Traffic detection data quality dynamic evaluation method and system based on meteorological-flow decoupling

Abstract

The invention relates to the technical field of traffic data detection, in particular to a traffic detection data quality dynamic evaluation method and system based on meteorological-flow decoupling, and the method comprises the steps that a data synchronous collection module collects data and outputs a multi-modal data stream; extracting a depth feature vector from the data stream through an independent feature representation module; performing causal decoupling through a weather-flow causal decoupling module, and outputting potential characterization; interaction feature representation is generated through a cross-domain feature interaction analysis module; the data comprehensive quality evaluation module generates a data quality evaluation vector and a comprehensive data quality score based on the potential characterization and the interactive feature characterization. The method has the beneficial effects that the independent contributions of the weather and the traffic flow to the data quality can be clearly separated. When the data quality is reduced, whether the dominant factor is severe weather or abnormal traffic flow can be clearly indicated, an accurate decision basis is provided for data restoration and sensor calibration or maintenance, and the defects of traditional black box assessment are overcome.

Description

Technical content

[0001] This invention belongs to the field of traffic data detection technology, specifically a dynamic evaluation method and system for traffic detection data quality based on meteorological-traffic flow decoupling. Background Technology

[0002] With the rapid development of intelligent transportation systems, traffic detection data serves as the core foundation supporting road network status perception, signal control optimization, and travel guidance decision-making. Its quality and reliability directly impact the accuracy and robustness of upper-layer applications. Current mainstream data quality assessment methods are mostly based on static thresholds, historical statistics, or simple anomaly detection rules, which struggle to effectively address the complex interference caused by the dynamic coupling of meteorological conditions (such as rain, snow, fog, and strong light) and traffic flow (such as peak congestion, sudden accidents, and holiday tidal flows) in real-world road environments. Such interference often leads to systematic biases, packet loss, or false detections in the output data of detection equipment. Furthermore, traditional assessment models, lacking decoupled modeling of meteorological and traffic flow factors, cannot distinguish whether data distortion originates from environmental noise or actual traffic condition changes, resulting in misjudgments or missed detections.

[0003] Therefore, there is an urgent need for a technical solution that can dynamically decouple the effects of weather and traffic flow and thereby achieve accurate quantitative assessment of data quality, so as to support the reliable operation and intelligent decision-making of intelligent transportation systems in complex environments. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for dynamic evaluation of traffic detection data quality based on meteorological-traffic flow decoupling, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a dynamic evaluation method for traffic detection data quality based on meteorological-traffic flow decoupling, comprising the following steps:

[0006] Traffic flow time-series data and environmental meteorological data are acquired, and the traffic flow time-series data and environmental meteorological data are timestamped and normalized to form a synchronous multimodal input data sequence; the multimodal input data sequence is input to an independent feature representation module to generate traffic flow depth feature vector and meteorological condition depth feature vector respectively.

[0007] The traffic flow deep feature vector and the meteorological condition deep feature vector are concatenated into a fused feature vector, which is then input into the meteorological-traffic causal decoupling module. This module performs structural decoupling based on a constructed causal graph model. Through causal intervention, it decouples the mutually independent potential representations of traffic flow impact and meteorological condition impact from the fused feature vector. These potential representations are then input into a cross-domain feature interaction analysis module to generate an interactive feature representation that quantifies the combined impact of meteorology and traffic flow. Finally, the potential representations of traffic flow impact, meteorological condition impact, and the interactive feature representation are concatenated into a comprehensive feature vector, which is input into a data quality comprehensive evaluation module to generate a data quality evaluation vector and a comprehensive data quality score. The data quality evaluation vector includes numerical indicators representing data completeness, accuracy, and timeliness.

[0008] As a further aspect of the present invention, the traffic flow time series data includes cross-sectional traffic flow, average vehicle speed, and lane time occupancy rate collected by roadside detectors; the environmental meteorological data includes rainfall intensity, atmospheric visibility, ambient temperature, wind speed, and road surface condition parameters; the timestamp alignment processing adopts an interpolation algorithm to adjust the sampling frequency of the environmental meteorological data to be consistent with the sampling frequency of the traffic flow time series data.

[0009] As a further embodiment of the present invention, the independent feature representation module includes a temporal convolutional network encoder and a multilayer perceptron encoder; the temporal convolutional network encoder is used to process the traffic flow temporal data, and it includes causal convolutional layers with exponentially growing inflation coefficients; the multilayer perceptron encoder is used to process the environmental meteorological data, and it includes multiple fully connected hidden layers, with the number of neurons decreasing layer by layer, and the activation function adopting a scaling exponential linear unit.

[0010] As a further aspect of the present invention, the meteorological-traffic flow causal decoupling module achieves decoupling in the following way: constructing a causal graph model of meteorological conditions (W), traffic flow (T), detection data (D), and data quality (Q);

[0011] Based on the causal graph model, intervention operations on traffic flow variables and meteorological condition variables are simulated respectively. A first condition generation network learns the mapping from meteorological condition variables to observed data variables under fixed traffic flow variables, and extracts potential representations of the impact of meteorological conditions from the network. A second condition generation network learns the mapping from traffic flow variables to observed data variables under fixed meteorological condition variables, and extracts potential representations of the impact of traffic flow from the network. The first and second condition generation networks are subjected to intervention consistency constraints during training.

[0012] As a further aspect of the present invention, it also includes calculating the average causal effect of meteorological conditions on data quality, the average causal effect of traffic flow on data quality, and the interactive causal effect of meteorological conditions and traffic flow on data quality; dynamically adjusting the weight of the cross-domain feature interaction analysis module in the comprehensive evaluation based on the strength of the interactive causal effect; and determining the dominant factors affecting data quality based on the relative magnitude of the average causal effect.

[0013] As a further aspect of the present invention, the data integrity index is the ratio of the actual number of data entries received to the theoretically expected number of data entries received within the evaluation period; the data accuracy index is calculated based on the error between the data of the evaluated detector and the data of the high-precision reference system; the high-precision reference system is specifically a lidar. The data timeliness index is calculated based on the average delay time from data acquisition to data storage.

[0014] As a further aspect of the present invention, it also includes weighted summation of the data integrity index, data accuracy index, and data timeliness index according to a preset weight vector to obtain a comprehensive data quality score; wherein, the weight vector is dynamically configured according to the traffic management application scenario, in the real-time traffic guidance scenario, the data accuracy index and the timeliness index are given higher weights; in the long-term traffic planning scenario, the data integrity index is given higher weights.

[0015] On the other hand, this invention provides a dynamic evaluation system for traffic detection data quality based on meteorological-traffic decoupling, comprising: a data synchronization acquisition module for acquiring time-series traffic flow data and environmental meteorological data, performing timestamp alignment and data preprocessing, and outputting a synchronized multimodal data stream; an independent feature representation module connected to the data synchronization acquisition module, including a temporal convolutional network encoder and a multilayer perceptron encoder, for extracting traffic flow depth feature vectors and meteorological condition depth feature vectors from the multimodal data stream; and a meteorological-traffic causal decoupling module connected to the independent feature representation module, employing a causal intervention architecture based on a causal graph model to perform causal intervention on the traffic flow depth feature vectors and meteorological condition depth feature vectors. The system performs causal decoupling on the deep feature vectors of traffic flow and outputs potential representations of traffic flow and meteorological conditions. A cross-domain feature interaction analysis module, connected to the meteorological-traffic causal decoupling module, analyzes the interaction between these two representations using a multi-head self-attention mechanism, generating interactive feature representations. A comprehensive data quality assessment module, connected to both the meteorological-traffic causal decoupling module and the cross-domain feature interaction analysis module, includes a feedforward neural network regression model. Using a non-linear activation function, it generates a data quality assessment vector and a comprehensive data quality score based on the potential representations of traffic flow, meteorological conditions, and interactive feature representations.

[0016] As a further embodiment of the present invention, the data synchronization acquisition module includes: a traffic data interface unit, which communicates with the roadside traffic detector cluster via a message queue telemetry transmission protocol; a meteorological data interface unit, which requests meteorological data application interface via a hypertext transfer security protocol; and a data synchronization processor, which buffers, resamples, and encapsulates the received data to form data frames with a unified frequency and format.

[0017] As a further aspect of the present invention, the data quality assessment vector output by the data quality comprehensive assessment module is transmitted to the data quality visualization and alarm unit; the data quality visualization and alarm unit is used to display the data quality assessment vector and the comprehensive data quality score in real time, and generate an alarm event when any indicator is lower than the preset maintenance threshold, and record alarm information, meteorological data, traffic flow data and decoupled representation to the system log.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0019] This invention, through a causal decoupling module, can clearly separate the independent contributions of weather and traffic flow to data quality. When data quality deteriorates, it can clearly identify whether the dominant factor is severe weather or abnormal traffic flow, providing a precise basis for decision-making in data repair, sensor calibration, or maintenance, overcoming the shortcomings of traditional "black box" assessments.

[0020] This invention not only analyzes the independent effects of each factor, but also specifically models the nonlinear coupling effect between weather and traffic flow through a cross-domain interactive analysis module. This enables the invention to accurately capture and evaluate complex and compound scenarios such as "the exacerbation of video detection accuracy due to congestion under low visibility conditions," greatly improving the robustness of evaluation under extreme or special conditions.

[0021] The output of this invention is no longer a single quality score, but a multi-dimensional vector covering completeness, accuracy and timeliness, providing traffic managers with a comprehensive view of data availability, enabling them to make more refined decisions based on the specific requirements of different applications for different dimensions of data quality.

[0022] The deep learning model constructed in this invention can automatically learn the complex mapping relationship between weather, flow, and data quality from massive historical data, without the need for manually setting cumbersome evaluation rules. It can also process the input data stream in real time and dynamically and continuously output the data quality evaluation results, fully meeting the real-time requirements of intelligent transportation systems for data quality monitoring. Attached Figure Description

[0023] Figure 1 This is a structural block diagram of the traffic detection data quality dynamic evaluation system based on meteorological-traffic flow decoupling proposed in this invention; Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation

[0024] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” and “described” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the described features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0025] Please see Figures 1 to 2 This invention provides a dynamic evaluation method for traffic detection data quality based on meteorological-traffic decoupling, comprising the following steps:

[0026] S1: Multi-source data synchronous acquisition and preprocessing acquires traffic flow data and environmental meteorological data. Traffic flow data comes from traffic detectors deployed at the target cross-section, collecting and uploading three core parameters every minute: cross-section traffic flow (the total number of vehicles passing through the detection cross-section per unit time), average vehicle speed, and lane occupancy rate (the proportion of time the detector is occupied by vehicles to the total sampling period). Environmental meteorological data is acquired through meteorological sensors deployed at the same location as the traffic detectors or by calling the public application programming interface of the local meteorological bureau, including rainfall intensity, atmospheric visibility, ambient temperature, wind speed, and road surface condition index (a dimensionless slipperiness coefficient, with values ​​between 0 and 1; the larger the number, the more slippery the road surface). The raw update frequency of meteorological data is once every 3-5 minutes. After receiving the data, the data synchronization processor first temporarily stores the data in a buffer queue. For meteorological data, a linear interpolation algorithm is used, with an interval of one minute, to perform interpolation calculations between two measured data points to align their timestamps with the traffic flow data every minute. The interpolation formula is: in, This represents the timestamp of the k-th interpolation point; Indicates the timestamp Interpolation at the location; and This represents the timestamps of two adjacent measured meteorological data points before and after the k-th interpolation point, satisfying the following conditions: ; and These represent timestamps. and The actual observed meteorological parameter values; The time factor is a coefficient between 0 and 1. After alignment, all parameters are normalized so that their values ​​fall within the range [0,1]. Finally, an 8-dimensional synchronous data sequence is generated every minute.

[0027] S2: Independent Deep Feature Extraction

[0028] Synchronous multimodal input data sequences are fed into an independent feature representation module, which includes a parallel temporal convolutional network encoder and a multilayer perceptron encoder. The synchronous data sequences from the past two hours (120 points in total) are used as a sliding window input. The temporal data is then fed into the temporal convolutional network encoder. This encoder consists of four stacked residual blocks, each containing a one-dimensional causal convolutional layer, a weight normalization layer, and a ReLU activation function layer, with dilation coefficients of 1, 2, 4, and 8 respectively, to capture multi-scale temporal dependencies. The final output is a 32-dimensional traffic flow deep feature vector. Simultaneously, meteorological data within the same window is input into a multilayer perceptron encoder. This encoder first performs global average pooling on each meteorological parameter along the time dimension, obtaining a 5-dimensional summary vector. This vector is then passed through two fully connected layers (16 and 8 neurons respectively), using the SELU activation function, ultimately outputting a 16-dimensional meteorological condition deep feature vector.

[0029] S3: Meteorological-Flow Causal Decoupling

[0030] Traffic flow depth feature vector and meteorological condition depth feature vector Concatenate into a fused feature vector The data is input to the meteorological-flow causal decoupling module. This module operates based on a pre-defined causal graph model. The causal graph contains four variables: meteorological condition W (corresponding to...). Traffic flow T (corresponding to) ), Observation data D (corresponding to The data quality Q (target variable) is defined as: W → D ← T, where there are unobserved confounding factors U between W and T (such as holidays or large events), and D affects Q.

[0031] To separate the independent influences of meteorological conditions and traffic flow on observed data from the fused feature vectors, the meteorological-traffic causal decoupling module performs the following two virtual intervention operations based on causal intervention, and uses a conditional generation network to realize the data mapping under the intervention:

[0032] implement That is, in the cause-and-effect graph, the traffic flow variable T is fixed as a baseline value. The causal diagram after intervention is denoted as All edges pointing to T are cut off, and T is no longer affected by the original causal variables. This is reflected in the causal graph after intervention. In this case, the generation of observation data D depends solely on meteorological conditions W, and the module trains the first conditional generation network. Let's learn about mapping relationships: in, This indicates that the traffic flow is fixed at... Under the given conditions, the estimated value of the observed data. (Network) The bottleneck features of the intermediate layer were extracted as a potential representation of the influence of meteorological conditions. .

[0033] implement That is, fixing the meteorological condition variable W as a baseline value. The causal relationship after intervention is denoted as All edges pointing to W are cut off in the causal graph after intervention. In this case, the generation of observation data D depends solely on traffic flow T, and the module trains a second conditional generation network. Let's learn about mapping relationships: in, This indicates that the meteorological conditions are fixed as Under these conditions, the estimated value of the observed data. (Network) The bottleneck features of the intermediate layer were extracted as a potential representation of the impact on traffic flow. .

[0034] The training process of the meteorological-flow causal decoupling module is as follows:

[0035] Using historical synchronous data, training sample triples (W, T, D) are constructed, where W represents meteorological features, T represents traffic flow features, and D represents the corresponding fused features. A dataset is constructed to simulate do(T) intervention (fixed traffic flow). ,in This represents the baseline value for traffic flow. Similarly, to simulate do(W) intervention (fixed weather conditions), a dataset is constructed. ,in The baseline value for the set meteorological conditions;

[0036] Using datasets Training the first-condition generator network The input is meteorological feature W, the learning process reconstructs and fuses feature D, and the training objective is to minimize the reconstruction error.

[0037] Using the training set Training the second conditional generation network Input traffic flow features T, learn, reconstruct, and fuse features D;

[0038] Intervention Consistency Constraints:

[0039] When the meteorological characteristics W of the input sample change while the traffic characteristics T remain fixed, the constraints... Maintain stability; conversely, when traffic characteristics T change while meteorological characteristics W remain fixed, the constraint... To maintain stability, the intervention consistency constraint is implemented by adding a regularization term to the training loss function, which penalizes the behavior when T (or W) is fixed. (or The fluctuations caused by changes in W (or T); after training, for any fused feature vector D input online, it is simultaneously input into the trained... and The network consists of two networks that propagate forward and directly output the decoupled latent representations. and ;

[0040] The meteorological-flow causal decoupling module also includes a causal contribution quantification unit, which is based on a trained... and The network obtains the average causal effect of meteorological conditions through counterfactual reasoning. Average causal effect of traffic flow The interaction causal effect of weather-flow (ICE) is also considered, and the weight allocation in the decoupling process is dynamically adjusted based on the causal effect.

[0041] like If so, then meteorological conditions are determined to be the dominant factor;

[0042] like If so, then traffic flow is determined to be the dominant factor;

[0043] Otherwise, it is determined to be a compound effect; To determine the threshold, a value of 0.6 is used in this embodiment.

[0044] In step S4, based on the interaction causality effect (ICE) in step S3, the importance weight of the cross-domain feature interaction analysis module is determined. : in This represents the base weight, with a value of 1.0. Indicates the magnification factor. This represents the interaction effect threshold, with a value of 0.15, indicating when the interaction causal effect... Exceeding the threshold At that time, importance weight Increase linearly on top of the base weight; otherwise, keep the base weight unchanged.

[0045] Decoupling will yield potential characterizations of traffic flow impacts. and the potential characterization of the meteorological conditions mentioned above. Input cross-domain feature interaction analysis module. This module first analyzes the cross-domain features based on importance weights. Adjust the configuration of the attention mechanism when When the value is large, it indicates a significant interaction effect, and the module will increase the attention given to the interaction; otherwise, the basic configuration will be maintained.

[0046] Specifically, Linear projection as query matrix ,Will The linear projection is given by the key matrix K and the value matrix V. Calculate the scaled dot product attention: in, The dimension of the key vector is sixteen in this embodiment. This attention mechanism calculates the attention weights for all meteorological feature dimensions at each traffic flow feature dimension, thereby quantifying the sensitivity of a specific traffic state to different meteorological conditions. The module is configured with four attention heads, each independently calculating its attention output. The outputs of the four heads are then concatenated along the feature dimensions, followed by a linear projection layer to finally generate a 16-dimensional interactive feature representation. .

[0047] In step S5, the potential impact representations of traffic flow, meteorological conditions, and weighted interaction features are concatenated along the feature dimension to form a comprehensive feature vector. This vector is then input to the data quality comprehensive evaluation module. This module is a feedforward neural network regression model with three hidden layers, using SELU as the activation function. The output layer has three neurons, representing: integrity index... The ratio of the actual number of valid data packets received within the evaluation period (2 hours) to the theoretical value (120 packets);

[0048] Accuracy indicators Calculate the root mean square error between the average vehicle speed measured by the traffic detector and the reference data from the cross-sectional lidar. And through the formula: Normalization is performed. Among them... The root mean square error between the average vehicle speed measured by the traffic detector and the lidar reference data is... When it is 0, =1; when Much larger hour, Approaching 0, In this embodiment, a reference speed value of 50 kilometers per hour is used for normalization.

[0049] Timeliness indicators : ,in This refers to the average latency from data collection to data entry into the central database. The preset maximum allowable delay time is 30 seconds. Use a very small positive number to avoid the denominator being zero, and ensure this through a truncation operation. The value does not exceed 1.

[0050] The regression model uses a large amount of labeled historical data for supervised learning during the training phase; the labels are the actual calculated values ​​of the three indicators mentioned above. After training, the model predicts the quality indicators of these three dimensions in real time based on the input comprehensive feature vector. In addition, the system is configured with a preset weight vector. The sum of all weighted components is one, and this configuration is tailored to the specific traffic management application scenario. In scenarios used for real-time traffic guidance, accuracy and timeliness are more important, and settings can be adjusted accordingly. In scenarios involving long-term transportation planning, completeness is even more critical, and settings can be configured... The overall data quality score S is calculated by weighted summation: + ,in , and The weights for completeness, accuracy, and timeliness indicators are respectively, and they satisfy the following conditions: + + =1.

[0051] Based on the current comprehensive data quality score S and the preliminary dominant factor determination results obtained from the causal contribution quantification unit, the system outputs the final data quality assessment conclusion and dominant factor attribution.

[0052] A dynamic evaluation system for traffic detection data quality based on meteorological-traffic flow decoupling includes a data synchronization acquisition module, an independent feature characterization module, a meteorological-traffic flow causal decoupling module, a cross-domain feature interaction analysis module, and a comprehensive data quality evaluation module. The data synchronization acquisition module comprises a traffic data interface unit, a meteorological data interface unit, and a data synchronization processor. The traffic data interface unit establishes a persistent connection with the roadside traffic detector cluster via a message queue telemetry transmission protocol, receiving data packets containing traffic flow, speed, and occupancy in real time, and parsing and encapsulating them into an internal data structure. The meteorological data interface unit initiates requests to the meteorological bureau's application programming interface (API) every five minutes via a hypertext transfer security protocol to obtain structured meteorological data and performs format parsing. The data synchronization processor has a built-in first-in-first-out (FIFO) data buffer queue for aligning data streams from different sources and calls a linear interpolation algorithm resampler to unify the meteorological data to a sampling frequency of once per minute, ultimately encapsulating it into a unified data frame structure for use by downstream modules. The independent feature representation module includes a temporal convolutional network encoder and a multilayer perceptron encoder. The temporal convolutional network encoder processes the traffic flow time-series data and contains causal convolutional layers with exponentially growing expansion coefficients. The multilayer perceptron encoder processes the environmental meteorological data and contains multiple fully connected hidden layers with the number of neurons decreasing layer by layer. The activation function uses scaled exponential linear units. The meteorological-traffic causal decoupling module adopts a causal intervention architecture based on a causal graph model to causally decouple the deep feature vectors of traffic flow and meteorological conditions, outputting potential representations of the impact of traffic flow and meteorological conditions. The cross-domain feature interaction analysis module is configured with four attention heads and uses an attention mechanism to calculate the attention weights of each traffic flow feature dimension on all meteorological feature dimensions, including linear transformation of query, key, and value, scaled dot product attention calculation, multi-head stitching, and output projection. The data quality comprehensive evaluation module includes a feedforward neural network regression model and connects to a data quality visualization and alarm unit at its backend. This alarm unit continuously receives the three-dimensional evaluation vector and comprehensive score output by the evaluation module and displays them in real time on a dashboard in the graphical user interface. The system presets three maintenance thresholds, corresponding to completeness, accuracy, and timeliness, respectively, such as 0.85, 0.80, and 0.90. When any indicator falls below its corresponding threshold, the alarm unit immediately triggers an alarm event, generating a complete log record that includes the event time, alarm type, current meteorological data snapshot, traffic flow data snapshot, decoupled potential impact characteristics of traffic flow and meteorological conditions, and interaction feature characteristics. This log record is then pushed to maintenance personnel for accurate fault tracing analysis.For example, if the alarm is dominated by the high rainfall intensity feature in the potential representation of meteorological conditions, it can be preliminarily determined that the data quality problem is caused by the interference of severe weather on the sensor; if the alarm is dominated by the high time occupancy feature in the potential representation of traffic flow, the problem may be caused by detector saturation or obstruction due to traffic congestion.

[0053] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A dynamic evaluation method for traffic detection data quality based on meteorological-traffic flow decoupling, characterized in that, The process includes the following steps: acquiring time-series traffic flow data and environmental meteorological data, and performing timestamp alignment and normalization to form a synchronized multimodal input data sequence; inputting the multimodal input data sequence into an independent feature representation module to generate a traffic flow deep feature vector and a meteorological condition deep feature vector, respectively; concatenating the traffic flow deep feature vector and the meteorological condition deep feature vector into a fused feature vector, and inputting it into a meteorological-traffic causal decoupling module; the meteorological-traffic causal decoupling module, based on a constructed causal graph model, decouples the mutually independent potential representations of traffic flow impact and meteorological condition impact from the fused feature vector through causal intervention operations; inputting the potential representations of traffic flow impact and meteorological condition impact into a cross-domain feature interaction analysis module to generate an interactive feature representation of the nonlinear coupling effect between meteorological conditions and traffic flow; The potential impact characteristics of traffic flow, meteorological conditions, and interaction features are concatenated into a comprehensive feature vector, which is then input into the data quality comprehensive assessment module to generate a data quality assessment vector and a comprehensive data quality score. The data quality assessment vector includes at least data integrity indicators, data accuracy indicators, and data timeliness indicators.

2. The method for dynamic evaluation of traffic detection data quality based on meteorological-traffic flow decoupling according to claim 1, characterized in that: The traffic flow time series data includes cross-sectional traffic flow, average vehicle speed, and lane time occupancy rate collected by roadside detectors; the environmental meteorological data includes rainfall intensity, atmospheric visibility, ambient temperature, wind speed, and road surface condition parameters; the timestamp alignment process uses an interpolation algorithm to adjust the sampling frequency of the environmental meteorological data to be consistent with the sampling frequency of the traffic flow time series data.

3. The method for dynamic evaluation of traffic detection data quality based on meteorological-traffic flow decoupling according to claim 1, characterized in that: The independent feature representation module includes a temporal convolutional network encoder and a multilayer perceptron encoder; the temporal convolutional network encoder is used to process the traffic flow time series data, and it contains causal convolutional layers with exponentially growing expansion coefficients; the multilayer perceptron encoder is used to process the environmental meteorological data.

4. The method for dynamic evaluation of traffic detection data quality based on meteorological-traffic flow decoupling according to claim 1, characterized in that, The meteorological-traffic causal decoupling module achieves decoupling in the following ways: It constructs a causal graph model of meteorological conditions (W), traffic flow (T), detection data (D), and data quality (Q); based on the causal graph model, it simulates intervention operations on traffic flow variables and intervention operations on meteorological condition variables, respectively; it learns the mapping from meteorological condition variables to observed data variables under fixed traffic flow variable conditions through a first condition generation network, and extracts potential representations of the impact of meteorological conditions from this network; it learns the mapping from traffic flow variables to observed data variables under fixed meteorological condition variables through a second condition generation network, and extracts potential representations of the impact of traffic flow from this network; wherein, the first condition generation network and the second condition generation network are subjected to intervention consistency constraints during training.

5. The method for dynamic evaluation of traffic detection data quality based on meteorological-traffic decoupling according to claim 4, characterized in that, It also includes calculating the average causal effect of meteorological conditions on data quality, the average causal effect of traffic flow on data quality, and the interactive causal effect of meteorological conditions and traffic flow on data quality; dynamically adjusting the weight of the cross-domain feature interaction analysis module in the comprehensive evaluation based on the strength of the interactive causal effect; and determining the dominant factors affecting data quality based on the relative magnitude of the average causal effect.

6. The method for dynamic evaluation of traffic detection data quality based on meteorological-traffic flow decoupling according to claim 1, characterized in that: The data integrity index is the ratio of the actual number of data entries received to the theoretical number of data entries that should be received within the evaluation period; the data accuracy index is calculated based on the error between the data of the evaluated detector and the data of the high-precision reference system; and the data timeliness index is calculated based on the average delay time from data acquisition to data storage.

7. The method for dynamic evaluation of traffic detection data quality based on meteorological-flow decoupling according to claim 6, characterized in that, It also includes weighted summation of the data integrity index, data accuracy index, and data timeliness index according to a preset weight vector to obtain a comprehensive data quality score; wherein, the weight vector is dynamically configured according to the traffic management application scenario. In the real-time traffic guidance scenario, the data accuracy index and the timeliness index are given higher weights; in the long-term traffic planning scenario, the data integrity index is given higher weights.

8. A dynamic evaluation system for traffic detection data quality based on meteorological-traffic flow decoupling, characterized in that, include: The data synchronization acquisition module is used to acquire and synchronously process traffic flow time-series data and environmental meteorological data, and output multimodal data streams; An independent feature characterization module, connected to the data synchronization acquisition module, is used to extract traffic flow deep feature vectors and meteorological condition deep feature vectors from the multimodal data stream; a meteorological-traffic causal decoupling module, connected to the independent feature characterization module, is used to perform causal decoupling on the traffic flow deep feature vectors and meteorological condition deep feature vectors, and output potential characterizations of traffic flow impact and meteorological condition impact. The cross-domain feature interaction analysis module, connected to the meteorological-traffic causal decoupling module, is used to analyze the interaction relationship between the potential representation of traffic flow impact and the potential representation of meteorological conditions impact, and generate interactive feature representations. The data quality comprehensive assessment module is connected to the meteorological-traffic causal decoupling module and the cross-domain feature interaction analysis module. Based on the potential impact characteristics of traffic flow, the potential impact characteristics of meteorological conditions, and the interaction feature characteristics, it generates a data quality assessment vector and a comprehensive data quality score.

9. The dynamic evaluation system for traffic detection data quality based on meteorological-flow decoupling according to claim 8, characterized in that, The data synchronization acquisition module includes: a traffic data interface unit, which communicates with the roadside traffic detector cluster via a message queue telemetry transmission protocol; a meteorological data interface unit, which requests meteorological data application interface via a hypertext transfer security protocol; and a data synchronization processor, which buffers, resamples, and encapsulates the received data to form data frames with a unified frequency and format.

10. The dynamic evaluation system for traffic detection data quality based on meteorological-flow decoupling according to claim 8, characterized in that, Also includes: The data quality visualization and alarm unit is connected to the data quality comprehensive evaluation module; the data quality visualization and alarm unit is used to display the data quality evaluation vector and comprehensive data quality score in real time, and generate alarm events and logs when any evaluation index is lower than a preset threshold.

Images