A method for predicting urban road traffic congestion index

Abstract

This application discloses a method for predicting urban road traffic congestion index, comprising the following steps: acquiring congestion index data of a target road segment, decomposing it to obtain intrinsic mode functions (IMFs) and residual terms of different frequencies; training each IMF component to obtain prediction results for IMFs of different frequencies; superimposing the prediction results of each subsequence to integrate the predicted value of the congestion index; verifying the predicted value based on evaluation indicators and quantifying the deviation between the predicted value and the actual value. This prediction method acquires high-quality congestion index data, fully utilizes the temporal information of the sequence data to improve prediction accuracy; and superimposes and reconstructs the prediction results of each subsequence to maintain the consistency and continuity of the prediction results. Ultimately, it achieves the effect of accurately predicting the congestion index during peak hours while mining complex data fluctuations, thus improving the prediction accuracy of the congestion index. This method brings a new development to alleviating urban traffic congestion problems and improves prediction accuracy.

Description

Technical Field

[0001] This application belongs to the field of intelligent transportation, specifically a method for predicting urban road traffic congestion index. Background Technology

[0002] In recent years, with the continuous advancement of urban modernization, traffic congestion has become a bottleneck problem for the development of many cities. How to effectively analyze historical congestion index data, predict congestion indices, understand the causes and bottleneck areas of urban traffic congestion, and provide data support for formulating targeted traffic management measures is an important issue that needs to be addressed.

[0003] Existing research on traffic congestion index prediction typically employs four methods: statistical analysis prediction models, nonlinear theoretical models, machine learning prediction models, and combined prediction models. While current research has made significant improvements in congestion index prediction, there are still areas for refinement. First, existing studies primarily use single prediction models or combinations of multiple machine learning models to predict road congestion indices, which has limitations in uncovering the complex fluctuation structure of the congestion index. Second, the prediction accuracy of traditional methods such as nonlinear regression needs improvement. Furthermore, existing prediction models perform well when the congestion index sequence is stable, but their accuracy needs to be improved when the peak congestion index fluctuates significantly. Summary of the Invention

[0004] To address the aforementioned problems in existing technologies, this application provides a method for predicting urban road traffic congestion index. This method mines complex fluctuations in data while accurately predicting the congestion index during peak hours, thereby improving the prediction accuracy of the congestion index and providing a new prediction method for alleviating urban traffic congestion.

[0005] To achieve the above objectives, this application adopts the following technical solution: a method for predicting urban road traffic congestion index, comprising the following steps:

[0006] Obtain congestion index data for the target road segment and decompose it to obtain intrinsic mode functions and residual terms at different frequencies;

[0007] Train each intrinsic mode function component to obtain prediction results for intrinsic mode functions of different frequencies;

[0008] The prediction results of each subsequence are superimposed to construct the overall congestion index prediction.

[0009] The predicted values ​​are verified based on the evaluation indicators, and the deviation between the predicted values ​​and the actual values ​​is quantified.

[0010] The acquisition of congestion index data for the target road segment includes:

[0011] Obtain historical congestion index data for the target road segment and preprocess the historical congestion index data;

[0012] Descriptive statistical analysis was performed on the processed data to obtain congestion index data.

[0013] The preprocessing includes handling outliers, duplicates, and missing values ​​in historical congestion index data. The descriptive statistics include analysis of mean, variance, skewness, kurtosis, normality test, and serial autocorrelation test.

[0014] The decomposition yields intrinsic mode functions and residual terms at different frequencies, including:

[0015] Add N times the same length of adaptive noise to the original signal sequence to obtain a new signal sequence. Perform N decompositions on the new sequence based on empirical mode decomposition, and the average value is the first characteristic mode function.

[0016] Obtain the first margin sequence, and calculate the second feature mode function based on the margin sequence;

[0017] Repeated calculations yield the marginal sequences and k+1 characteristic modal components of stage k+1 and stage k;

[0018] Repeat the calculation until the extreme point of the margin sequence is no greater than 2, and obtain K intrinsic mode functions and residual terms.

[0019] Among them, the new signal sequence x i (t) The expression for the first characteristic mode function IMF1 is:

[0020] x i (t)=x(t)+εω i (t);

[0021]

[0022] Where x(t) is the original signal sequence, ε is the adaptive coefficient, and ω i (t) is a noise sequence.

[0023] The expressions for the first margin sequence r1(t) and the second characteristic mode function IMF2 are as follows:

[0024] r1(t) = x(t) - IMF1(t);

[0025]

[0026] Wherein, the marginal sequence r of stage k+1 and stage k. k (t) and k+1 characteristic mode components IMF k+1 ,

[0027] The expression is:

[0028] r k (t)=r k-1 (t)-IMF k (t);

[0029]

[0030] The expressions for the K intrinsic mode functions and residual terms are as follows:

[0031]

[0032] The training of each intrinsic mode function component includes:

[0033] By resetting the gate, the weight matrix W is processed. Xr Get the output R of the reset gate t The expression is:

[0034] R t =σ(X) t W xr +H t-1 W hr +b r );

[0035] Where R t X represents the gate output being reset at time t; t Represents the current input; H t-1 Represents the hidden state at the previous moment; δ represents the sigmoid function; W xr W hr Represents the weight parameter; b r Represents the deviation parameter;

[0036] By updating the door, the hidden state H of the previous level is revealed. t-1 Part of the data is passed to the next layer for training, expressed as:

[0037] Z t =σ(X) t W xz +H t-1 W hz +b z

[0038] Z t δ represents the output of the gate updated at time t; W represents the sigmoid function. xz W hz Represents the weight parameter; b z Represents the deviation parameter;

[0039] The candidate hidden state is obtained by element-wise multiplying the output of the reset gate at time t with the hidden state at time t-1, and then combining the calculated value with the input at time t. The expression is:

[0040]

[0041] in W represents the candidate hidden state at time t; xh W hh Represents the weight parameter; b h Represents the deviation parameter; ☉ represents element-wise multiplication;

[0042] Based on update gate Z t For H t-1 and The hidden state H at time t is obtained by performing a combination. t The expression is:

[0043]

[0044] Where H t-1 H represents the hidden state of the previous moment; t Represents the current hidden state; Z represents the current candidate hidden state; t Represents the output of the currently updated gate; ☉ represents element-wise multiplication;

[0045] Different weights are assigned based on the relevance of different positions in the input sequence, as expressed in the following expression:

[0046]

[0047] Where Q represents the query vector; K represents the key vector; and V represents the value vector.

[0048] The evaluation indicators include the coefficient of determination, mean absolute percentage error, and mean absolute error.

[0049] The beneficial effects of this application are:

[0050] This application provides a method for predicting urban road traffic congestion index. It acquires high-quality congestion index data, accurately and comprehensively captures the frequency characteristics of the data, and fully utilizes the temporal information of the sequence data to improve prediction accuracy. The prediction results of each subsequence are superimposed and reconstructed to maintain consistency and continuity, ultimately achieving the goal of accurately predicting peak-hour congestion index while mining complex data fluctuations, thus improving the prediction accuracy of the congestion index. This method brings a new development to alleviating urban traffic congestion and improves prediction accuracy, possessing positive application value for urban traffic management and planning. Attached Figure Description

[0051] Figure 1 This is a schematic flowchart illustrating the prediction method in an embodiment of this application;

[0052] Figure 2 This is a schematic diagram of the framework of the prediction model of the prediction method shown in the embodiments of this application;

[0053] Figure 3 This is a GRU network structure diagram of the prediction method shown in the embodiments of this application.

[0054] Figure 4 This is a GRU prediction flowchart illustrating the prediction method in an embodiment of this application. Detailed Implementation

[0055] Embodiments of this application will now be described in more detail with reference to the accompanying drawings. While embodiments of this application are shown in the drawings, it should be understood that this application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to make this application more thorough and complete, and to fully convey the scope of this application to those skilled in the art.

[0056] It should be understood that although the terms "first," "second," "third," etc., may be used in this application to describe various information, this information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0057] In the description of this application, it should be understood that the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application.

[0058] Unless otherwise expressly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0059] Example 1

[0060] A method for predicting urban road traffic congestion index is proposed, based on a gated recurrent neural network attention model using CEEMDAN-GRU-ATTENTION complex empirical mode decomposition. CEEMDAN-GRU-ATTENTION is a deep learning model that combines CEEMDAN's complete ensemble empirical mode decomposition with adaptive noise reduction, GRU gated recurrent units, and an attention mechanism. CEEMDAN is used to decompose and denoise time-series data, then a GRU network is used to model long-term dependencies, and finally, the attention mechanism dynamically focuses on important feature information. By combining these three components, the CEEMDAN-GRU-ATTENTION model efficiently extracts complex features from time-series data, achieving more accurate prediction and analysis.

[0061] like Figure 1-4 The method for predicting urban road traffic congestion index shown includes the following steps:

[0062] S1. Obtain congestion index data for the target road segment and decompose it to obtain the intrinsic mode functions and residual terms at different frequencies. This includes the following steps:

[0063] S11. Data Preprocessing and Analysis: Obtain historical congestion index data for road segments, and preprocess the congestion index data. Data preprocessing includes handling outliers, duplicates, and missing values, removing outliers and duplicates, and imputing missing values. Descriptive statistical analysis is then performed on the processed data, including analysis of mean, variance, skewness, and kurtosis. Jarque Bera normality tests and Ljung-Box tests are performed to determine whether the dataset approximates a normal distribution and whether autocorrelation in the time series data is significant.

[0064] S12. Data Decomposition: The congestion index data is decomposed using the CEEMDAN model to obtain intrinsic mode functions (IMFs) and residual terms at different frequencies.

[0065] The specific method is as follows:

[0066] S121. A new signal sequence x is obtained by adding N times the same length of adaptive noise to the original signal sequence x(t). i (t). The new sequence is decomposed N times using EMD, and the average value is the first characteristic mode function IMF1, expressed as follows:

[0067] x i (t)=x(t)+εω i (t

[0068]

[0069] Where ε is the adaptive coefficient and is the noise sequence.

[0070] S122. Calculate the first margin sequence r1(t), as shown in the following expression:

[0071] r1(t) = x(t) - IMF1(t)

[0072] S123. Based on S202, calculate the second characteristic mode function IMF2, with the following expression:

[0073]

[0074] S124. Repeat the calculation to obtain the marginal sequence r of stage k+1 and stage k. k (t) and k+1 characteristic mode components IMF k+1 The expression is as follows:

[0075] r k (t)=r k-1 (t)-IMF k (t

[0076]

[0077] S125. Repeat the above steps until the extreme points of the margin sequence are no greater than 2. Through these steps, the original sequence can be decomposed into K characteristic mode functions and a margin sequence, as shown in the following expression:

[0078]

[0079] In this invention, the original nonlinear and nonstationary time series data is decomposed into 15 IMF components and a residual term R. (n) .

[0080] S2. Train each intrinsic mode function component to obtain the prediction results of intrinsic mode functions at different frequencies.

[0081] A congestion index prediction model based on GRU-ATTENTION was constructed. The model was trained and used to predict the intrinsic mode functions (IMFs) of each IMF, yielding prediction results for IMFs at different frequencies. The activation function chosen was ReLU, representing the rectified linear unit, expressed as follows:

[0082] f(x) = max(0,x)

[0083] The loss function chosen is mean squared error (MSE), expressed as follows:

[0084]

[0085] Where n represents the total number of samples; y represents the true value; This represents the predicted value.

[0086] The Adam optimizer was selected, and the learning rate was set to 0.0003 to avoid overfitting of the model.

[0087] The steps for running the GRU model are as follows:

[0088] Step 1: Clean the initial road congestion index data, including deleting duplicate values, identifying and repairing outliers and missing values, and normalizing the data.

[0089] Step 2: Initialize the network and set the basic initial parameters of the GRU model, including the number of hidden layer units, input stride, number of iterations, and forgetting rate;

[0090] Step 3: Divide the road congestion index dataset into a training set and a validation set in an 8:2 ratio;

[0091] Step 4: Perform iterative training of the GRU model. The dataset will output predicted values ​​in the GRU model. Calculate the error between the model's predicted values ​​and the true values.

[0092] Step 5: Calculate the loss function and use the Adam optimizer to continuously improve the model's parameters, making the model's predicted values ​​more accurate.

[0093] Step 6: Based on the saved loss history, draw loss maps for the training and validation sets.

[0094] Step 7: Output the evaluation metrics of the model, and analyze the prediction effect and accuracy of the GRU model based on the magnitude of the evaluation metrics.

[0095] Step 8: Draw a curve comparing the predicted data and the actual data.

[0096] Based on this, the specific method of S3 is as follows:

[0097] S21. By resetting the gate, the weight matrix W is... Xr Element-wise multiplication is performed, followed by processing with the sigmoid function to obtain the output R of the reset gate. t The expression is as follows:

[0098] R t =σ(X) t W xr +H t-1 W hr +b r

[0099] Where R t Represents the moment to reset the gate's output; X t Represents the current input; H t-1 Represents the hidden state at the previous moment; δ represents the sigmoid function; W xr W hr Represents the weight parameter; b r This represents the deviation parameter.

[0100] S22. By updating the door, the hidden state H of the previous layer is revealed. t-1 Part of the data is passed to the next layer for training, as shown in the following expression:

[0101] Z t =σ(X) t W xz +H t-1 W hz +b z

[0102] Z t δ represents the output of the gate updated at time t; W represents the sigmoid function. xz W hz Represents the weight parameter; b z This represents the deviation parameter.

[0103] S23. Multiply the output of the reset gate at time t element-wise with the hidden state at time t-1. If the calculated value is approximately 0, it indicates that the hidden state information at time t-1 needs to be selectively ignored or "forgotten"; conversely, if the calculated value is approximately 1, it means that the hidden state information at time t-1 should be stored. After performing element-wise multiplication, merge the calculated value with the input at time t. Finally, set an appropriate activation function in the fully connected layer to process the result obtained in the previous step, thereby obtaining the candidate hidden states. The expression is as follows:

[0104]

[0105] in W represents the candidate hidden state at time t; xh W hh Represents the weight parameter; b h ☉ represents the deviation parameter; ☉ represents element-wise multiplication.

[0106] S24, Using Update Gate Z t For H t-1 and The hidden state H at time t is obtained by performing a combination. t The expression is as follows:

[0107]

[0108] Where H t-1 H represents the hidden state of the previous moment; t Represents the current hidden state; Z represents the current candidate hidden state; t ☉ represents the output of the current update gate; ☉ represents element-wise multiplication.

[0109] S25. Combining the self-attention mechanism allows the model to assign different weights based on the relevance of different positions in the input sequence, as shown in the following expression:

[0110]

[0111] Where Q represents the query vector; K represents the key vector; and V represents the value vector.

[0112] S3. Overlay the prediction results of each subsequence and integrate the predicted value of the congestion index.

[0113] Based on the Intrinsic Mode Functions (IMFs) and residual terms of different frequencies obtained from the CEEMDAN decomposition, predictions are made for each IMF and residual term to obtain prediction results for IMFs of different frequencies. The prediction result of each IMF is added to the residual term to obtain the superposition result of each IMF and residual term. The superposition results of each IMF and residual term are then added together to obtain the predicted value of the congestion index.

[0114] Through this superposition and reconstruction process, the predicted IMFs and residual terms of different frequencies are combined to obtain the overall congestion index prediction result. By comprehensively considering the influence of different frequency components, the prediction accuracy and precision of the congestion index are improved.

[0115] When performing overlay reconstruction, it is essential to maintain temporal consistency among the various IMFs and residual terms to ensure the continuity and accuracy of the prediction results. Therefore, the correct overlay order and the consideration of the weight allocation of each IMF and residual term are crucial. Appropriate algorithms or rules can be used to determine the appropriate overlay weights to obtain the best prediction results.

[0116] S4. Verify the predicted values ​​based on the evaluation indicators and quantify the deviation between the predicted values ​​and the actual values.

[0117] The selected evaluation indicators include the coefficient of determination (R²). 2 ), mean absolute percentage error (MAPE) and mean absolute error (MAE).

[0118] Coefficient of determination R 2 R-value, also known as the goodness of fit, reflects how accurately a model fits the data. 2 The value range of is generally [0,1]. The closer its value is to 1, the better the model's fit. The expression is as follows:

[0119]

[0120] Mean Absolute Error (MAE) is the average of absolute errors and effectively reflects the actual error of the predicted values. Its expression is as follows:

[0121]

[0122] Mean Absolute Percentage Error (MAPE) represents the accuracy of the model. The smaller the value, the better the accuracy of the prediction model. The expression is as follows:

[0123]

[0124] Where n represents the total number of samples; y i Represents the i-th true value; This represents the i-th predicted value; This represents the average of the i true values.

[0125] Example 2

[0126] The data collected over the past three months included approximately 13,000 data entries, with each entry having a time granularity of 10 minutes. The dataset contained six indicators, including congestion index, commute time, and free-flow speed.

[0127] In the data preprocessing stage, outliers and duplicates in the original data were removed, and missing values ​​were imputed. The processed data were then analyzed, including descriptive statistical analysis, Jarque-Bera normality test, and Ljung-Box test.

[0128] In the data decomposition phase, the preprocessed data was decomposed using the CEEMDAN model, resulting in 15 IMF components and one residual term, which were then subjected to descriptive statistics. Different IMF components represent the characteristics of the original signal at different time scales, while the residual reflects the trend of the original congestion index sequence.

[0129] In the model training and prediction phase, a GRU-ATTENTION model is constructed to mine the time-scale features contained in each subsequence. Each subsequence is then trained and predicted separately to obtain the prediction results of the neural network model for different subsequences.

[0130] In the overlay and reconstruction stage, the prediction results of each subsequence are overlaid and reconstructed to integrate the predicted values ​​of the congestion index sequence.

[0131] In the evaluation and analysis phase, the output R... 2 The performance of the constructed network model is analyzed using three metrics: , MAE, and MAPE. The final model's goodness of fit Rfit is also considered. 2 The accuracy rate reaches 94.4%, and the mean absolute percentage error (MAPE) reaches 6.3%, meaning the model's accuracy is 93.7%, with a mean absolute error (MAE) of only 0.11. Therefore, the output indicators also show that the constructed CEEMDAN-GRU-ATTENTION model has a good predictive effect on road congestion indices.

[0132] Finally, it should be noted that in this document, relationships such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "include," "contain," or any other variations are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0133] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0134] The various embodiments of this application have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for predicting urban road traffic congestion index, characterized in that, The following steps are involved: Obtain congestion index data for the target road segment, and decompose the congestion index data using the CEEMDAN model to obtain intrinsic mode functions and residual terms at different frequencies. A congestion index prediction model based on GRU-ATTENTION was constructed, and each intrinsic mode function component was trained to obtain the prediction results of the intrinsic mode functions at different frequencies. The training of each of the intrinsic mode function components includes: The weight matrix is ​​processed by resetting the gate. Get the output of the reset gate The expression is: ； in represent The output of the gate is reset at all times; Represents the current input; Represents the hidden state of the previous moment; δ represents the sigmoid function; , Represents the weighting parameter; Represents the deviation parameter; By updating the door, the hidden state of the previous level is revealed. Part of the data is passed to the next layer for training, expressed as: ； in represent The output of the gate is updated continuously; Represents the sigmoid function; , Represents the weighting parameter; Represents the deviation parameter; use The output of the time-reset gate is reset. The hidden states at time t are multiplied element-wise, and the calculated values ​​are combined with the input at time t to obtain the candidate hidden states. The expression is: ； in Represents the candidate hidden state at time t; , Represents the weighting parameter; Represents the deviation parameter; ☉ represents element-wise multiplication; Based on the update gate right and The hidden state at time t is obtained by performing a combination. The expression is: ； in Represents the hidden state of the previous moment; Represents the current hidden state; Represents the current hidden state of the candidate; Represents the output of the currently updated gate; ☉ represents element-wise multiplication; Different weights are assigned based on the relevance of different positions in the input sequence, as expressed in the following expression: ； Where Q represents the query vector; K represents the key vector; and V represents the value vector. Based on the Intrinsic Mode Functions (IMFs) and residual terms of different frequencies obtained from CEEMDAN decomposition, each IMF and residual term is predicted to obtain the prediction results of IMFs of different frequencies. The prediction results of each IMF are added to the residual terms to obtain the superposition results of each IMF and residual term. The superposition results of each IMF and residual term are added to obtain the predicted value of the congestion index. The prediction results of each subsequence are superimposed to construct the integrated predicted value of the congestion index. The predicted value is verified based on the evaluation index, and the deviation between the predicted value and the actual value is quantified.

2. The method for predicting urban road traffic congestion index as described in claim 1, characterized in that, The acquisition of congestion index data for the target road segment includes: Obtain historical congestion index data for the target road segment and preprocess the historical congestion index data; Descriptive statistical analysis is performed on the processed data to obtain the congestion index data.

3. The method for predicting urban road traffic congestion index as described in claim 2, characterized in that, The preprocessing includes handling outliers, duplicates, and missing values ​​in the historical congestion index data. The descriptive statistics include analysis of mean, variance, skewness, and kurtosis, as well as normality tests and serial autocorrelation tests.

4. The method for predicting urban road traffic congestion index as described in claim 1, characterized in that, The decomposition yields intrinsic mode functions and residual terms at different frequencies, including: Add N times the same length of adaptive noise to the original signal sequence to obtain a new signal sequence. Perform N decompositions on the new sequence based on empirical mode decomposition, and the average value is the first characteristic mode function. Obtain the first margin sequence, and calculate the second feature mode function based on the margin sequence; Repeated calculations Phase and the Marginal sequence of stages and Characteristic modal components; Repeat the calculation until the extreme point of the margin sequence is no greater than 2, to obtain K intrinsic mode functions and residual terms.

5. The method for predicting urban road traffic congestion index as described in claim 4, characterized in that, The new signal sequence The first characteristic mode function The expression is: ； ； in It is the original signal sequence. It is an adaptive coefficient. It is a noise sequence.

6. The method for predicting urban road traffic congestion index as described in claim 4, characterized in that, The first margin sequence The second characteristic mode function The expression is: ； 。 7. The method for predicting urban road traffic congestion index as described in claim 4, characterized in that, The Phase and the Marginal sequence of stages and Characteristic mode components The expression is: ； 。 8. The method for predicting urban road traffic congestion index as described in claim 4, characterized in that, The expressions for the K intrinsic mode functions and the residual term are as follows: 。 9. The method for predicting urban road traffic congestion index as described in claim 1, characterized in that, The evaluation metrics include the coefficient of determination, mean absolute percentage error, and mean absolute error.

Images