Short-term traffic flow prediction method, system and equipment based on comprehensive cloud model

Abstract

The invention relates to the technical field of traffic flow prediction, in particular to a short-time traffic flow prediction method and system based on a comprehensive cloud model, and the method comprises the steps: obtaining traffic flow time sequence data of a target road section, extracting the spatial-temporal characteristics of traffic flow based on the traffic flow time sequence data, and obtaining a prediction result; respectively inputting the spatial-temporal characteristics into a first prediction model and a second prediction model to respectively obtain a first preliminary prediction sequence and a second preliminary prediction sequence, and generating a first prediction cloud droplet and a second prediction cloud droplet corresponding to a certain moment in the future; fusing the first predicted cloud droplet with a historically generated cloud droplet to obtain a first fused cloud droplet; fusing the second prediction cloud droplet with the first fusion cloud droplet to obtain a second fusion cloud droplet; obtaining a fusion cloud model based on the first prediction model, the second prediction model and a historical cloud model; and obtaining a final prediction value corresponding to the same future moment based on the digital characteristics of the fusion cloud model. The prediction value is combined with historical data, and the accuracy is higher.

Description

Technical Field

[0001] This invention relates to the field of traffic flow prediction technology, specifically to a method, system, and device for short-term traffic flow prediction based on a comprehensive cloud model. Background Technology

[0002] Blasting operations require precise selection of off-peak traffic periods to minimize safety risks. However, existing prediction methods often lead to biased timing of blasting operations due to insufficient use of historical data and lack of fusion of multi-source heterogeneous data, making it impossible to effectively define dynamic safety warning periods.

[0003] Traditional models such as ARIMA or Kalman filtering rely solely on short-term time-series data, neglecting the correlation between long-term historical traffic flow characteristics and blasting events, making it difficult to characterize the nonlinear traffic fluctuations induced by construction. Furthermore, single-method models, such as single neural networks or statistical models, struggle to account for the randomness and contingency of traffic flow, resulting in large prediction errors, low accuracy, and insufficient reliability in dynamic construction scenarios. Summary of the Invention

[0004] This invention provides a method, system, and device for short-term traffic flow prediction based on a comprehensive cloud model.

[0005] The technical solution of this invention is as follows: The short-term traffic flow prediction method based on the integrated cloud model includes the following steps: S1. Obtain the time-series traffic flow data of the target road segment; S2. Extract the spatiotemporal features of traffic flow based on traffic flow time series data; S3, cloud prediction and fusion; S3.1 Input the spatiotemporal features into the first prediction model and the second prediction model respectively to obtain the first preliminary prediction sequence and the second preliminary prediction sequence respectively; S3.2 For each preliminary prediction sequence, calculate its cloud model digital features and generate a first and second prediction cloud droplet corresponding to a future time. The prediction cloud droplet includes the prediction value and its degree of certainty. S3.3 Obtain historical traffic flow dataset, construct digital features of historical cloud model based on historical traffic flow dataset, and generate historical generated cloud droplets and their certainty corresponding to the same future time. S3.4 Obtain the fused cloud model digital features based on the digital features of the first prediction model, the digital features of the second prediction model, and the digital features of the historical cloud model; The first conflict coefficient is calculated based on the first predicted cloud droplet certainty and the historical generated cloud droplet certainty. The first fusion certainty is calculated based on the first conflict coefficient, the first predicted cloud droplet certainty, and the historical generated cloud droplet certainty. The second conflict coefficient is calculated based on the first fusion certainty and the second predicted cloud droplet certainty. The second fusion certainty is calculated based on the second conflict coefficient, the first fusion certainty, and the second predicted cloud droplet certainty. The final predicted value corresponding to the same future moment is obtained based on the digital features and second fusion certainty of the fused cloud model.

[0006] The final predicted value is calculated as follows: ; in, This is the final predicted value. , These are the expectation and random entropy of the fused cloud model, respectively. This represents the second degree of fusion certainty.

[0007] Follows a normal distribution ,in, , These are the entropy and hyper-entropy of the fused cloud model, respectively.

[0008] The fused cloud model digital features are obtained from the digital features of the first prediction model, the digital features of the second prediction model, and the digital features of the historical cloud model. The specific calculation method is as follows: The digital characteristics of the converged cloud model include its expectation, entropy, and hyperentropy, which are calculated as follows: , , , in, , and These are the expectation, entropy, and hyperentropy of the first prediction model, respectively. , and These are the expectation, entropy, and hyperentropy of the first prediction model, respectively. , and These are the expectation, entropy, and hyperentropy of the historical cloud prediction model, respectively.

[0009] Based on historical traffic flow datasets, digital features of the historical cloud model are constructed. These digital features include the expectation, entropy, and hyperentropy of the historical cloud model, specifically: ; ; ; in, , , Let be the expectation, entropy, and hyperentropy of the historical cloud model, respectively, and m be the size of the historical dataset. Data collected from historical traffic flow datasets. This represents the sample variance.

[0010] The first prediction model is a temporal convolutional network model.

[0011] The second prediction model is the Long Short-Term Memory (LSTM) network model.

[0012] Predict the final values ​​for multiple future moments, and select the future moment with the smallest final predicted value as the construction period.

[0013] A short-term traffic flow prediction system based on a comprehensive cloud model is used to implement the aforementioned short-term traffic flow prediction method based on a comprehensive cloud model, including: The data acquisition module acquires the time-series traffic flow data for the target road segment; The feature extraction module extracts the spatiotemporal features of traffic flow based on traffic flow time-series data; The prediction module inputs the spatiotemporal features into the first prediction model and the second prediction model respectively to obtain the first preliminary prediction sequence and the second preliminary prediction sequence respectively; for each preliminary prediction sequence, it calculates its cloud model digital features and generates the first and second predicted cloud drops corresponding to a certain future moment, wherein the predicted cloud drops include the predicted value and its certainty; it acquires historical traffic flow dataset, constructs historical cloud model digital features based on the historical traffic flow dataset, and generates historical generated cloud drops and their certainty corresponding to the same future moment; The fusion module obtains the fused cloud model digital features based on the digital features of the first prediction model, the digital features of the second prediction model, and the digital features of the historical cloud model; calculates the first conflict coefficient based on the certainty of the first predicted cloud droplets and the certainty of the historical generated cloud droplets; calculates the first fusion certainty based on the first conflict coefficient, the certainty of the first predicted cloud droplets, and the certainty of the historical generated cloud droplets; calculates the second conflict coefficient based on the first fusion certainty and the certainty of the second predicted cloud droplets; and obtains the final predicted value corresponding to the same future moment based on the digital features of the fused cloud model and the second fusion certainty.

[0014] An electronic device includes: one or more processors; and a memory associated with the one or more processors, the memory storing program instructions that, when read and executed by the one or more processors, perform the aforementioned short-term traffic flow prediction method based on a comprehensive cloud model.

[0015] The beneficial effects of this application on accurately predicting short-term traffic flow based on a comprehensive cloud model are as follows: 1. Compared with traditional single-model or simple model averaging methods, this invention avoids the limitations of a single model, mines data features from different angles, and can provide prediction results with higher accuracy and smaller error.

[0016] 2. The output of this invention is not only a predicted value, but also a prediction result with deterministic metrics. By converting point predictions into cloud droplets, the reliability of the prediction is quantified.

[0017] 3. The introduction of historical cloud models provides a measurable standard for current predictions, objectively measuring whether current predictions deviate from the norm, providing a reliable basis for decision-making, and greatly improving the robustness of predictions. Detailed Implementation

[0018] Example: The technical solution of this invention is as follows: The short-term traffic flow prediction method based on the integrated cloud model includes the following steps: S1. Obtain the traffic flow time sequence data of the target road segment.

[0019] Traffic flow time-series data of the target construction section is obtained, and anomaly processing based on generative adversarial networks is used to process the traffic flow time-series data to identify and label abnormal data, fill in missing values, and obtain complete traffic flow time-series data after removing abnormal data.

[0020] S2. Extract the spatiotemporal features of traffic flow based on traffic flow time series data.

[0021] Variational mode decomposition is used to decompose traffic flow time series data into multiple intrinsic mode components. Graph convolutional networks are then used to model each intrinsic mode component and the road network topology to extract the spatiotemporal features of traffic flow.

[0022] S3, cloud prediction and fusion.

[0023] S3.1 Input the spatiotemporal features into the first prediction model and the second prediction model respectively to obtain the first preliminary prediction sequence and the second preliminary prediction sequence respectively; S3.2 For each preliminary prediction sequence, calculate its cloud model digital features and generate a first and second prediction cloud droplet corresponding to a future time. The prediction cloud droplet includes the prediction value and its degree of certainty. S3.3 Obtain historical traffic flow dataset, construct historical cloud model based on historical traffic flow dataset, and generate historical generated cloud droplets and their determinism corresponding to the same future time. S3.1 Input the spatiotemporal features into the first prediction model and the second prediction model respectively to obtain the first preliminary prediction sequence and the second preliminary prediction sequence respectively; S3.2 For each preliminary prediction sequence, calculate its cloud model digital features and generate a first and second prediction cloud droplet corresponding to a future time. The prediction cloud droplet includes the prediction value and its degree of certainty. S3.3 Obtain historical traffic flow dataset, construct digital features of historical cloud model based on historical traffic flow dataset, and generate historical generated cloud droplets and their certainty corresponding to the same future time. S3.4 Obtain the fused cloud model digital features based on the digital features of the first prediction model, the digital features of the second prediction model, and the digital features of the historical cloud model; The first conflict coefficient is calculated based on the first predicted cloud droplet certainty and the historical generated cloud droplet certainty. The first fusion certainty is calculated based on the first conflict coefficient, the first predicted cloud droplet certainty, and the historical generated cloud droplet certainty. The second conflict coefficient is calculated based on the first fusion certainty and the second predicted cloud droplet certainty. The second fusion certainty is calculated based on the second conflict coefficient, the first fusion certainty, and the second predicted cloud droplet certainty. The final predicted value corresponding to the same future moment is obtained based on the digital features and second fusion certainty of the fused cloud model.

[0024] The following describes in detail how to perform the above steps: S3.1 Input the spatiotemporal features into the first prediction model and the second prediction model respectively to obtain the first preliminary prediction sequence and the second preliminary prediction sequence respectively.

[0025] Specifically, the first prediction model is a temporal convolutional network model, and the second prediction model is a long short-term memory network model.

[0026] Among them, the temporal convolutional network model has the core advantage of capturing long-term dependencies. Through dilated convolution, it is easy to see very long-term historical information; and it can perform parallel computation, the convolution operation can be highly parallelized, the training speed is fast, the memory utilization is high, and the receptive field is flexible. This can also be achieved by stacking layers and adjusting the dilation coefficient, which can easily control the size of the receptive field.

[0027] Long Short-Term Memory (LSTM) network models have the core advantage of capturing short-term patterns and sequence dynamics. Their gating mechanism can finely regulate the retention and forgetting of information, but it must be calculated sequentially by time step, and the training speed is relatively slow. The size of the receptive field is more determined by the network structure and parameter dynamics, which makes LTM network models more sensitive to recent input changes and may capture some sudden transient patterns.

[0028] The goal of this process is to enable different types of models to extract different patterns from the same set of well-extracted spatiotemporal features, thereby generating preliminary prediction sequences with high diversity and strong complementarity, providing high-quality and diversified support for subsequent fusion.

[0029] S3.2 For each preliminary prediction sequence, calculate its cloud model digital features and generate a first prediction cloud droplet and a second prediction cloud droplet corresponding to a certain future time. The prediction cloud droplet contains the prediction value and its degree of certainty.

[0030] Inverse cloud computing is performed on the two preliminary prediction sequences. The input is a complete data sequence, and the output is a cloud model digital feature (E) describing the overall characteristics of the sequence. x E n H e ).

[0031] E x (Expectation): The average value of the sequence is the center value of the cloud droplet in the universe of discourse.

[0032] E n (Entropy): Represents the ambiguity and uncertainty of a sequence. The higher the entropy, the more dispersed the data.

[0033] H e (Hyperentropy): It is a measure of the uncertainty of entropy, representing the entropy of entropy, and reflecting the thickness and dispersion of clouds.

[0034] Then, using the obtained digital features, a positive cloud generator is used to generate a predicted cloud droplet corresponding to a certain future moment and the certainty that the cloud droplet belongs to this cloud. The certainty is a value between 0 and 1, indicating how likely the predicted value is to belong to the concept described by the model.

[0035] Specifically, a cloud droplet is the most basic unit in the cloud model. It is a two-dimensional point, for example: (x,μ)=(3000, 85), where x is a specific value, predicting traffic flow as x vehicles / month; μ is the degree of certainty of the value, with an 85% confidence that the value 3000 is reliable.

[0036] Therefore, we can know that a cloud droplet is a data structure that binds a definite numerical value x together with its definite value μ.

[0037] S3.3 Obtain historical traffic flow dataset, construct digital features of historical cloud model based on historical traffic flow dataset, and generate historical generated cloud droplets and their certainty corresponding to the same future time. It should be noted that the traffic flow time series data of the target road segment and the historical traffic flow dataset are two different concepts in this application, and they play completely different roles.

[0038] Traffic flow time series data for the target road segment is used for current and future predictions. This includes recent and current data, which is usually continuous data from several months or a year before a certain point in the future. It is raw, unprocessed data that needs to be cleaned and feature extracted before it can be input into the prediction model.

[0039] Historical traffic flow datasets are used to learn historical patterns and serve as a benchmark or prior knowledge for evaluating the reliability of current predictions. They consist of long-term data, typically several years, used to uncover periodic and trend patterns. Moreover, historical traffic flow datasets are pre-processed and clean datasets.

[0040] Specifically, the historical traffic flow dataset is Where m is the amount of data. Based on historical traffic flow datasets, a historical cloud model is established, and the digital features of the historical cloud model are calculated, including the expectation, entropy, and hyperentropy of the historical cloud model. It is the average value of historical data, representing the average level or central trend of traffic flow over a historical period. It is the most core feature of the historical cloud model; the entropy of the historical cloud model This reflects the range or dispersion of historical data around the central value. The larger the value, the more drastic the historical traffic flow fluctuations and the greater the uncertainty; the hyperentropy of the historical cloud model. The stability and reliability of the entropy of the historical cloud model were measured. The larger the value, the more unstable the dispersion of the historical data itself is. Thus, the historical data is no longer a jumble of scattered points, but is summarized into three numerical characteristics. These characteristics together constitute a mathematical model that can describe the "average level," "fluctuation range," and "fluctuation stability" of historical traffic flow. The specific calculation method is as follows: ; ; ; in, , , Let be the expectation, entropy, and hyperentropy of the historical cloud model, respectively, and m be the size of the historical dataset. Data collected from historical traffic flow datasets. This represents the sample variance.

[0041] Therefore, this invention does not simply rely on the result of a single predictive model, but rather treats the predictions of two different models as two pieces of evidence with certainty, and then introduces a third piece of evidence from historical data as a reference benchmark. By analyzing the mutual support and conflict relationships among these three pieces of evidence, a more reliable and robust fused prediction result is ultimately obtained.

[0042] S3.4 Obtain the fused cloud model digital features based on the digital features of the first prediction model, the digital features of the second prediction model, and the digital features of the historical cloud model; The digital characteristics of the converged cloud model include expectation, entropy, and hyperentropy, among which the expectation of the converged cloud model... This represents the predicted central value after the fusion of all information sources. Its calculation formula is typically a weighted average of the expectations of each sub-model, with the weights being the reciprocals of the entropy of each model; the entropy of the fused cloud model... This represents the overall uncertainty after fusion. The uncertainty of the fused cloud model should be greater than or equal to that of any single predictive cloud model. Therefore, the entropy of each sub-model is directly added together to characterize the overall dispersion of the fused model. This is a conservative and reasonable estimate, ensuring that the uncertainty range of the final prediction result can cover the possibilities of all sub-models. Its calculation formula is usually the sum of the entropies of each sub-model; the super-entropy of the fused cloud model. Measured The specific calculation method for its stability and reliability is as follows: , , , in, , and These are the expectation, entropy, and hyperentropy of the first prediction model, respectively. , and These are the expectation, entropy, and hyperentropy of the first prediction model, respectively. , and These are the expectation, entropy, and hyperentropy of the historical cloud prediction model, respectively.

[0043] The first conflict coefficient is calculated based on the first predicted cloud droplet certainty and the historical generated cloud droplet certainty. The specific calculation method is as follows: , in, The first conflict coefficient, For the first predicted cloud droplet certainty, Determine the historical generation of cloud droplet certainty.

[0044] Based on the first conflict coefficient, the first predicted cloud droplet certainty, and the historical generated cloud droplet certainty, the first fusion certainty is calculated. The specific calculation method is as follows: , in, This represents the first degree of fusion certainty.

[0045] Based on the first fusion certainty and the second predicted cloud droplet certainty, the specific calculation method for the second conflict coefficient is as follows: , in, For the second predicted cloud droplet certainty, This is the second conflict coefficient.

[0046] Based on the second conflict coefficient, the first fusion certainty, and the second predicted cloud droplet certainty, the second fusion certainty is calculated. The specific calculation method is as follows: in, This represents the second degree of fusion certainty.

[0047] The final predicted value corresponding to the same future moment is obtained based on the digital features and second fusion certainty of the fused cloud model.

[0048] The final predicted value is calculated by the inverse cloud generator, which is the reverse process of the forward cloud generator. It deduces the corresponding specific predicted value from the known digital characteristics of the cloud model and the specific degree of certainty.

[0049] The final predicted value is determined by the expectation of the fused cloud model and a random fluctuation term. The magnitude of the fluctuation term is jointly determined by the entropy of the fused cloud model and the second fusion certainty. The higher the second fusion certainty, the smaller the fluctuation range, and the closer the predicted value is to the center of the expectation of the fused cloud model.

[0050] The final predicted value is calculated as follows: ; in, This is the final predicted value, representing the specific traffic flow at a future point in time. , These are the expectation and random entropy of the fused cloud model, respectively. This represents the second degree of fusion certainty.

[0051] Random Entropy It is a random number that follows a normal distribution. , Introduction This reflects the randomness of cloud models, causing slight fluctuations in each generated prediction. In practical engineering applications, to simplify calculations, sometimes the values ​​are directly taken... = This yields a definite output value, which is then multiplied by... This is the cloud droplet transformation term, whose function is to map the deterministic metric within the interval (0,1) back to the range of traffic flow. The larger the value, the smaller the value of this item, meaning the predicted value is more likely to fall within the range of [previous value]. nearby.

[0052] Predict the final values ​​for multiple future moments, and select the future moment with the smallest final predicted value as the construction period.

[0053] Specific applications include: The monthly traffic flow of a certain highway in the coming year is predicted, and the results are shown in the table below: The table shows that the peak traffic volume on the highway in a certain area is expected to be in January, with a traffic volume of 54,614 pcu / month, while the lowest traffic volume is expected in May, at 24,687 pcu / month. Therefore, it is recommended that the bridge demolition be scheduled for May or November.

[0054] Based on application examples, it can be seen that, according to the cloud fusion model of this application, the predicted value of the future short-term traffic flow and the deterministic measure of the predicted value are output; the predicted value and the deterministic measure can be sent to the traffic dynamic management and control system to generate traffic safety management and control strategies for the construction area.

[0055] Thus, this application overcomes the problems of low accuracy and poor reliability of traditional prediction models in the specific scenario of highway bridge blasting construction, due to the integrated, random, volatile and accidental nature of traffic flow. It achieves accurate prediction of short-term traffic flow, providing reliable data support for delineating dynamic safety warning zones and formulating traffic control plans, thereby fundamentally reducing construction safety risks.

[0056] A short-term traffic flow prediction system based on a comprehensive cloud model is used to implement the aforementioned short-term traffic flow prediction method based on a comprehensive cloud model, including: The data acquisition module acquires the time-series traffic flow data for the target road segment; The feature extraction module extracts the spatiotemporal features of traffic flow based on traffic flow time-series data; The prediction module inputs the spatiotemporal features into the first prediction model and the second prediction model respectively to obtain the first preliminary prediction sequence and the second preliminary prediction sequence respectively; for each preliminary prediction sequence, it calculates its cloud model digital features and generates the first and second predicted cloud drops corresponding to a certain future moment, wherein the predicted cloud drops include the predicted value and its certainty; it acquires historical traffic flow dataset, constructs historical cloud model digital features based on the historical traffic flow dataset, and generates historical generated cloud drops and their certainty corresponding to the same future moment; The fusion module obtains the fused cloud model digital features based on the digital features of the first prediction model, the digital features of the second prediction model, and the digital features of the historical cloud model; calculates the first conflict coefficient based on the certainty of the first predicted cloud droplets and the certainty of the historical generated cloud droplets; calculates the first fusion certainty based on the first conflict coefficient, the certainty of the first predicted cloud droplets, and the certainty of the historical generated cloud droplets; calculates the second conflict coefficient based on the first fusion certainty and the certainty of the second predicted cloud droplets; and obtains the final predicted value corresponding to the same future moment based on the digital features of the fused cloud model and the second fusion certainty.

[0057] An electronic device includes: one or more processors; and a memory associated with the one or more processors, the memory storing program instructions that, when read and executed by the one or more processors, perform the aforementioned short-term traffic flow prediction method based on a comprehensive cloud model.

Claims

1. A short-term traffic flow prediction method based on a comprehensive cloud model, characterized in that, Includes the following steps: S1. Obtain the time-series traffic flow data of the target road segment; S2. Extract the spatiotemporal features of traffic flow based on traffic flow time series data; S3, cloud prediction and fusion; S3.1 Input the spatiotemporal features into the first prediction model and the second prediction model respectively to obtain the first preliminary prediction sequence and the second preliminary prediction sequence respectively; S3.2 For each preliminary prediction sequence, calculate its cloud model digital features and generate a first and second prediction cloud droplet corresponding to a future time. The prediction cloud droplet includes the prediction value and its degree of certainty. S3.3 Obtain historical traffic flow dataset, construct digital features of historical cloud model based on historical traffic flow dataset, and generate historical generated cloud droplets and their certainty corresponding to the same future time. S3.4 Obtain the fused cloud model digital features based on the digital features of the first prediction model, the digital features of the second prediction model, and the digital features of the historical cloud model; The first conflict coefficient is calculated based on the first predicted cloud droplet certainty and the historical generated cloud droplet certainty. The first fusion certainty is calculated based on the first conflict coefficient, the first predicted cloud droplet certainty, and the historical generated cloud droplet certainty. The second conflict coefficient is calculated based on the first fusion certainty and the second predicted cloud droplet certainty. The second fusion certainty is calculated based on the second conflict coefficient, the first fusion certainty, and the second predicted cloud droplet certainty. The final predicted value corresponding to the same future moment is obtained based on the digital features and second fusion certainty of the fused cloud model.

2. The short-term traffic flow prediction method based on a comprehensive cloud model according to claim 1, characterized in that, The final predicted value is calculated as follows: ； in, This is the final predicted value. , These are the expectation and random entropy of the fused cloud model, respectively. This represents the second degree of fusion certainty.

3. The short-term traffic flow prediction method based on a comprehensive cloud model according to claim 2, characterized in that, The Follows a normal distribution ,in, , These are the entropy and hyper-entropy of the fused cloud model, respectively.

4. The short-term traffic flow prediction method based on a comprehensive cloud model according to claim 1, characterized in that, The fused cloud model digital features are obtained based on the digital features of the first prediction model, the digital features of the second prediction model, and the digital features of the historical cloud model. The specific calculation method is as follows: The digital characteristics of the converged cloud model include its expectation, entropy, and hyperentropy, which are calculated as follows: ， ， ， in, , and These are the expectation, entropy, and hyperentropy of the first prediction model, respectively. , and These are the expectation, entropy, and hyperentropy of the first prediction model, respectively. , and These are the expectation, entropy, and hyperentropy of the historical cloud prediction model, respectively.

5. The short-term traffic flow prediction method based on a comprehensive cloud model according to claim 1, characterized in that, Based on historical traffic flow datasets, digital features of the historical cloud model are constructed. These digital features include the expectation, entropy, and hyperentropy of the historical cloud model, specifically: ； ； ； in, , , Let be the expectation, entropy, and hyperentropy of the historical cloud model, respectively, and m be the size of the historical dataset. Data collected from historical traffic flow datasets. This represents the sample variance.

6. The short-term traffic flow prediction method based on a comprehensive cloud model according to claim 1, characterized in that, The first prediction model is a temporal convolutional network model.

7. The short-term traffic flow prediction method based on a comprehensive cloud model according to claim 1, characterized in that, The second prediction model is a long short-term memory network model.

8. The short-term traffic flow prediction method based on a comprehensive cloud model according to claim 1, characterized in that, Predict the final values ​​for multiple future moments, and select the future moment with the smallest final predicted value as the construction period.

9. A short-term traffic flow prediction system based on a comprehensive cloud model, used to implement the short-term traffic flow prediction method based on a comprehensive cloud model as described in any one of claims 1-8, characterized in that, include: The data acquisition module acquires the time-series traffic flow data for the target road segment; The feature extraction module extracts the spatiotemporal features of traffic flow based on traffic flow time-series data; The prediction module inputs the spatiotemporal features into the first prediction model and the second prediction model respectively to obtain the first preliminary prediction sequence and the second preliminary prediction sequence respectively; for each preliminary prediction sequence, it calculates its cloud model digital features and generates the first and second predicted cloud drops corresponding to a certain future moment, wherein the predicted cloud drops include the predicted value and its certainty; it acquires historical traffic flow dataset, constructs historical cloud model digital features based on the historical traffic flow dataset, and generates historical generated cloud drops and their certainty corresponding to the same future moment; The fusion module obtains the fused cloud model digital features based on the digital features of the first prediction model, the digital features of the second prediction model, and the digital features of the historical cloud model. The first conflict coefficient is calculated based on the first predicted cloud droplet certainty and the historical generated cloud droplet certainty. The first fusion certainty is calculated based on the first conflict coefficient, the first predicted cloud droplet certainty, and the historical generated cloud droplet certainty. The second conflict coefficient is calculated based on the first fusion certainty and the second predicted cloud droplet certainty. The second fusion certainty is calculated based on the second conflict coefficient, the first fusion certainty, and the second predicted cloud droplet certainty. The final predicted value corresponding to the same future moment is obtained based on the digital features and second fusion certainty of the fused cloud model.

10. An electronic device, characterized in that, include: One or more processors; The system also includes a memory associated with the one or more processors, the memory being used to store program instructions that, when read and executed by the one or more processors, perform the short-term traffic flow prediction method based on the integrated cloud model as described in any one of claims 1-8.

Images