Traffic data supplementing model determination method, traffic data supplementing method and related device

Abstract

The application discloses a traffic data supplement model determination method, a traffic data supplement method and related devices, and relates to the technical field of traffic data supplement. The determination method comprises the following steps: performing random missing processing and missing mapping processing on a complete traffic data set to obtain a missing vector set; training a denoising stack autoencoder by using the complete traffic data set and the missing vector set, obtaining a trained denoising stack autoencoder, and inputting the missing vector set into the trained denoising stack autoencoder to perform dimension increasing processing, thereby obtaining first high-dimensional feature vector data; training a generative adversarial network by using the complete traffic data set and the first high-dimensional feature vector data, thereby obtaining a trained generative adversarial network; and sequentially connecting an encoder of the trained denoising stack autoencoder, the generative adversarial network and a decoder of the denoising stack autoencoder, thereby obtaining a traffic data supplement model. The application can accurately and efficiently supplement missing parts in traffic data.

Description

Technical Field

[0001] This application relates to the field of traffic data compensation technology, and in particular to a method for determining a traffic data compensation model, a traffic data compensation method, and related apparatus. Background Technology

[0002] With the rapid development of Intelligent Transportation Systems (ITS), the collection, analysis, and application of traffic data have become central to modern traffic management and planning. ITS acquires vast amounts of real-time information on traffic flow, speed, and density through sensors, cameras, and other data acquisition devices. This data provides crucial support for traffic flow prediction, congestion management, and route optimization. However, due to sensor malfunctions, data transmission errors, and environmental interference, the collected traffic data often contains missing values. This data gap not only affects the completeness of the data but also poses a serious challenge to the accuracy of subsequent data analysis and model predictions. Therefore, effectively compensating for missing data and restoring the original distribution and statistical characteristics of the data has become a critical issue that urgently needs to be addressed in the field of intelligent transportation systems.

[0003] Existing data imputation techniques, such as mean imputation, linear interpolation, and k-nearest neighbor (kNN), perform reasonably well when dealing with small-scale datasets or datasets with low missing rates, but often fall short when faced with large-scale traffic data with high missing rates. Summary of the Invention

[0004] The purpose of this application is to provide a method for determining a traffic data compensation model, a traffic data compensation method, and related apparatus, which can accurately and efficiently compensate for missing parts in traffic data, providing strong support for traffic data analysis, prediction, and decision-making.

[0005] To achieve the above objectives, this application provides the following solution:

[0006] Firstly, this application provides a method for determining a traffic data compensation model, including:

[0007] Random missing data processing and missing data mapping are performed on the complete traffic dataset to obtain a missing vector set; the complete traffic dataset includes traffic flow data, spatiotemporal data, and environmental data;

[0008] A denoising stacked autoencoder is trained using the complete traffic dataset and the missing vector set to obtain a trained denoising stacked autoencoder.

[0009] The missing vector set is input into the encoder of the trained denoising stacked autoencoder for dimensionality upscaling to obtain the first high-dimensional feature vector data.

[0010] The complete traffic dataset and the first high-dimensional feature vector data are used to train a generative adversarial network to obtain a trained generative adversarial network.

[0011] The encoder of the trained denoising stacked autoencoder, the trained generative adversarial network, and the decoder of the trained denoising stacked autoencoder are sequentially connected to obtain the traffic data compensation model.

[0012] Secondly, this application provides a method for supplementing traffic data, including:

[0013] Obtain raw traffic data;

[0014] The original traffic data is input into the traffic data compensation model to obtain the compensated traffic data; the traffic data compensation model is a model trained according to the traffic data compensation model determination method.

[0015] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the traffic data compensation model determination method or the traffic data compensation method described above.

[0016] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the traffic data compensation model determination method or the traffic data compensation method described above.

[0017] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the traffic data compensation model determination method or the traffic data compensation method described above.

[0018] According to the specific embodiments provided in this application, the following technical effects are disclosed:

[0019] This application provides a method for determining a traffic data compensation model, a traffic data compensation method, and related apparatus. This application can process datasets containing multiple dimensions such as traffic flow, spatiotemporal information, and environmental data, comprehensively considering the impact of various factors on traffic conditions. By using a denoising stacked autoencoder for data processing, it can remove noise from the original data while processing missing data, avoiding errors caused by missing data and improving the predictive performance of subsequent models. Furthermore, generative adversarial networks (GANs) improve the quality of generated data through adversarial training, reduce overfitting during the recovery of missing data, and enhance the model's generalization ability. Combining GANs with a denoising stacked autoencoder allows the model to better capture the inherent patterns in the generated data, accurately and efficiently compensating for missing parts of traffic data, improving data completeness and accuracy, and providing strong support for traffic data analysis, prediction, and decision-making. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is an application environment diagram of a traffic data compensation model determination method according to an embodiment of this application;

[0022] Figure 2 A flowchart illustrating a method for determining a traffic data compensation model according to an embodiment of this application;

[0023] Figure 3 This is a schematic diagram of the structure of a noise reduction autoencoder provided in an embodiment of this application;

[0024] Figure 4 This is a schematic diagram of the structure of a noise-reducing stack-type self-encoder provided in an embodiment of this application;

[0025] Figure 5 This is a schematic diagram of the structure of a traffic data compensation model provided in an embodiment of this application;

[0026] Figure 6 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0027] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0028] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0029] The applicant's research found that traditional methods, such as mean imputation, linear interpolation, and k-nearest neighbors (kNN), typically rely on simple statistics or nearby data points, lacking a deep understanding of the data's inherent structure and patterns. Traffic data exhibits complex spatiotemporal correlations and highly nonlinear characteristics, making it difficult for traditional methods to effectively restore the original data distribution, resulting in significant differences between the imputed data and the true data. Furthermore, traditional methods suffer from low computational efficiency and excessive time complexity when processing large-scale datasets, failing to meet the demands of real-time or near-real-time traffic flow prediction and decision support. Therefore, a novel data imputation method is urgently needed that can efficiently process large-scale traffic data with high missing rates while preserving the original data distribution and statistical properties.

[0030] Therefore, the traffic data compensation model determination method provided in this application embodiment can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on another server. Terminal 102 can send a complete traffic dataset to server 104. After receiving the complete traffic dataset, server 104 performs random missing data processing and defect mapping on the complete traffic dataset to obtain a missing vector set. It then trains a denoising stacked autoencoder using the complete traffic dataset and the missing vector set, obtaining a trained denoising stacked autoencoder, and inputs it into the missing vector set for dimensionality upscaling to obtain the first high-dimensional feature vector data. It then trains a generative adversarial network (GAN) using the complete traffic dataset and the first high-dimensional feature vector data, obtaining a trained GAN. Finally, it sequentially connects the encoder of the trained denoising stacked autoencoder, the GAN, and the decoder of the denoising stacked autoencoder to obtain a traffic data compensation model. Server 104 can then feed back the obtained traffic data compensation model to terminal 102. In addition, in some embodiments, the traffic data compensation model determination method can also be implemented by the server 104 or the terminal 102 separately. For example, the terminal 102 can directly process the complete traffic dataset, or the server 104 can obtain the complete traffic dataset from the data storage system and process it.

[0031] The terminal 102 can be, but is not limited to, various desktop computers, laptops, and IoT devices. The server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers, or it can be a cloud server.

[0032] In one exemplary embodiment, such as Figure 2 As shown, a method for determining a traffic data compensation model is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps 201 to 205. Wherein:

[0033] Step 201: Perform random missing data processing and missing data mapping on the complete traffic dataset to obtain a missing vector set; the complete traffic dataset includes traffic flow data, spatiotemporal data and environmental data.

[0034] Step 202: Train a denoising stacked autoencoder using the complete traffic dataset and the missing vector set to obtain the trained denoising stacked autoencoder.

[0035] Step 203: Input the missing vector set into the encoder of the trained denoising stacked autoencoder for dimensionality upscaling to obtain the first high-dimensional feature vector data.

[0036] Step 204: Train the generative adversarial network using the complete traffic dataset and the first high-dimensional feature vector data to obtain the trained generative adversarial network.

[0037] Step 205: The encoder of the trained denoising stacked autoencoder, the trained generative adversarial network, and the decoder of the trained denoising stacked autoencoder are sequentially connected to obtain the traffic data compensation model.

[0038] By implementing steps 201 to 205 above, a traffic data compensation model can be obtained. This model can better capture the inherent patterns of the data when generating missing data, accurately and efficiently compensate for the missing parts of the traffic data, improve the integrity and accuracy of the data, and provide strong support for traffic data analysis, prediction and decision-making.

[0039] Furthermore, the Denoising Stacked Autoencoder (DSAE) has two basic components: an autoencoder (AE) and a denoising autoencoder (DAE). A single AE extracts features from the raw input data, while stacked AEs can be stacked into a deep neural network to obtain an abstract representation of the input data by progressively extracting features. A single DAE is a stochastic version of an AE, capable of capturing the statistical correlations between inputs. To better compensate for missing traffic data, this application uses DSAE, a model that retains the functional advantages of both DAE and stacked AE. The DAE primarily performs denoising of missing traffic data, while the stacked AE mainly helps extract features from the traffic data. DSAE achieves data recovery through feature extraction and statistical correlation learning. The composition structure and training algorithm of DSAE are described in detail below.

[0040] Specifically, DAE is a variant of AE, and its composition structure is as follows: Figure 3 As shown. The input part of the DAE feeds the missing original input data into the input layer. The purpose of DAE training is to reconstruct the original data from the missing original input data, and the reconstructed original data is obtained from the network's output layer. The missing vector of the original input data x. Obtained through corruption mapping:

[0041]

[0042] Once the mapping is complete, The input layer is fed into the network. Encoder f θInput layer data The representation h mapped to a hidden layer, i.e. f θ It is a nonlinear transformation function of the following form:

[0043]

[0044] Where θ represents the encoder parameters in the denoising autoencoder, including W and b; W is the weight matrix between the hidden layer and the input layer, and b is the bias vector of the hidden layer. The decoder g' in the denoising autoencoder... θ The hidden layer representation h is demapped into a form that corresponds to the input vector. The reconstructed vector y is output in the output layer, i.e., y = g' θ (h). g' θ It is also a non-linear transformation function, with the following form:

[0045]

[0046] Where θ' represents the parameters of the decoder, including W' and b'; W' is the weight matrix between the output layer and the hidden layer, and b' is the bias vector of the output layer; s is the activation function.

[0047] To reconstruct clean original input data, the reconstruction error must be a measure of the error between the reconstructed vector y and the clean original vector x. Training a DAE is essentially minimizing this re-error, achieved by solving the following optimization problem:

[0048] (θ,θ′)=argminL(X,Y);

[0049] Where X is the set of input vectors x; Y is the set of corresponding reconstructed vectors y; and L is the loss function, usually defined as follows:

[0050]

[0051] Where L(X,Y) is the loss function value; M is the number of reconstructed vectors; N is the dimension of the reconstructed vectors; x ij Let y be the i-th input vector of dimension j; ij Let be the i-th reconstructed vector of dimension j.

[0052] After training, the hidden layer representation of the DAE is considered a useful representation of the input because it is possible to recover the corresponding clean input data from a missing data point. Furthermore, for a new missing vector fed into the input layer, the trained DAE can recover its corresponding clean vector.

[0053] DSAE consists of a bottom-level DAE and stacked AEs in the middle layers, and its structure is as follows: Figure 4 As shown. The purpose of the bottom layer DAE is to recover a clean input vector from the missing vector, while the stacked AEs in the middle layers extract features from the input layer.

[0054] For the missing vector fed into the input layer, it is mapped to the reconstructed vector y using the following formula:

[0055]

[0056] Where W1 is the weight matrix between the first hidden layer and the input layer; b1 is the bias vector of the first hidden layer; h l W is the hidden layer output representation of the l-th hidden layer; l It is the weight matrix between the l-th hidden layer and the (l-1)-th hidden layer; b l W is the bias vector of the l-th hidden layer; l+1 It is the weight matrix between the output layer and the (l+1)th hidden layer; b l+1 It is the bias vector of the output layer.

[0057] The process of training DSAE is also a process of minimizing reconstruction error, which is achieved by solving the following optimization problem:

[0058]

[0059] Where θ represents the inclusion of W l and b l The parameters are (1 = 1, 2, ..., L+1); X is the complete traffic dataset, Y is the set of data vectors reconstructed by the model, and L is the loss function, defined as:

[0060]

[0061] Where L(X,Y) is the loss function value; M is the number of reconstructed vectors; N is the dimension of the reconstructed vectors; x ij Let y be the i-th input vector of dimension j; ij Let be the i-th reconstructed vector of dimension j.

[0062] Unlike training an AE or DAE, this process involves two steps: pre-training and global fine-tuning. Pre-training trains the parameters of each AE (i.e., hidden layer), while global fine-tuning adjusts all the weights and biases of the DSAE. After training, the DSAE is able to recover its corresponding clean input vector from the missing vector.

[0063] Furthermore, a denoising stacked autoencoder is trained using the complete traffic dataset and the missing vector set to obtain the trained denoising stacked autoencoder, specifically including:

[0064] Determine if the current hidden layer is the first hidden layer.

[0065] When the current hidden layer is the first hidden layer, the missing vector set is input from the input layer, passes through the first hidden layer, and the reconstructed vector of the first hidden layer is obtained. Based on the complete traffic dataset, the reconstructed vector of the first hidden layer and the loss function, the first hidden layer is updated to obtain the updated first hidden layer.

[0066] When the current hidden layer is not the first hidden layer, the reconstruction vector of the previous hidden layer is input into the current hidden layer to obtain the reconstruction vector of the current hidden layer. Based on the complete traffic dataset, the reconstruction vector of the current hidden layer and the loss function, the parameters of the current hidden layer are updated using the backpropagation algorithm to obtain the updated current hidden layer.

[0067] Each hidden layer is used as the current hidden layer in turn until each hidden layer is updated to obtain the first denoising stacked autoencoder; the first denoising stacked autoencoder is obtained by training each hidden layer of the denoising stacked autoencoder one by one.

[0068] The missing vector set is input into the first denoising stack autoencoder to obtain the output data of the first denoising stack autoencoder.

[0069] Based on the complete traffic dataset, the output data of the first denoising stacked autoencoder, and the loss function, the backpropagation algorithm is used to update all parameters of the first denoising stacked autoencoder. That is, the trained denoising stacked autoencoder is obtained by globally fine-tuning the first denoising stacked autoencoder.

[0070] Furthermore, the generative adversarial network (GAN) model in this embodiment is the core of the algorithm architecture. Both the generator and discriminator are deep neural network structures composed of various activation functions, including PRELU, Dropout, Log Sigmoid, and Tanh. Specifically, the hidden layer of the generator uses PRELU, the output layer uses Dropout, the hidden layer of the decoder uses Tanh, and the output layer uses Log Sigmoid. During network training, the generator G fits the distribution of the entire data and generates samples z, while the discriminator D is responsible for determining whether sample z comes from G or is sampled from real data x. The goal of the GAN is to continuously train G to make the generated data increasingly closer to real samples, while continuously optimizing D to distinguish the source of the data as much as possible. The optimization functions for G and D can be expressed as follows:

[0071] minmaxV(D,G)=Ex~pdata(x)[logD(x)]+Ez~p(z)[log(1-D(G(z)))];

[0072] Where V(D,G) is the optimization function; logD(x) is the logarithm of the discriminator's output probability for the real data x; and log(1-D(G(z))) is the logarithm of the discriminator's output probability for the generator's generated data G(z).

[0073] In the function V(D,G), the first term is the entropy of the data from the true distribution pdata(x) passing through the discriminator, which attempts to maximize it to 1. The second term is the entropy of the data from the random input p(z) passing through the generator, which generates a fake sample that is identified as fake by the discriminator; in this term, the discriminator attempts to minimize it to 0. Therefore, overall, the discriminator tries to maximize the function V(D,G). On the other hand, the generator's task is exactly the opposite; it tries to minimize the function V(D,G), minimizing the difference between real and fake data. When the discriminator D's classification accuracy converges to 50%, it indicates that the generator G has rendered the discriminator D unable to distinguish the data source, and the model training is considered complete. At this point, the high-dimensional feature vector of the denoising autoencoder (after upscaling the encoder) can be used as input to generate a new dataset using the generator G.

[0074] Furthermore, a generative adversarial network is trained using the complete traffic dataset and the first high-dimensional feature vector data to obtain a trained generative adversarial network, specifically including:

[0075] The first high-dimensional feature vector data is input into the generator to obtain the generated data output by the generator.

[0076] The complete traffic dataset and the generated data output by the generator are input into the discriminator to obtain the probability distribution;

[0077] An optimization function is constructed, and the generator and discriminator are updated according to the probability distribution and the optimization function to obtain a trained generative adversarial network.

[0078] Based on the aforementioned algorithm principles and preliminary hyperparameter experiments, this application combines a denoising stacked autoencoder with a generative adversarial network, using traffic data to compensate for the model structure as follows: Figure 5As shown. The learning rate of the denoising stacked autoencoder is 0.005. The encoder consists of an input layer, seven hidden layers, and an output layer connected in sequence, and the decoder consists of an input layer, seven hidden layers, and an output layer connected in sequence. Considering that the experimental data selected in this embodiment has a maximum dimension of 18-dimensional dataset, and weighing the computational cost of the network and the representational ability of the feature vectors for the data, the encoder is set to upscale the original traffic data to 200-300 dimensions through a deep neural network structure. At this point, its information content meets the sample dimension required for training the generative adversarial network. In subsequent dataset experiments, the original traffic data and 216-dimensional feature vectors are mutually converted. During the training process, the encoder performs 2% random erasure on the input signal and adds Gaussian white noise with a mean of 0.

[0079] After the denoising stacked autoencoder is trained, its encoder part is used to increase the dimensionality of the missing vectors. The transformed first high-dimensional feature vector is then fed into the generative adversarial network for training. The generator G has a learning rate of 0.001 and consists of one input layer, fourteen hidden layers, and one output layer. The discriminator D has a learning rate of 0.005 and consists of one input layer, eighteen hidden layers, and one output layer. In the architecture design, this application makes the discriminator slightly more powerful than the generator, making the discriminator more rigorous during training and generating data that is closer to real data. Training is complete when the discriminator's classification accuracy approaches 50%. On a single Tesla GPU, the training time for the model ranges from approximately 60 to 500 minutes, depending on the dataset size.

[0080] Compared with existing technologies, the traffic data compensation model generated in this application has the following advantages:

[0081] (1) This application can effectively handle missing data rates of up to 80%, while traditional methods often fail to provide accurate data compensation under high missing data rates, resulting in a significant decrease in the performance of the prediction model. By combining the use of Denoising Stacked Autoencoder (DSAE) and Generative Adversarial Network (GAN), DSAE enhances the robustness of the model to missing data by introducing random noise during training, and achieves data upsizing through multi-layer autoencoder stacking to extract richer feature representations; GAN uses these high-dimensional feature vectors to generate samples similar to the original data, thereby generating high-quality compensation data under high missing data rates.

[0082] (2) Since the discriminator of the generative adversarial network is responsible for distinguishing between generated data and real data, the quality of generated data is improved through adversarial training, making the generated data more similar to real data in terms of statistical characteristics.

[0083] (3) Compared with traditional data imputation methods, the method in this application can better preserve the overall distribution of the data and enhance the generalization ability of the subsequent prediction model. By using DSAE in the pre-training stage to simulate the actual situation of missing data through random masking and noise injection, the model has better generalization ability when facing real missing data.

[0084] (4) The method of this application utilizes an optimized algorithm architecture design, particularly the combined use of a denoising stacked autoencoder and a generative adversarial network. The pre-training phase of DSAE can process multiple data batches in parallel, while the training process of GAN reduces the model training time through optimized network structure and parameter settings.

[0085] (5) By combining DSAE and GAN, the generated data is not only similar to the original data in terms of statistical characteristics, but also better reflects the dynamic changes of traffic flow, providing higher quality input data for traffic flow prediction models, thereby improving the accuracy and reliability of prediction.

[0086] This application also provides an application scenario in which the above-described traffic data compensation model determination method is applied. Specifically, the traffic data compensation model determination method provided in this embodiment can be applied in a traffic data compensation scenario. The traffic data compensation scenario includes a raw traffic data acquisition stage and a raw traffic data compensation stage; the raw traffic data is input into the traffic data compensation model to obtain the compensated traffic data. The traffic data compensation model determination method provided in this embodiment belongs to the training stage of the traffic data compensation model within the raw traffic data compensation stage.

[0087] In one exemplary embodiment, a compensation method for a traffic data compensation model is provided, comprising:

[0088] Obtain raw traffic data.

[0089] The original traffic data is input into the traffic data compensation model to obtain the compensated traffic data.

[0090] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 6As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores processed data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a traffic data compensation model determination method.

[0091] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0092] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0093] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0094] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0095] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0096] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0097] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0098] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for determining a traffic data compensation model, characterized in that, The method for determining the traffic data compensation model includes: Random missing data processing and missing data mapping are performed on the complete traffic dataset to obtain a missing vector set; the complete traffic dataset includes traffic flow data, spatiotemporal data, and environmental data; A denoising stacked autoencoder is trained using the complete traffic dataset and the missing vector set to obtain a trained denoising stacked autoencoder. The missing vector set is input into the encoder of the trained denoising stacked autoencoder for dimensionality upscaling to obtain the first high-dimensional feature vector data. The complete traffic dataset and the first high-dimensional feature vector data are used to train a generative adversarial network to obtain a trained generative adversarial network. The encoder of the trained denoising stacked autoencoder, the trained generative adversarial network, and the decoder of the trained denoising stacked autoencoder are sequentially connected to obtain the traffic data compensation model. The denoising stacked autoencoder comprises an input layer, multiple hidden layers, and an output layer connected in sequence. The complete traffic dataset and the missing vector set are input into the denoising stacked autoencoder, and the denoising stacked autoencoder is trained to obtain a trained denoising stacked autoencoder, specifically including: Obtain the first denoising stacked autoencoder; the first denoising stacked autoencoder is obtained by training each hidden layer of the denoising stacked autoencoder one by one; The missing vector set is input into the first denoising stack autoencoder to obtain the output data of the first denoising stack autoencoder. Based on the complete traffic dataset, the output data of the first denoising stacked autoencoder, and the loss function, the first denoising stacked autoencoder is updated to obtain a trained denoising stacked autoencoder. Obtaining the first denoising stack-based autoencoder specifically includes: Determine if the current hidden layer is the first hidden layer; When the current hidden layer is the first hidden layer, the missing vector set is input from the input layer, passes through the first hidden layer, and the reconstructed vector of the first hidden layer is obtained. Based on the complete traffic dataset, the reconstructed vector of the first hidden layer, and the loss function, the first hidden layer is updated to obtain the updated first hidden layer. When the current hidden layer is not the first hidden layer, the reconstruction vector of the previous hidden layer is input into the current hidden layer to obtain the reconstruction vector of the current hidden layer. Based on the complete traffic dataset, the reconstruction vector of the current hidden layer, and the loss function, the current hidden layer is updated to obtain the updated current hidden layer. Each hidden layer is used as the current hidden layer in turn until each hidden layer is updated, thus obtaining the first denoising stack-based autoencoder.

2. The method for determining the traffic data compensation model according to claim 1, characterized in that, The generative adversarial network (GAN) includes a generator and a discriminator. The complete traffic dataset and the first high-dimensional feature vector data are input into the GAN to train it, resulting in a trained GAN. Specifically, this includes: The first high-dimensional feature vector data is input into the generator to obtain the generated data output by the generator; The complete traffic dataset and the generated data are input into the discriminator to obtain the probability distribution; An optimization function is constructed, and the generator and the discriminator are updated according to the probability distribution and the optimization function to obtain a trained generative adversarial network.

3. The method for determining the traffic data compensation model according to claim 1, characterized in that, The trained denoising stacked autoencoder includes an encoder comprising an input layer, seven hidden layers, and an output layer connected in sequence; the trained denoising stacked autoencoder also includes a decoder comprising an input layer, seven hidden layers, and an output layer connected in sequence; the trained generative adversarial network includes a trained generator and a discriminator, wherein the trained generator comprises an input layer, fourteen hidden layers, and an output layer connected in sequence; and the trained decoder comprises an input layer, eighteen hidden layers, and an output layer connected in sequence.

4. The method for determining the traffic data compensation model according to claim 3, characterized in that, The hidden layer of the trained generator uses the PRELU activation function; the output layer of the trained generator uses the Dropout activation function; the hidden layer of the trained decoder uses the Tanh activation function; and the output layer of the trained decoder uses the Log Sigmoid activation function.

5. A method for supplementing traffic data, characterized in that, The traffic data compensation methods include: Obtain raw traffic data; The original traffic data is input into the traffic data compensation model to obtain the compensated traffic data; the traffic data compensation model is the model obtained by the traffic data compensation model determination method according to any one of claims 1-4.

6. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the traffic data compensation model determination method according to any one of claims 1-4 or the traffic data compensation method according to claim 5.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the traffic data compensation model determination method according to any one of claims 1-4 or the traffic data compensation method according to claim 5.

8. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the traffic data compensation model determination method according to any one of claims 1-4 or the traffic data compensation method according to claim 5.

Images