A highway traffic parameter estimation method based on evaluation index improved FCM

Abstract

This invention proposes an improved FCM (Fast Moving Mode Decomposition) method for highway traffic parameter estimation based on evaluation indices, primarily by improving the FCM state classification algorithm. First, highway traffic data is stabilized and denoised using Empirical Mode Decomposition (EMD) to reduce errors caused by data fluctuations, constructing a data matrix for model input. Next, a multi-dimensional traffic state evaluation index system is established at both the cross-section and road segment levels, and an improved FCM algorithm is proposed based on this system to achieve accurate traffic state classification. Finally, combining traffic state classification with deep learning methods, an adaptive time series analysis and prediction model is constructed to perform short-term predictions of traffic flow and speed, thereby obtaining dynamic traffic parameter estimation results. This invention incorporates road physical structure and utilizes the improved FCM algorithm for dynamic evaluation of road network traffic states, helping highway managers improve traffic management and scheduling capabilities and optimize the overall operational performance of the road network.

Description

Technical Field

[0001] This invention belongs to the field of intelligent transportation systems and traffic state estimation technology, specifically relating to a method for estimating highway traffic parameters based on an improved FCM (Functional Traffic Management) index. Background Technology

[0002] Traffic flow theory is crucial for studying the operational status and changing patterns of road traffic flow. Key parameters describing traffic flow characteristics include traffic volume, traffic speed, and traffic density. Calculating and analyzing traffic parameters at different time scales is essential for improving the accuracy of highway management. Current technology can dynamically acquire traffic parameter data, but it struggles to obtain traffic parameters for future moments.

[0003] In modern traffic system management, traffic state estimation has always been a crucial task. Traditional traffic parameter estimation methods are often affected by error accumulation and sudden events, resulting in limited accuracy. The FCM (Fuzzy C-Means Algorithm) algorithm integrates fuzzy theory into cluster analysis and is a clustering method based on the optimal objective function.

[0004] In recent years, driven by interdisciplinary research and artificial intelligence technology, research on traffic parameter prediction based on deep learning has attracted much attention. Among these methods, bidirectional long short-term memory (LSTM) network models can learn from traffic flow time-series data in both forward and backward directions, giving the hidden layer outputs richer spatiotemporal characteristics. Therefore, constructing a traffic parameter estimation model that combines the advantages of these methods is of great significance to traffic technology research, enabling a more accurate understanding of traffic conditions and providing more precise information for intelligent traffic management, traffic planning, and road safety. Summary of the Invention

[0005] The problem to be solved by this invention is to provide a method for estimating highway traffic parameters based on an improved FCM (Functional Traffic Management) index, which is used to estimate dynamic traffic parameters.

[0006] This invention adopts the following technical solution: a method for estimating highway traffic parameters based on an improved FCM (Functional Traffic Flow Mechanism) using evaluation indicators, comprising the following steps:

[0007] S1. Collect highway traffic data and preprocess it to construct a stable traffic data matrix;

[0008] S2. Construct a traffic state evaluation index system at two scales: cross section level and road segment level. Introduce the sample size attribute to improve the fuzzy C-means clustering algorithm, classify traffic states, and use classified data to improve the estimation accuracy of highway traffic parameters.

[0009] S3. Construct an adaptive time series analysis and prediction model, and extract and mine the spatiotemporal features in traffic flow data by introducing a bidirectional long short-term memory network model with an attention mechanism.

[0010] S4. Using the stabilized traffic data matrix and traffic state division results, train the traffic parameter estimation model, verify and adjust the traffic parameter estimation model, and complete the training of the traffic parameter estimation model.

[0011] S5. Dynamically input the traffic data matrix into the trained traffic parameter estimation model to obtain dynamic estimation parameters of highway traffic parameters.

[0012] Further, in step S1, highway traffic data is collected and preprocessed, including the following sub-steps:

[0013] S11. Collect raw data of microscopic vehicle motion parameters on the highway;

[0014] S12. Handle missing values ​​in the original data: Check for missing and outlier values ​​in the original data, and replace the missing and outlier values ​​with the mean of the upper and lower values ​​of the missing and outlier values.

[0015] S13. Construct a sample, selecting traffic flow, average vehicle speed, and road physical structure as influencing factors input samples;

[0016] S14. The empirical mode decomposition method is used to perform stationarization decomposition on the sample data, decomposing the time series data into several intrinsic mode functions. The empirical mode decomposition formula is as follows:

[0017]

[0018] Where I(t) is the input signal; IMF i (t) represents the i-th eigenmode function; R k (t) represents the residual sequence; k is the number of intrinsic mode functions;

[0019] S15. Reassemble the decomposed components to construct a stable data matrix;

[0020] S16. The input data is normalized using the maximum-minimum normalization method, and the processed data is divided into a training set and a test set. The normalized data will be in the interval [0, 1]. The calculation formula is as follows:

[0021]

[0022] Where, x scale The data represents the data after normalization using the min-max normalization method; x represents the input data; x max ,x minThese represent the maximum and minimum values ​​of the input data, respectively.

[0023] Furthermore, in step S2, a traffic state evaluation index system is constructed at two scales: cross-section level and road segment level. An FCM algorithm is used to classify traffic states, and the accuracy of parameter estimation is improved by classifying data. The process includes the following steps:

[0024] S21. Select cross-section level indicators to describe the traffic status and its changing patterns of a single cross-section, including cross-section traffic flow, average cross-section speed, and cross-section physical structure.

[0025] S22. Select road segment-level indicators to describe the traffic conditions and their changing patterns within the road segment, including road segment traffic volume, average travel speed of the road segment, road physical structure, and traffic volume in kilometers.

[0026] S23. Initialize the cluster centers and membership matrix, and perform equalization correction on the sample size;

[0027] First, perform initial clustering on the n sample data to obtain the initial category to which each sample belongs;

[0028] Then, based on the traditional FCM algorithm, the sample size attribute is introduced to improve the objective function J, resulting in the improved objective function J′;

[0029] S24. Calculate the cluster centers based on the membership matrix and update the membership matrix;

[0030] S25. Use cross-validation for validation. For n data samples, take 1 value as the validation set in each iteration and the remaining n-1 values ​​as the training set. Repeat the iteration to ensure that each sample point is used as the validation set.

[0031] S26. Output cluster centers and accuracy to complete traffic state classification.

[0032] Further, in step S3, an adaptive time series analysis prediction model is constructed, including the following steps:

[0033] S31. Construct BiLSTM hidden layers. The BiLSTM model is trained from two directions and the training structure is linearly fused, so it can capture the bidirectional influence of time well.

[0034] S32. Construct an Attention layer, integrating the attention mechanism with the BiLSTM model. Based on the BiLSTM model, introduce an attention mechanism to calculate different weights for the hidden layer output vector of the BiLSTM model. The attention mechanism uses the output vector h of the hidden layer in the BiLSTM model at time t as the weight. t As input, the attention weight α is calculated. tCalculation formula:

[0035] u t =tanh(Wh) t +b)

[0036]

[0037] Let the output of the Attention layer at time t be S. t Calculation formula:

[0038]

[0039] Among them, u t The output vector h of the BiLSTM model at time t t The importance of the result; W is the sum of the weight matrices; b is the bias value;

[0040] S33. Construct a Drop layer and use the dropout function to randomly delete neurons during training to prevent the neural network from overfitting during training.

[0041] S34. Construct a fully connected layer, select the sigmoid function as the activation function, and output the predicted value.

[0042] Further, in step S4, the traffic parameter estimation model is trained using the stabilized data matrix and the traffic state classification results, including the following steps:

[0043] S41. Divide the data into a training set and a test set;

[0044] S42. Set the error value and the maximum number of iterations. Train the model in each iteration. If the result is less than the target error value, end the iteration. Otherwise, update the network parameter values ​​and continue the iteration.

[0045] S43. Select the Adam algorithm to optimize model weights;

[0046] S44. Train the model based on the training data and the set hyperparameters;

[0047] S45. Evaluate the model's performance using the validation set and adjust the model as needed;

[0048] S46. Grid search to find optimal parameters:

[0049] Different combinations of parameters in a neural network will predict the results. Therefore, the grid search method is used to adjust parameters such as the number of BiLSTM layer units, the number of iterations, and the size of the time window to find the optimal combination of parameters.

[0050] Furthermore, in step S5, the traffic data matrix is ​​dynamically input into the traffic parameter estimation model trained in step S4 to obtain the dynamic short-term traffic flow estimation parameters of the highway, which are used to dynamically evaluate the traffic status of the road network.

[0051] The present invention also provides: an electronic device, comprising:

[0052] One or more processors;

[0053] A storage device on which one or more programs are stored;

[0054] When the one or more programs are executed by the one or more processors, the one or more processors implement any of the above-mentioned highway traffic parameter estimation methods based on evaluation indicators to improve FCM.

[0055] The present invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps in any of the above-mentioned methods for estimating highway traffic parameters based on evaluation indicators to improve FCM.

[0056] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0057] 1. The present invention provides a dynamic estimation method for highway traffic parameters. Taking into account the influence of road physical structure on traffic conditions, it constructs cross-section-level and road segment-level road operation status evaluation indicators from the perspectives of road segments and road networks, respectively. By introducing an improved FCM algorithm, it dynamically evaluates the traffic status of the road network, making the evaluation of highway operation status more accurate and effective.

[0058] 2. The present invention provides a dynamic estimation method for highway traffic parameters. It introduces empirical mode decomposition to decompose time-series data and construct a stationary data matrix. Noise removal of historical traffic flow data reduces interference from data fluctuations, improves data quality, and thus enhances the model's prediction accuracy and stability.

[0059] 3. This invention combines an improved FCM algorithm for traffic state classification with an adaptive time series analysis prediction model, and uses a BiLSTM model incorporating an attention mechanism for traffic flow parameter estimation. This provides a novel research perspective for the application of deep learning technology in short-term traffic flow estimation. This method offers a new and effective solution for future traffic management strategies, helping to better understand and estimate traffic flow states, thereby providing strong decision support for traffic management departments. Attached Figure Description

[0060] Figure 1 This is a flowchart of the highway traffic parameter estimation method based on the improved FCM based on evaluation index of the present invention;

[0061] Figure 2 This is a schematic diagram of the transformation of the model input matrix in an embodiment of the present invention;

[0062] Figure 3 These are the cross-section and road segment level road operation status evaluation indicators in this embodiment of the invention;

[0063] Figure 4 This is a schematic diagram of the improved FCM algorithm according to an embodiment of the present invention;

[0064] Figure 5 This is a schematic diagram of the adaptive time series analysis and prediction model according to an embodiment of the present invention;

[0065] Figure 6 This is a schematic diagram illustrating the training of the adaptive time series analysis and prediction model according to an embodiment of the present invention;

[0066] Figure 7 This is a comparison chart of the highway traffic parameter estimation results based on the improved FCM based on evaluation indicators and actual data. Detailed Implementation

[0067] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the application will be further described in detail below with reference to the accompanying drawings. The described embodiments are only a part of the embodiments involved in this invention. All non-innovative embodiments based on this invention by other researchers in the art are within the protection scope of this invention.

[0068] See Figure 1 This invention discloses a method for estimating highway traffic parameters based on an improved FCM (Functional Traffic Management) index, comprising the following steps:

[0069] S1: Collect traffic data from highways and preprocess it, including handling missing values ​​and constructing samples, performing EMD (Empirical Mode Decomposition), and performing stationary decomposition on the sample data to remove noise from historical traffic flow and speed data, and constructing a stationary data matrix.

[0070] S2: A traffic state evaluation index system is constructed at two scales: cross-section level and road segment level. An improved FCM (Fuzzy C-Means Algorithm) algorithm is used to classify traffic states and improve the accuracy of parameter estimation by classifying data.

[0071] S3: Construct an adaptive time series analysis and prediction model. By introducing BiLSTM (Bi-directional Long Short-Term Memory) with an attention mechanism, fully extract and mine the spatiotemporal features in traffic flow data. Its basic structure includes: BiLSTM hidden layer, attention layer, drop layer and fully connected layer.

[0072] S4: Using the stabilized data matrix and traffic state classification results, train the traffic parameter estimation model, perform model validation and adjustment, and complete the model training.

[0073] S5: Dynamically input the traffic data matrix into the trained traffic parameter estimation model to obtain dynamic short-term traffic flow estimation parameters for highways.

[0074] In one embodiment of the present invention, such as Figure 2 As shown, step S1 includes the following steps S11 to S16.

[0075] Step S11: Collect microscopic vehicle motion parameters on the highway:

[0076] In this embodiment, a basic section of a highway (approximately 50 kilometers long) is selected, within which multiple radar detector sections and multiple ETC gantry sections are deployed to provide the required traffic parameters, including vehicle speed and traffic flow. This highway traffic parameter prediction uses 18 days (24 hours a day) of detection data as the training set and 10 days of detection data as the test set.

[0077] Step S12: Perform anomaly analysis on the collected data, clean the original dataset, and filter and delete irrelevant and duplicate data. Handle missing and outlier values ​​by checking for missing and outlier values ​​in the original dataset and replacing them with the mean of the values ​​above and below the missing and outlier values.

[0078] Step S13: Construct the sample, selecting traffic flow, average vehicle speed and road physical structure from the repaired sample as influencing factors.

[0079] Step S14: Use the EMD model to perform stationarization decomposition on the sample data, decomposing the time series data into several intrinsic mode functions. EMD model expression formula:

[0080]

[0081] Where I(t) is the input signal; IMF i (t) represents the i-th eigenmode function; R k(t) represents the residual sequence; k is the number of intrinsic mode functions;

[0082] The original flow and velocity time series data are input signals into the EMD algorithm for decomposition. The original flow and velocity time series data are decomposed into multiple intrinsic mode functions and a residual sequence, respectively. The flow and velocity time series data after EMD decomposition reduce the errors caused by data fluctuations.

[0083] Step S15: The multiple intrinsic mode components of the decomposed flow and velocity are fused into a matrix as the data input matrix of the adaptive time series analysis prediction model, thereby improving the estimation accuracy.

[0084] Step S16: Normalize the input data using the max-min normalization method. After linear transformation, the result is mapped to [0, 1]. The specific calculation formula is as follows:

[0085]

[0086] Where, x scale The data represents the data normalized by the Min-Max Scaler; x represents the input data; x max ,x min These represent the maximum and minimum values ​​of the input data, respectively.

[0087] In one embodiment of the present invention, such as Figure 3 As shown, step S2 constructs a traffic state evaluation index system at two scales: cross-section level and road segment level, specifically including the following sub-steps S21 to S22.

[0088] Step S21: Select cross-section level indicators to describe the traffic status and its changing patterns of a single cross-section, including cross-section traffic flow, average cross-section speed, and cross-section physical structure.

[0089] Step S22: Select road segment level indicators to describe the traffic conditions and their changing patterns within the road segment, including road segment traffic volume, average travel speed, road physical structure, and traffic volume in kilometers.

[0090] In one embodiment of the present invention, such as Figure 4 As shown, step S2 uses an improved FCM algorithm to classify traffic states and improves the accuracy of parameter estimation by classifying data. Specifically, it includes the following sub-steps S23 to S26.

[0091] Step S23: Initialize cluster centers and membership matrix, and perform equalization correction on sample size:

[0092] First, for n sample data X n ={X1,X2,X3,2,Xn Perform initial clustering to obtain the initial category to which each sample belongs.

[0093] Then, based on the traditional FCM algorithm, the sample size attribute is introduced to improve the objective function J. The improved objective function J′ is given by the formula:

[0094]

[0095] Where, d ij u represents the distance metric between j data samples and the i-th cluster category. ij u is Let J represent the membership degree of the j-th data sample to the i-th cluster category and the membership degree of the s-th data sample to the i-th cluster category, respectively. i ′ represents the objective function for the improved i-th cluster category, and m, n, and c represent the membership factor, the number of data samples, and the number of cluster samples, respectively.

[0096] Step S24: Calculate the cluster centers based on the membership matrix and update the membership matrix; iterate the calculation of the cluster centers until ||c (r+1) -c (r) <ε||;

[0097] Among them, c (r) c (r+1) Let represent the cluster centers in the r-th and r+1-th iterations, respectively, and ε represent the iteration stopping threshold.

[0098] Step S25: Perform validation using cross-validation:

[0099] Cross-validation exhaustively tests every sample, making it a comprehensive and highly accurate validation method. The basic idea of ​​cross-validation is that for n data samples, in each iteration, one value is selected as the validation set, and the remaining n-1 values ​​are used as the training set. This process is repeated iteratively, ensuring that each sample point is used as part of the validation set.

[0100] Take the first n-1 data X n-1 ={X1,X2,X3,...,X n-1 To re-cluster the training set and obtain new membership degrees and cluster centers, X n Given the test set, determine X. n If the clustering result of the test set is inconsistent with the initial category, the number of errors n′ = 1 is calculated. This process is repeated iteratively. Each time the clustering result of the test set is inconsistent with the initial category, n′ + 1 is calculated. The accuracy of the model clustering can be expressed by the formula:

[0101]

[0102] Where P represents the accuracy of the clustering model; n′ represents the number of clustering errors, and the larger the P, the better the model performance.

[0103] Step S26: Output cluster centers and accuracy to complete traffic state classification.

[0104] In one embodiment of the present invention, such as Figure 5 As shown, step S3 constructs an adaptive time series analysis prediction model, which utilizes a BiLSTM model with an attention mechanism to fully extract and mine the spatiotemporal features in traffic flow data. Specifically, it includes the following sub-steps S31 to S34.

[0105] Step S31: Construct the BiLSTM hidden layer. The BiLSTM model is trained from two directions and the training structure is linearly fused, so it can effectively capture the bidirectional influence of time series. The formula for calculating the forward LSTM is:

[0106]

[0107] Inverse LSTM calculation formula:

[0108]

[0109] Forward and backward LSTM fusion output h:

[0110] h = [h] F,t ,h B,t ] = W F h F,t +W B h B,t +b

[0111] Where σ is the activation function; b is the corresponding bias value; and W f W B These are the weight matrices for the hidden layers of the forward and backward LSTM models, respectively, h. F,t For the output of the forward LSTM calculation, h B,t The output result is calculated for the inverse LSTM.

[0112] Step S32: Construct an Attention layer. Based on the BiLSTM model, introduce an attention mechanism to calculate the different weights of the hidden layer output vector of the BiLSTM model, using the output vector H of the hidden layer in the BiLSTM model at time t as an example. t As input, the attention weight α is calculated. t Calculation formula:

[0113] u t =tanh(Wh) t +b)

[0114]

[0115] Let the output of the Attention layer at time t be S. t Calculation formula:

[0116]

[0117] Among them, u t The output vector h of the BiLSTM model at time t t The importance of the result; W is the weight matrix; b is the bias value.

[0118] Step S33: Construct a Drop layer and use the dropout function to randomly remove neurons during training to prevent the neural network from overfitting during training.

[0119] Step S34: Construct a fully connected layer, select the sigmoid function as the activation function, expand the feature matrix into a one-dimensional vector, and output the prediction results of traffic flow state parameters.

[0120] In one embodiment of the present invention, such as Figure 6 As shown, step S4 includes the following sub-steps S41 to S46.

[0121] Step S41: Divide the data into a training set and a test set. The training set is used to debug the estimation model, and the test set is used to test the actual learning ability of the model.

[0122] In this embodiment, the training set and the test set are divided into 64% and 36% respectively.

[0123] Step S42: Using a set error value and a maximum number of iterations, train the model in each iteration. If the result is less than the target error value, end the iteration; otherwise, update the network parameter values ​​and continue the iteration.

[0124] Step S43: Select the Adam algorithm to optimize the model weights.

[0125] Step S44: Train the model based on the training data and the set hyperparameters.

[0126] Step S45: Use the grid search method to perform hyperparameter tuning.

[0127] Different combinations of parameters in a neural network will predict the results. Therefore, grid search is used to adjust parameters such as the number of units, number of iterations, and time window size of the adaptive time series analysis prediction model in order to find the optimal combination of parameters.

[0128] In this embodiment, as Figure 7As shown, the root mean square error (RMSE) and mean absolute error (MAE) are used to evaluate the model's predictive performance. Larger RMSE and MAE values ​​indicate larger errors, while RMSE is more sensitive to larger deviations, and MAE is more sensitive to smaller errors. In this embodiment, the RMSE and MAE indicators for both flow rate and velocity prediction achieve good results, outperforming traditional ARIMA, LSTM, and BiLSTM models.

[0129] After completing the traffic model training of this embodiment of the invention, step S5 inputs traffic data into the trained traffic state estimation model to obtain the dynamic traffic state parameters of the highway.

[0130] In this embodiment of the invention, an electronic device is also provided, comprising: one or more processors; a storage device having one or more programs stored thereon; when the one or more programs are executed by the one or more processors, the one or more processors implement the improved FCM state division method for dynamic estimation of highway traffic parameters described in any of the above embodiments.

[0131] In this embodiment of the invention, a computer-readable storage medium is also provided, on which a computer program is stored. When the program is executed by a processor, it implements the steps in any of the highway traffic parameter estimation methods based on evaluation indicators to improve FCM in the above embodiments.

[0132] This embodiment presents a dynamic estimation method for highway traffic parameters based on an improved FCM state classification using an evaluation index system. It decomposes time-series data to reduce interference from data fluctuations and improve data quality. A traffic state evaluation index system is constructed at both the cross-section and road segment levels. An improved FCM algorithm is used for traffic state classification, and parameter estimation accuracy is improved through data classification. An adaptive time series analysis and prediction model is built, performing forward and backward learning on long-term time-series data to fully extract effective information, deeply explore the spatiotemporal characteristics of traffic flow data, and assign weights to effective information. This method demonstrates good performance for short-term highway traffic parameter estimation and can improve traffic estimation accuracy.

[0133] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. It should be noted that the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various improvements and modifications can be made without departing from the spirit of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for estimating highway traffic parameters based on an improved FCM (Functional Traffic Management) index, characterized in that: Includes the following steps: S1. Collect highway traffic data and preprocess it to construct a stable traffic data matrix; S2. Construct a traffic state evaluation index system at two scales: cross section level and road segment level. Introduce the sample size attribute to improve the fuzzy C-means clustering algorithm, classify traffic states, and use classified data to improve the estimation accuracy of highway traffic parameters. S3. Construct an adaptive time series analysis and prediction model, and extract and mine the spatiotemporal features in traffic flow data by introducing a bidirectional long short-term memory network model with an attention mechanism. S4. Using the stabilized traffic data matrix and traffic state division results, train the traffic parameter estimation model, verify and adjust the traffic parameter estimation model, and complete the training of the traffic parameter estimation model. S5. Dynamically input the traffic data matrix into the trained traffic parameter estimation model to obtain dynamic estimation parameters of highway traffic parameters; In step S2, a traffic state evaluation index system is constructed at two scales: cross-section level and road segment level. An improved FCM algorithm is used to classify traffic states, and the accuracy of parameter estimation is improved by using classified data. The process includes the following steps: S21. Select cross-sectional level indicators to describe the traffic conditions and their changing patterns of a single cross-section; the cross-sectional level indicators include: cross-sectional traffic flow, cross-sectional average vehicle speed, and cross-sectional physical structure. S22. Select road segment-level indicators to describe the traffic conditions and their changing patterns within the road segment; the road segment-level indicators include: road segment traffic volume, road segment average travel speed, road segment physical structure, and road segment traffic volume in kilometers. S23. Initialize cluster centers and membership matrices, and adjust sample size for balance; introduce sample size attribute to the objective function based on the traditional FCM algorithm. Improvements are made, and the improved objective function is... For the formula: ； in, This represents the distance metric between j data samples and the i-th cluster category. , Let represent the membership degree of the j-th data sample to the i-th cluster category and the membership degree of the s-th data sample to the i-th cluster category, respectively. Let represent the objective function for the improved i-th cluster category, where m, n, and c represent the membership factor, the number of data samples, and the number of cluster samples, respectively. S24. Calculate cluster centers based on the membership matrix and update the membership matrix; iterate the calculation of cluster centers until... ; in, , Let these represent the cluster centers in the r-th and r+1-th iterations, respectively. Indicates the iteration stopping threshold; S25. Using cross-validation, for n data samples, each iteration takes one value as the validation set and the remaining n-1 values ​​as the training set. This process is repeated until every sample point has been used as the validation set. The accuracy formula for model clustering is as follows: ； in, Indicates the accuracy of the clustering model. Indicates the number of clustering errors; S26. Output cluster centers and accuracy to complete traffic state classification.

2. The method for estimating highway traffic parameters based on an improved FCM using evaluation indices as described in claim 1, characterized in that, Step S1 involves collecting and preprocessing highway traffic data, including the following sub-steps: S11. Collect raw data of microscopic vehicle motion parameters on the highway; S12. Handle missing values ​​in the original data: Check for missing and outlier values ​​in the original data, and replace the missing and outlier values ​​with the mean of the upper and lower values ​​of the missing and outlier values. S13. Construct a sample, selecting traffic flow, average vehicle speed, and road physical structure as influencing factors input samples; S14. The empirical mode decomposition method is used to perform stationarization decomposition on the sample data, decomposing the time series data into several intrinsic mode functions. The empirical mode decomposition formula is as follows: ； in, For input signals; For the first One intrinsic mode function; The sequence represents the residuals; k represents the number of intrinsic mode functions. S15. Reassemble the decomposed components to construct a stable data matrix; S16. The input data is normalized using the maximum-minimum normalization method, and the processed data is divided into a training set and a test set. The normalized data will be in the interval [0, 1]. The calculation formula is as follows: ； in, This represents the data after normalization using the min-max normalization method; Represents input data; These represent the maximum and minimum values ​​of the input data, respectively.

3. The method for estimating highway traffic parameters based on an improved FCM using evaluation indices as described in claim 1, characterized in that, In step S3, the bidirectional long short-term memory network model is represented as BiLSTM. An adaptive time series analysis and prediction model is constructed. By introducing an attention mechanism into the BiLSTM, the spatiotemporal features in traffic flow data are fully extracted and mined, including the following steps: S31. Constructing BiLSTM hidden layers: The bidirectional long short-term memory neural network model is trained from two directions and the training structure is linearly fused to capture the bidirectional influence of time series. S32. Constructing the Attention Layer: Introduce an attention mechanism on the basis of the BiLSTM model and integrate the attention mechanism with the BiLSTM model; S33. Construct a Drop layer and use the dropout function to randomly delete neurons during training to prevent the neural network from overfitting during training. S34. Construct a fully connected layer, select the sigmoid function as the activation function, and output the predicted value of the adaptive time series analysis prediction model.

4. The method for estimating highway traffic parameters based on an improved FCM using evaluation indices as described in claim 3, characterized in that, In step S31, the bidirectional long short-term memory neural network model is trained from two directions and the training structures are linearly fused, including forward LSTM computation and backward LSTM computation, and the forward and backward LSTM are fused and output. The formula is as follows: ； in, This corresponds to the bias value. These are the weight matrices of the hidden layers of the forward and backward LSTM models, respectively. The output results are calculated for the forward LSTM. The output result is calculated for the inverse LSTM.

5. The method for estimating highway traffic parameters based on an improved FCM using evaluation indices as described in claim 3, characterized in that, In step S32, an attention mechanism is introduced based on the BiLSTM model to calculate different weights of the hidden layer output vector of the BiLSTM model. The attention mechanism uses the output vector of the hidden layer in the BiLSTM model at time t as the basis for the calculation. As input, calculate attention weights Calculation formula: ； ； set up Attention layer output at each time step Calculation formula: ； in, The output vector of the BiLSTM model at time t Impact on the importance of the outcome; It is a weight matrix; This is the bias value.

6. The method for estimating highway traffic parameters based on an improved FCM using evaluation indices according to claim 1, characterized in that, In step S4, the traffic parameter estimation model is trained using the stabilized data matrix and the traffic state classification results, including the following steps: S41. Divide the data into a training set and a test set; S42. Set the error value and the maximum number of iterations. Train the model in each iteration. If the result is less than the target error value, end the iteration. Otherwise, update the network parameter values ​​and continue the iteration. S43. Select the Adam algorithm to optimize model weights; S44. Train the model based on the training data and the set hyperparameters; S45. Evaluate the model's performance using the validation set and adjust the model as needed; S46. Grid search for optimal parameters: Use grid search to adjust the number of BiLSTM layer cells, number of iterations, and time window size to find the optimal combination of parameters.

7. The method for estimating highway traffic parameters based on an improved FCM using evaluation indices as described in claim 1, characterized in that, In step S5, the traffic data matrix is ​​dynamically input into the traffic parameter estimation model trained in step S4 to obtain the dynamic short-term traffic flow estimation parameters of the highway, which are used to dynamically evaluate the traffic status of the road network.

8. An electronic device, characterized in that, include: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the steps in the highway traffic parameter estimation method based on the improved FCM according to any one of claims 1 to 7.

Images