1 is fed back to the user.

2. The real-time interaction method for user requests of the massive road surface monitoring data sharing platform according to claim 1, wherein In step 2, the specific process of selecting 1 candidate prediction model from the multiple trained prediction models is: Select a trained recognition model with good recognition effect and high recognition efficiency from the multiple trained recognition models as the candidate recognition model.

3. The real-time interaction method for user requests of the massive pavement monitoring data sharing platform according to claim 1, characterized in that, Between data annotation and model training, it also includes: performing dimensionality reduction processing on the training samples, and using the training samples after dimensionality reduction processing for model training.

4. The real-time interaction method for user requests of the massive road surface monitoring data sharing platform according to claim 3, characterized in that, The specific process of dimensionality reduction processing is: Calculate multiple statistical indicators of each data group, calculate the grey correlation degree between each statistical indicator and the corresponding label, sort the multiple grey correlation degrees of each data group obtained from largest to smallest, select the statistical indicators corresponding to the top pre-number of grey correlation degrees from the sorted multiple grey correlation degrees of each data group, and use the selected statistical indicators and corresponding labels of each data group as the training samples after dimensionality reduction processing.

5. The real-time interaction method for user requests of a massive pavement monitoring data sharing platform according to claim 4, characterized in that, The multiple statistical indicators include mean value, maximum value, and peak value.

6. The real-time interaction method for user requests of a massive road surface monitoring data sharing platform according to claim 4 or 5, characterized in that In step 1, before using all the data groups after dimensionality reduction processing as training samples, it also includes: Adjust the ratio of the data groups with label 0 and label 1 to 1:1, and use the adjusted data groups as training samples.

7. The real-time interaction method for user requests of the massive pavement monitoring data sharing platform according to claim 1, characterized in that, In step 3, after performing data segmentation on the data for the certain period of time, it also includes: performing dimensionality reduction processing on the multiple data groups after segmentation, and inputting the multiple data groups after dimensionality reduction into the candidate recognition model.

Images