Traffic time sequence low-cost storage and data reconstruction method based on feature clustering
By performing time-frequency decomposition and supervised clustering of traffic time series, combining road topology directed graphs, a feature semantic library is constructed, which solves the problems of high cost and data reconstruction accuracy of traditional storage methods, and realizes low-cost storage and efficient data reconstruction. Support the development of smart transportation.
Patent Information
- Application Number
- CN202411042753.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-31
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-07-31
AI Technical Summary
The traditional traffic time series data storage method cannot meet the needs of big data, resulting in high storage costs. In addition, existing compression methods are difficult to balance compression ratio and reconstruction accuracy when reconstructing data, and lack versatility and robustness.
By performing time-frequency decomposition of segmented traffic time series, inputting the road topology directed graph into the constructed traffic sequence compression and reconstruction depth model, supervised clustering is carried out to build a feature semantic library, and using this library for data compression and reconstruction.
Reduces the storage space requirement for large-scale traffic flow data, reduces storage costs, and can accurately reconstruct data when data is damaged or lost, ensuring data integrity and accuracy, while providing strong data support to promote smart transportation technology development.
Smart Images

Figure CN118965029A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of intelligent transportation and artificial intelligence technology, and specifically relates to a low-cost storage and data reconstruction method for traffic time series based on feature clustering. Background Art
[0002] The intelligent traffic management system relies on a large amount of traffic time series data, providing rich and important basic data for traffic management, road planning, travel services and related research and development work. Through in-depth analysis and application of these data, the intelligent traffic management system can not only realize the efficient scheduling of road resources and improve road traffic efficiency, but also enhance road safety, optimize people's travel experience, and ultimately promote the sustainable development of the entire intelligent transportation field. However, with the increasing perfection of the intelligent transportation system, the scale of traffic time series data has also shown explosive growth. Traditional data storage methods can no longer meet the needs of big data storage and bring high storage costs. For example, the DAIR-V2X vehicle-road cooperative trajectory prediction dataset provides more than 200,000 traffic segments, each segment is 10 seconds, and the total length is about 556 hours of traffic data. The dataset size exceeds 30G; the highway network PeMS traffic dataset collects data from sensors on about 40,000 highways in real time, processes about 2GB of time series data every day, and stores more than 10 years of data in total. Therefore, it is necessary to study a low-cost data storage method and realize the reconstruction of original data when necessary, so as to store and manage traffic time series data more efficiently and economically.
[0003] Traffic time series data is the basis for the realization of smart transportation. Its collection, format and storage methods are as follows:
[0004] The collection of traffic time series usually relies on various sensors, monitoring devices and data sources. These devices and sources can capture information related to vehicles and traffic environment, providing rich traffic data. The collection methods of traffic time series data include but are not limited to traffic sensors, GPS devices, traffic light control systems, weather sensors, vehicle sensors and various traffic applications and platforms. Traffic sensors, such as ground induction coils, infrared sensors and microwave sensors, are used to detect the presence, number and speed of vehicles; GPS devices provide vehicle location, speed and driving trajectory information for vehicle tracking, travel time measurement and real-time navigation; traffic light control systems record traffic light control signals; weather sensors provide meteorological information such as temperature, precipitation, visibility and wind speed for analyzing the impact of weather on traffic; vehicle sensors include speed sensors, ABS sensors, tire pressure sensors, etc., which are used to provide vehicle status and performance data to support vehicle performance monitoring and health management; traffic applications and platforms, such as traffic navigation applications and smart traffic management systems, collect real-time traffic data through users' mobile phones or vehicle-mounted devices to provide real-time traffic information, navigation services and traffic management. These collection methods are often used in combination to provide comprehensive traffic time series data to support the needs of applications such as traffic management, travel services, and research.
[0005] Existing low-cost storage solutions for traffic data mainly include lossless compression and lossy compression. Lossless compression is a method that can compress data without losing any data, and the compressed data can be completely restored to its original state. Existing lossless compression methods include Huffman coding, Run-Length coding, LZW algorithm, etc. Lossful compression is a method that loses some data during the compression process and can achieve a higher compression ratio. Existing lossy compression methods include sequence approximation methods and sequence transformation methods. Sequence approximation methods include low-density sampling, multi-sequence aggregation, segment representation, etc.; sequence transformation methods include Fourier transform, discrete cosine transform, wavelet transform, etc.
[0006] The above lossless compression methods all have the disadvantage of relatively low compression ratio and are not suitable for compressing large amounts of data. The above lossy compression sequence approximation methods such as low-density sampling, multi-sequence aggregation, and segment representation can achieve a higher compression ratio, but will lose a large amount of data, which is not conducive to data reuse and accurate analysis. The sequence transformation method of the above lossy compression method is affected by the time-frequency characteristics of the data itself, the selection of transformation parameters, etc., and it is difficult to balance the compression ratio and reconstruction accuracy, and lacks versatility and robustness. Summary of the invention
[0007] In order to solve the above problems existing in the prior art, the present invention provides a maintenance method and system for train equipment failure. The technical problem to be solved by the present invention is achieved through the following technical solutions:
[0008] In a first aspect, the present invention provides a low-cost storage and data reconstruction method for traffic time series based on feature clustering, comprising:
[0009] S100, by performing time-frequency decomposition on the segmented traffic time series, an n1-dimensional frequency sequence is obtained, and the sequence is combined with the road topology directed graph and input into the constructed traffic sequence compression and reconstruction deep model;
[0010] S200, executing the training process of the traffic sequence compression and reconstruction deep model, and in the training process, first extracting the features of the input object and then performing supervised clustering to obtain an n2-dimensional feature class set, and forming a feature semantic library, and repeatedly adjusting the classification parameters of the supervised clustering by designing a loss function until the training cutoff condition is reached, and obtaining a traffic time series feature database when the training cutoff condition is reached;
[0011] S300, according to the time series timestamp and the road network in the reconstruction requirements, extract the sequence initial value, the n2-dimensional feature class set and the road topology directed graph corresponding to the road network from the traffic time series feature database, and use the three to reconstruct the reconstructed traffic flow time series.
[0012] In a first aspect, the present invention provides a low-cost storage and data reconstruction device for traffic time series based on feature clustering, comprising:
[0013] An input module is configured to obtain an n1-dimensional frequency sequence by performing time-frequency decomposition on the segmented traffic time series, and input the sequence into the constructed traffic sequence compression and reconstruction deep model in combination with the road topology directed graph;
[0014] The training module is configured to execute the training process of the traffic sequence compression and reconstruction deep model, and in the training process, the features of the input object are first extracted and then supervised clustering is performed to obtain an n2-dimensional feature class set, and a feature semantic library is formed, and the classification parameters of the supervised clustering are repeatedly adjusted by designing a loss function until a training cutoff condition is reached, so as to obtain a traffic time series feature database when the training cutoff condition is reached;
[0015] The reconstruction module is configured to extract the sequence initial value, the n2-dimensional feature class set and the road topology directed graph corresponding to the road network from the traffic time series feature database according to the time series timestamp and the road network in the reconstruction requirements, and use the three to reconstruct the reconstructed traffic flow time series.
[0016] Beneficial effects:
[0017] (1) The present invention reduces the space required for storing large-scale traffic flow data, thereby reducing storage costs.
[0018] (2) When data is damaged or lost, the method of the present invention can be used to reconstruct the damaged or lost data, thereby ensuring the integrity and accuracy of the data.
[0019] (3) The method of the present invention provides strong data support for the intelligent traffic management system, thereby promoting the development of intelligent traffic technology. It can also be used for the prediction of traffic time series and has broad application prospects.
[0020] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] Figure 1 It is a flow chart of a low-cost storage and data reconstruction method of traffic time series based on feature clustering provided by the present invention;
[0022] Figure 2 It is an input and output diagram of the traffic time series feature database provided by the present invention;
[0023] Figure 3 It is a flow chart of the training of the traffic sequence compression and reconstruction deep model based on feature clustering provided by the present invention;
[0024] Figure 4 It is a flow chart of traffic time series reconstruction provided by the present invention;
[0025] Figure 5 It is a schematic diagram of the prediction results of the predicted traffic time series provided by the present invention. DETAILED DESCRIPTION
[0026] The present invention is further described in detail below with reference to specific embodiments, but the embodiments of the present invention are not limited thereto.
[0027] First, as Figure 1 As shown, the present invention provides a low-cost storage and data reconstruction method for traffic time series based on feature clustering, comprising:
[0028] S100, by performing time-frequency decomposition on the segmented traffic time series, an n1-dimensional frequency sequence is obtained, and the sequence is combined with the road topology directed graph and input into the constructed traffic sequence compression and reconstruction deep model;
[0029] In this step, the traffic time series is represented by a, and all segmented traffic time series are represented by [a1, a2, …a i, ]express.
[0030] S200, executing the training process of the traffic sequence compression and reconstruction deep model, and in the training process, first extracting the features of the input object and then performing supervised clustering to obtain an n2-dimensional feature class set, and forming a feature semantic library, and repeatedly adjusting the classification parameters of the supervised clustering by designing a loss function until the training cutoff condition is reached, and obtaining a traffic time series feature database when the training cutoff condition is reached;
[0031] refer to Figure 2 The traffic time series feature database includes a road topology directed graph database, a traffic segmentation time series category library and a feature semantic library; the road topology directed graph is stored in the road topology directed graph database; the sequence initial value and the corresponding category are stored in the traffic segmentation time series category library; the n2-dimensional traffic sequence clustering feature f x The categories to which they belong are stored in the feature semantic library; and there is a data index relationship between the traffic segment time series category library and the feature semantic library.
[0032] The traffic time series category feature library stores a traffic time series category feature set in Refers to the category features of any dimension, saving the k after clustering i The traffic segment time series category library stores the timestamp, the road to which the segment traffic time series belongs, and the n2-dimensional category label data.
[0033] S300, according to the time series timestamp and the road network in the reconstruction requirements, extract the sequence initial value, the n2-dimensional feature class set and the road topology directed graph corresponding to the road network from the traffic time series feature database, and use the three to reconstruct the reconstructed traffic flow time series.
[0034] As an optional implementation manner of the present invention, S100 includes:
[0035] S110, dividing the traffic time series into equally spaced segmented traffic time series, constructing a road topology directed graph using the road grid distances between traffic time series acquisition devices, and normalizing both the segmented traffic time series and the road topology directed graph;
[0036] S120, decomposing the normalized segmented traffic time series into an n1-dimensional frequency series by using a time-frequency transform decomposition method, and inputting the frequency series together with the normalized road topology directed graph into the constructed traffic sequence compression and reconstruction deep model.
[0037] The time-frequency transform decomposition method of the present invention includes but is not limited to Fourier transform, wavelet transform, Hilbert transform and various improved time-frequency transforms.
[0038] As an optional implementation manner of the present invention, S110 includes:
[0039] S111, using a traffic time series collection device to collect a traffic time series, and dividing it into equally spaced segmented traffic time series; wherein the traffic time series includes motor vehicle flow, non-motor vehicle flow, traffic flow speed, traffic occupancy rate, meteorological information and vehicle status;
[0040] S112, calculating the road grid distance between the traffic time series collection devices, and processing it into an adjacency matrix data structure, thereby obtaining a road topology directed graph;
[0041] S113, normalizing the segmented traffic time series and the road topology directed graph by using a minimum-maximum normalization method to obtain a normalized segmented traffic time series and a normalized road topology directed graph.
[0042] Among them, the length of the equally spaced segmented traffic time series is Assume that the total length of the traffic time series is l a , then any To obtain a higher compression ratio. The predetermined formula of the minimum and maximum value normalization method is: where x nor is the standardized value, x is the original value of the sequence, and x min is the minimum value of the sequence, x max is the maximum value of the sequence.
[0043] Combination Figure 3 and Figure 4 As an optional implementation of the present invention, S200 includes: executing a training process for compressing and reconstructing a deep model of a traffic sequence, the training process including the following steps:
[0044] a. In the current training round, the n2-dimensional traffic sequence feature vector is extracted from the n1-dimensional frequency sequence using the traffic sequence compression and reconstruction deep model, and then the feature vector is clustered to obtain the multi-class features of each dimension in the n2 dimensions; the multi-class features of each dimension in the n2 dimensions, the road topology directed graph and the sequence initial value are combined into a traffic time series feature database; the features in the traffic time series feature database are fused and reconstructed with the deep features of the road topology directed graph to obtain a reconstructed segmented traffic time series;
[0045] b. Calculate the total loss function using the reconstructed segmented traffic time series and the segmented traffic time series in S100, and use the loss function to adjust the weight of the traffic sequence compression and reconstruction deep model of each training round; use the traffic sequence compression and reconstruction deep model after adjusting the weight as the traffic sequence compression and reconstruction deep model of the next training round;
[0046] e. Repeat a to b until the traffic sequence compression and reconstruction deep model reaches the training cutoff condition, and obtain the traffic time series feature database when the training cutoff condition is reached.
[0047] Among them, the training cutoff conditions include reaching the maximum number of training rounds or the loss function value no longer decreasing.
[0048] refer to Figure 3 , the traffic sequence compression and reconstruction deep model of the present invention adopts an automatic encoder encoder-decoder structure, including: a traffic sequence feature extraction encoding module, a road topology structure directed graph feature extraction module, a feature supervised clustering module, and a sequence reconstruction decoding module;
[0049] The traffic sequence feature extraction encoding module, the road topology structure directed graph feature extraction module, the feature supervised clustering module, and the sequence reconstruction decoding module are composed of multiple layers of neural network layers, including but not limited to Transformer layer, long short-term memory (LSTM) neural network layer, convolutional neural network (CNN) layer, fully connected layer (DNN) and various improved neural network layers. The activation functions used in the neural network layer include but are not limited to Rule, Tanh, Softmax, Sigmid and various improved activation functions.
[0050] The input of the traffic sequence feature extraction encoding module is the segmented traffic time sequence a i The decomposed n1-dimensional frequency sequence is output as the n2-dimensional traffic sequence feature vector f r , where n1≥n2;
[0051] The road topology directed graph feature extraction module inputs the road topology directed graph and outputs the depth feature f of the road topology directed graph. g ;
[0052] The input of the feature supervised clustering module is the n2-dimensional traffic sequence feature vector f r , the output is the n2-dimensional traffic sequence clustering feature f x and the category to which it belongs;
[0053] The input of the sequence reconstruction decoding module is the segmented traffic time sequence a i At least one sequence initial value, n2-dimensional traffic sequence clustering feature f x And the deep feature f g , the output is the reconstructed traffic time series a i .
[0054] Combination Figure 3 and Figure 4 In an optional embodiment, a includes:
[0055] a1, in the current training round, the traffic sequence feature extraction encoding module of the current training round is used to extract the features of the n1-dimensional frequency sequence, and output the n2-dimensional traffic sequence feature vector f r The normalized road topology directed graph is extracted using the road topology structure directed graph feature extraction module of the current training round. g , and output to the sequence reconstruction decoding module of the current training round;
[0056] a2, using the supervised clustering module of the current training round to cluster the n2-dimensional traffic sequence feature vector f r Each dimension of is classified to output n2-dimensional traffic sequence clustering feature f x The n2-dimensional traffic sequence is clustered into features f x The feature semantic library of the current training round is composed of the category it belongs to, and the feature semantic library of the current training round and the segmented traffic time series a i The sequence initial value and the normalized road topology directed graph in compose the traffic time series feature database of the current training round;
[0057] a3, select a segment traffic time series a from the traffic time series feature database of the current training round i , the normalized road topology directed graph and the n2-dimensional traffic sequence clustering feature f x Input into the sequence reconstruction decoding module of the current training round, and use the sequence reconstruction decoding module to first perform feature fusion and then feature extraction to output the segmented traffic time series reconstructed under the current training round
[0058] refer to Figure 3 In an optional embodiment, b includes:
[0059] b1, using the segmented traffic time series reconstructed under the current training round and segmented traffic time series a i , calculate the first loss value of the current training round;
[0060] Among them, the first loss value is the loss value calculated by the traffic sequence reconstruction loss loss1. The traffic sequence reconstruction loss loss1 is the error between the original traffic flow and the reconstructed traffic flow, which is used for the training and evaluation of the traffic sequence feature extraction encoding module, the road topology structure directed graph feature extraction module, and the sequence reconstruction decoding module.
[0061] b2, the segmented traffic time series reconstructed under the current training round and segmented traffic time series a i Both perform n2-dimensional time-frequency transform decomposition, and calculate the sum of the errors between each dimensional frequency sequences after the decomposition;
[0062] b3, taking the sum of the errors as the third loss value of the current training round, and calculating the second loss value of the current training round in combination with the first loss value of the current training round;
[0063] The second loss value is a loss value calculated by the error sum loss3, where loss3 is the sum of errors between frequency signals of each dimension after the reconstructed traffic flow and the original traffic flow are decomposed by n2-dimensional time-frequency transformation.
[0064] b4, using the first loss value of the current training round to adjust the weights of the sequence reconstruction decoding module and the road topology structure directed graph feature extraction module of the current training round, and using the second loss value of the current training round to calculate the weight of the feature supervised clustering module of the current training round;
[0065] Among them, the second loss value is the loss value calculated by the supervised clustering loss loss2, the supervised clustering loss loss2 = loss1 + loss3, and loss2 is used for the training and evaluation of the feature supervised clustering module.
[0066] b5, the adjusted sequence reconstruction decoding module, road topology structure directed graph feature extraction module, and feature supervised clustering module are used as the sequence reconstruction decoding module, road topology structure directed graph feature extraction module, and feature supervised clustering module of the next training round.
[0067] In an optional embodiment, the present invention can also update the road topology directed graph in the road topology directed graph database by adding vertices, directed edges, and distance weights; set an update threshold for the traffic segment time series category library, and when the threshold is met, update the sequence initial value and the category to which it belongs in the traffic segment time series category library.
[0068] It is worth noting that the operation mode of the traffic time series feature database includes the update of the road topology directed graph, the update of the traffic segmentation time series category, and the update of the feature semantic data. The road topology directed graph update updates the road topology network by adding vertices, directed edges, and distance weights. The feature semantic data update refers to the generation of new traffic time series categories and features in the traffic time series due to factors such as road structure, urban spatial structure, and traffic environment. A traffic time series category feature update threshold is set, and when the threshold is met, the traffic time series category feature is updated. The traffic segmentation time series category data update performs feature extraction on any newly added real-time traffic segmentation time series and clusters it into n2-dimensional feature classes, and updates the segmentation time series category data.
[0069] In a second aspect, the present invention provides a low-cost storage and data reconstruction device for traffic time series based on feature clustering, comprising:
[0070] An input module is configured to obtain an n1-dimensional frequency sequence by performing time-frequency decomposition on the segmented traffic time series, and input the sequence into the constructed traffic sequence compression and reconstruction deep model in combination with the road topology directed graph;
[0071] The training module is configured to execute the training process of the traffic sequence compression and reconstruction deep model, and in the training process, the features of the input object are first extracted and then supervised clustering is performed to obtain an n2-dimensional feature class set, and a feature semantic library is formed, and the classification parameters of the supervised clustering are repeatedly adjusted by designing a loss function until a training cutoff condition is reached, so as to obtain a traffic time series feature database when the training cutoff condition is reached;
[0072] The reconstruction module is configured to extract the sequence initial value, the n2-dimensional feature class set and the road topology directed graph corresponding to the road network from the traffic time series feature database according to the time series timestamp and the road network in the reconstruction requirements, and use the three to reconstruct the reconstructed traffic flow time series.
[0073] The effect of the present invention is explained below through simulation.
[0074] For a 24-hour traffic time series, it is segmented into 24 segments, each segment is 1 hour. A directed graph of road topology is calculated. All data are normalized to the interval [0, 1] using the minimum and maximum normalization method. The traffic time series is one of motor vehicle flow, non-motor vehicle flow, traffic flow speed, traffic occupancy rate, meteorological information, and vehicle status.
[0075] Each 1-hour segmented time series is decomposed into a 5-dimensional frequency series by using the time-frequency transformation decomposition method. The input of the traffic sequence feature extraction encoding module is the 5-dimensional frequency sequence after the decomposition of each 1-hour traffic segmented time series, and the output is a 5-dimensional traffic sequence feature vector f r The input of the feature supervised clustering module is the 5-dimensional traffic sequence feature, and the output is the 5-dimensional traffic sequence clustering feature f x The sequence reconstruction decoding module consists of a feature fusion layer, a feature extraction layer, and a multi-layer perceptron (MLP) output layer. The input is the initial value of the 2 time steps of the traffic segmentation time series, the 5-dimensional traffic sequence clustering feature value f x And the road topology directed graph depth feature f g , the output is the reconstructed traffic sequence The feature semantic library stores the traffic time series category feature set X f =[f x1 ,f x2 ,f x3 ,f x4 ,f x5 ]. For the category features of the 5-dimensional feature sequence, each 1-dimensional feature sequence is clustered into 10 categories, and the labels and their feature values are saved.
[0076] The segmented traffic time series category library stores the timestamp, the road to which it belongs, and the 5-dimensional category label data of the segmented time series. For 365 groups of 24-hour traffic time series data, the traffic time series category feature update threshold is set to 24 times. When the number of similar features occurs 24 times, the traffic time series category feature is updated. The segmented traffic time series category data is updated, and any newly added 1-hour segmented time series is feature extracted and clustered into a 5-dimensional feature class, and the segmented time series category data is updated.
[0077] According to the time series timestamp and the road network to which it belongs, the two time step initial values, the 5-dimensional feature class set and its road topology directed graph are extracted from the traffic time series feature database. The above data are input into the sequence reconstruction decoding module, and the output is the reconstructed traffic time series.
[0078] The present invention can recover data for lost traffic time series. For example, for lost 10-minute traffic time series data, the timestamp and the road section are used to recover the initial values of the two time steps in the previous hour from the traffic time series feature database, the 5-dimensional feature class set and the road topology directed graph. The above data are input into the sequence reconstruction decoding module, and the output is the restored traffic time series.
[0079] The present invention can predict traffic flow time series. The prediction process is as follows: Step 1: collect traffic flow data in real time and perform standardization. Step 2: extract the 5-dimensional feature class set and road topology directed graph of the road within 1 hour based on the traffic time series feature library. Step 3: input the above data into the sequence reconstruction decoding module and output the predicted value of the traffic flow data. Using the public data set PEMS-BAY, the traffic speed prediction with a time range of 60 minutes is performed. The results are as follows: Figure 5 As shown, from Figure 5 The error evaluation indicators are calculated for the prediction results, and the error evaluation indicators are MAE is 1.66, RMSE is 3.44, and MAPE is 3.75%.
[0080] The present invention provides a low-cost storage and data reconstruction method for traffic time series based on feature clustering through methods such as time series frequency decomposition, autoencoder, and supervised clustering, and constructs a traffic time series feature database, which can accurately reconstruct the original data while compressing and storing traffic time series data. The present invention can also be used for the reconstruction of damaged or lost traffic time series data, as well as the prediction of traffic time series data.
[0081] It is worth noting that the terms "first" and "second" in the present invention are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.
[0082] Although the present application is described herein in conjunction with various embodiments, in the process of implementing the claimed application, those skilled in the art may understand and implement other variations of the disclosed embodiments by reviewing the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality of components or steps.
[0083] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.
Claims
1. A low-cost storage and data reconstruction method for traffic time series based on feature clustering, characterized in that: include: S100, by performing time-frequency decomposition on the segmented traffic time series, an n1-dimensional frequency sequence is obtained, and the sequence is combined with the road topology directed graph and input into the constructed traffic sequence compression and reconstruction deep model; S200, executing the training process of the traffic sequence compression and reconstruction deep model, and in the training process, first extracting the features of the input object and then performing supervised clustering to obtain an n2-dimensional feature class set, and forming a feature semantic library, and repeatedly adjusting the classification parameters of the supervised clustering by designing a loss function until the training cutoff condition is reached, and obtaining a traffic time series feature database when the training cutoff condition is reached; S300, according to the time series timestamp and the road network in the reconstruction requirements, extract the sequence initial value, the n2-dimensional feature class set and the road topology directed graph corresponding to the road network from the traffic time series feature database, and use the three to reconstruct the reconstructed traffic flow time series.
2. The low-cost storage and data reconstruction method for traffic time series based on feature clustering according to claim 1 is characterized in that: S100 includes: S110, dividing the traffic time series into equally spaced segmented traffic time series, constructing a road topology directed graph using the road grid distances between traffic time series acquisition devices, and normalizing both the segmented traffic time series and the road topology directed graph; S120, decomposing the normalized segmented traffic time series into an n1-dimensional frequency series by using a time-frequency transform decomposition method, and inputting the frequency series together with the normalized road topology directed graph into the constructed traffic sequence compression and reconstruction deep model.
3. The low-cost storage and data reconstruction method for traffic time series based on feature clustering according to claim 2 is characterized in that: S110 includes: S111, using a traffic time series collection device to collect a traffic time series, and dividing it into equally spaced segmented traffic time series; wherein the traffic time series includes motor vehicle flow, non-motor vehicle flow, traffic flow speed, traffic occupancy rate, meteorological information and vehicle status; S112, calculating the road grid distance between the traffic time series collection devices, and processing it into an adjacency matrix data structure, thereby obtaining a road topology directed graph; S113, normalizing the segmented traffic time series and the road topology directed graph by using a minimum-maximum normalization method to obtain a normalized segmented traffic time series and a normalized road topology directed graph.
4. The low-cost storage and data reconstruction method for traffic time series based on feature clustering according to claim 1 is characterized in that: S200 includes: executing a training process for compressing and reconstructing a deep model of a traffic sequence, wherein the training process includes the following steps: a. In the current training round, the n2-dimensional traffic sequence feature vector is extracted from the n1-dimensional frequency sequence using the traffic sequence compression and reconstruction deep model, and then the feature vector is clustered to obtain the multi-class features of each dimension in the n2 dimensions; the multi-class features of each dimension in the n2 dimensions, the road topology directed graph and the sequence initial value are combined into a traffic time series feature database; the features in the traffic time series feature database are fused and reconstructed with the deep features of the road topology directed graph to obtain a reconstructed segmented traffic time series; b. Calculate the total loss function using the reconstructed segmented traffic time series and the segmented traffic time series in S100, and use the loss function to adjust the weight of the traffic sequence compression and reconstruction deep model of each training round; use the traffic sequence compression and reconstruction deep model after adjusting the weight as the traffic sequence compression and reconstruction deep model of the next training round; e. Repeat a to b until the traffic sequence compression and reconstruction deep model reaches the training cutoff condition, and obtain the traffic time series feature database when the training cutoff condition is reached.
5. The low-cost storage and data reconstruction method for traffic time series based on feature clustering according to claim 4 is characterized in that: The traffic sequence compression and reconstruction deep model includes: a traffic sequence feature extraction encoding module, a road topology structure directed graph feature extraction module, a feature supervised clustering module, and a sequence reconstruction decoding module; The input of the traffic sequence feature extraction encoding module is the segmented traffic time sequence a i The decomposed n1-dimensional frequency sequence is output as the n2-dimensional traffic sequence feature vector f r , where n1≥n2; The road topology directed graph feature extraction module inputs the road topology directed graph and outputs the depth feature f of the road topology directed graph. g ; The input of the feature supervised clustering module is the n2-dimensional traffic sequence feature vector f r , the output is the n2-dimensional traffic sequence clustering feature f x and the category to which it belongs; The input of the sequence reconstruction decoding module is the segmented traffic time sequence a i At least one sequence initial value, n2-dimensional traffic sequence clustering feature f x And the deep feature f g , the output is the reconstructed traffic time series a i .
6. The low-cost storage and data reconstruction method for traffic time series based on feature clustering according to claim 5 is characterized in that: a include: a1, in the current training round, the traffic sequence feature extraction encoding module of the current training round is used to extract the features of the n1-dimensional frequency sequence, and output the n2-dimensional traffic sequence feature vector f r The feature supervised clustering module of the current training round is used, and the road topology directed graph feature extraction module of the current training round is used to extract the normalized road topology directed graph depth feature f g , and output to the sequence reconstruction decoding module of the current training round; a2, using the supervised clustering module of the current training round to cluster the n2-dimensional traffic sequence feature vector f r Each dimension of is classified to output n2-dimensional traffic sequence clustering feature f x The n2-dimensional traffic sequence is clustered into features f x The feature semantic library of the current training round is composed of the category it belongs to, and the feature semantic library of the current training round and the segmented traffic time series a i The sequence initial value and the normalized road topology directed graph in compose the traffic time series feature database of the current training round; a3, select segmented traffic time series a from the traffic time series feature database of the current training round i , the normalized road topology directed graph and n2 dimensional traffic sequence clustering features f x Input into the sequence reconstruction decoding module of the current training round, and use the sequence reconstruction decoding module to first perform feature fusion and then feature extraction to output the segmented traffic time series a reconstructed under the current training round i .
7. The low-cost storage and data reconstruction method for traffic time series based on feature clustering according to claim 6 is characterized in that b include: b1, using the segmented traffic time series a reconstructed under the current training round i and segmented traffic time series a i , calculate the first loss value of the current training round; b2, the segmented traffic time series a reconstructed under the current training round i and segmented traffic time series a i Both perform n2-dimensional time-frequency transform decomposition, and calculate the sum of the errors between each dimensional frequency sequences after the decomposition; b3, taking the sum of the errors as the third loss value of the current training round, and calculating the second loss value of the current training round in combination with the first loss value of the current training round; b4, using the first loss value of the current training round to adjust the weights of the sequence reconstruction decoding module and the road topology structure directed graph feature extraction module of the current training round, and using the second loss value of the current training round to calculate the weight of the feature supervised clustering module of the current training round; b5, the adjusted sequence reconstruction decoding module, road topology structure directed graph feature extraction module, and feature supervised clustering module are used as the sequence reconstruction decoding module, road topology structure directed graph feature extraction module, and feature supervised clustering module of the next training round.
8. The low-cost storage and data reconstruction method for traffic time series based on feature clustering according to claim 6 is characterized in that: The traffic time series feature database includes a road topology directed graph database, a traffic segmentation time series category library and a feature semantic library; the road topology directed graph is stored in the road topology directed graph database; the sequence initial value and the corresponding category are stored in the traffic segmentation time series category library; the n2-dimensional traffic sequence clustering feature f x The categories to which they belong are stored in the feature semantic library; and there is a data index relationship between the traffic segment time series category library and the feature semantic library.
9. The low-cost storage and data reconstruction method for traffic time series based on feature clustering according to claim 8 is characterized in that: The low-cost storage and data reconstruction method of traffic time series based on feature clustering also includes: Updating the road topology directed graph in the road topology directed graph database by adding vertices, directed edges, and distance weights; The update threshold of the traffic segment time series category library is set. When the threshold is met, the initial value and category of the sequence in the traffic segment time series category library are updated.
10. A low-cost storage and data reconstruction device for traffic time series based on feature clustering, characterized in that: include: An input module is configured to obtain an n1-dimensional frequency sequence by performing time-frequency decomposition on the segmented traffic time series, and input the sequence into the constructed traffic sequence compression and reconstruction deep model in combination with the road topology directed graph; The training module is configured to execute the training process of the traffic sequence compression and reconstruction deep model, and in the training process, the features of the input object are first extracted and then supervised clustering is performed to obtain an n2-dimensional feature class set, and a feature semantic library is formed, and the classification parameters of the supervised clustering are repeatedly adjusted by designing a loss function until a training cutoff condition is reached, so as to obtain a traffic time series feature database when the training cutoff condition is reached; The reconstruction module is configured to extract the sequence initial value, the n2-dimensional feature class set and the road topology directed graph corresponding to the road network from the traffic time series feature database according to the time series timestamp and the road network in the reconstruction requirements, and use the three to reconstruct the reconstructed traffic flow time series.
Citation Information
Patent Citations
Traffic prediction method based on enhanced space-time diagram neural network
CN112241814A
Road network traffic state discrimination method based on clustering and graph convolutional network
CN113450562A
Unsupervised radar signal sorting method based on deep clustering
CN113971440A
Bearing fault prediction method based on fuzzy cognitive map
CN115901264A
Traffic flow prediction method based on clustering and heterogeneous graph neural network
CN116681176A
Cited By
Highway information evaluation method based on data management
CN120452189A
