CN116030637B - Traffic state prediction integration method - Google Patents
Traffic state prediction integration method Download PDFInfo
- Publication number
- CN116030637B CN116030637B CN202310314503.5A CN202310314503A CN116030637B CN 116030637 B CN116030637 B CN 116030637B CN 202310314503 A CN202310314503 A CN 202310314503A CN 116030637 B CN116030637 B CN 116030637B
- Authority
- CN
- China
- Prior art keywords
- model
- sub
- data
- traffic
- iteration
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
The invention discloses a traffic state prediction integration method, and belongs to the technical field of public traffic passenger flow detection. The invention builds a robust integrated model which maximizes the diversity among the components, and inputs each training data subset into a corresponding sub-model for training; updating each sub-model through a designed iterative algorithm, generating an optimal prediction model after iteration is finished, inputting a data set to be predicted into the generated sub-model, thereby obtaining a prediction result of each sub-model, and giving a final prediction value on the basis. The method and the system can effectively improve the short-time traffic state prediction accuracy under complex scenes (such as weather conditions, traffic jams or traffic accidents and the like).
Description
Technical Field
The invention relates to the field of traffic state prediction, in particular to a traffic state prediction integration method with robustness and maximized diversity.
Background
Under the condition of rapid development of social economy, the rapid increase of the automobile conservation amount can cause potential problems of traffic jam, traffic accidents, serious environmental and noise pollution and the like, wherein the traffic jam causes economic loss, trip time consumption and aggravates environmental pollution; traffic accidents seriously affect the development of social economy and the improvement of people's life, which brings barriers to the sustainable development of society. The traffic state prediction aims at exploring the traffic running state at the future time, so that the urban road can run efficiently, and high-quality traffic service is provided for traffic participants. Traffic state based prediction methods are considered as effective approaches to relieving traffic congestion, preventing traffic accidents, reducing emissions and fuel consumption. Thus, researchers in this field have proposed various types of short-time traffic state prediction methods for recent decades, including: short-time traffic state prediction methods based on SVM and improved model thereof, short-time traffic state prediction methods based on integrated learning, short-time traffic state prediction methods based on deep learning, and the like.
The central idea of ensemble learning is to correct errors back even if one model gets mispredicted by combining multiple well performing algorithms to expect a better, more comprehensive prediction model. In view of the better predictive ability of ensemble learning, it is widely used in traffic flow prediction. A large number of experimental results show that the prediction method based on the ensemble learning has the advantages of high accuracy, strong generalization capability and the like.
However, due to various uncontrollable factors (such as rain and snow weather conditions, traffic jams, traffic accidents, transmission distortion, sensor communication faults, etc.), the collected data inevitably contains abnormal data points (abnormal values for short), errors or missing data, which significantly affects the prediction accuracy of the model. Although some preprocessing methods (e.g., discard, fill, replace, deduplication, etc.) may be used to remove outliers from the sample data, there is no guarantee that these outlier data points will be completely removed. In addition, these outliers may be important features of model predictions, such as traffic data under traffic congestion or traffic accidents, to better reflect the reality of road operation. Therefore, improving the robustness of the integrated model is an important and meaningful research effort.
Disclosure of Invention
Aiming at the defects and shortcomings existing in the prior art, the invention aims to provide the traffic state prediction integration method with robustness and maximized diversity, which can ensure that higher prediction precision can be obtained even under the influence of external factors, and provides a basis for releasing traffic information and avoiding potential traffic jams of urban roads.
The invention discloses a traffic state prediction integration method, which comprises the following steps:
acquiring traffic sampling data from different traffic data sampling points as data to be predicted, wherein the data to be predicted is divided into a plurality of corresponding data subsets according to the corresponding traffic data sampling points;
inputting data to be predicted into a trained robust integrated model which maximizes diversity among all components, wherein the integrated model is also divided into a plurality of corresponding sub-models according to corresponding traffic data sampling point positions, each data subset of the data to be predicted is respectively input into the corresponding sub-model to obtain a prediction result of each sub-model, and the prediction result of each sub-model is synthesized to give a final prediction value;
the objective function of each sub-model is expressed as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,p∈(0,2]，||·|| 2 represents L 2 The distance of the norms,mrepresenting the sample type sequence number of each sub-model data,Cfor each sub-model data sample type total,c m is the first sub-modelmE, obtaining optimal parameters corresponding to the class sample in advance through an optimization algorithm m1 And e m2 Respectively matrix A m And B m Unit column vector of corresponding dimension, matrix a m Representing the first sub-modelmData set consisting of class samples, matrix B m Representing the division of each submodel bymThe remainder outside the class samplem-data set consisting of class 1 samples, matrix a m And B m Having the same dimension numbern，q m Is the first sub-modelmThe relaxation variable corresponding to the class sample is obtained by the constraint condition of the objective function,λ 1 is a preset parameter for balancing the prediction error and diversity of the integrated model,F(w) represents a diversity function, w m Representing the first sub-modelmWeight vector of class sample, b m Representing the first sub-modelmDeviation of class samples, w= [w 1 ,w 2 ,…,w L ] T For the diversity of the weight vectors of the model,Lis the total number of sub-models, w m 、b m 、F(w) andpobtaining an optimal value through iterative optimization; in the objective function, the first term and the second term belong to the prediction error of the integrated model, the third term represents diversity, and when optimization is completed, the first term and the second term are optimal.
Further, the method comprisesc m Obtained by a drosophila optimization algorithm.
Further, obtaining w through iterative optimization m 、b m 、F(w) andpthe optimal value of (2) is obtained by iteratively updating the augmentation vector of each sub-model after the iteration is finished, and the specific steps are as follows:
1) Setting the iteration timestAnd iteration steppInitial values of 1 and 0.1, respectively, initializing an augmentation vector z m 1 Diagonal matrix D m1 1 And D m2 1 Wherein D is m1 1 Is the first of (2)iDiagonal elementsAnd D m2 1 Is the first of (2)jThe diagonal elements are respectively:
a i representative matrix A m Middle (f)iLine data, b j Representative matrix B m Middle (f)jThe data of the rows are stored in a memory,m 1 representation matrix A m Is a function of the total number of rows of the system,m 2 representation matrix B m E mi1 Represent e m1 Middle (f)iLine data, e mj2 Represent e m2 Middle (f)jLine data;
initializing a diversity weighting vector w;
2) Calculating an augmentation vector z m t(+1) And according to the augmentation vector z m t(+1) Updating diagonal matrix D m1 t(+1) And D m2 t(+1) The method specifically comprises the following steps:
wherein, the liquid crystal display device comprises a liquid crystal display device,εthe regularization coefficient is I, and the I is an identity matrix;
3) When an augmentation vector z is obtained m t(+1) When determining diversity function using alternate direction multiplier methodF(w) specifically:
wherein, the liquid crystal display device comprises a liquid crystal display device,L k andL l respectively represent the firstkSum of alllSub-model, p k Representation ofL k Is p l Representation ofL l Is used for predicting the vector of the vector; matrix D t Is the first of (2)kLine 1lIs arranged as
By passing throughF(w) Lagrangian function, calculating the next iteration value w of the obtained diversity weighting vector t(+1) ：
w t(+1) =(α t e-β t )/( λ 1 D t )
Wherein alpha is t And beta t Representing the number of iterations astTime of dayF(w) non-negative lagrangian multipliers, e is a unit column vector of the corresponding dimension;
4) According to the augmentation vector z obtained in step 2) m t(+1) Obtaining a weight vector w m And deviation b m Combining the diversity function obtained in step 3)F(w) and iteration stepp，Calculating a sub-model objective function value of the iteration according to the sub-model objective function;
if the number of iterationstEqual to 1, update the iteration numbert=t+1A kind of electronic device with high-pressure air-conditioning systemp=p+0.1Returning to the step 2) to continue iteration;
if the number of iterationstIf the difference is greater than 1, comparing the sub-model objective function value of the current iteration with the sub-model objective function value of the previous iteration, if the difference is less than the preset threshold, or the number of iterationstIf the weight vector w is larger than the preset upper limit value, stopping iteration, wherein the weight vector w is obtained in the iteration m Deviation b m Function of diversityF(w) iteration steppIs the corresponding optimal value, otherwise the iteration times are updatedt=t+1A kind of electronic device with high-pressure air-conditioning systemp=p+0.1Returning to the step 2) to continue iteration.
Further, the preset threshold value is 0.001.
Further, the preset upper limit value is 20.
Further, after each data subset of the data to be predicted is respectively input into a corresponding sub-model to obtain a prediction result of each sub-model, weighted voting is performed on the prediction result of each sub-model to give a final prediction value.
Further, training data adopted for training the robust integrated model and maximizing the diversity among the components is to obtain a part of traffic sampling data from different traffic data sampling points as a training data set, and divide the traffic sampling data into a plurality of corresponding training subsets according to the corresponding traffic data sampling points.
Further, when training data is adopted, a part of the acquired traffic sampling data is selected as a test data set for testing the prediction result of the integrated model which is robust in training and maximizes the diversity among components.
The beneficial effects of the invention are as follows: the invention builds a robust integrated model which maximizes the diversity among the components, and inputs each training data subset into a corresponding sub-model for training; updating each sub-model through a designed iterative algorithm, generating an optimal prediction model after iteration is finished, inputting a data set to be predicted into the generated sub-model, thereby obtaining a prediction result of each sub-model, and giving a final prediction value on the basis. Therefore, the method and the system can effectively alleviate the influence of abnormal values in short-time traffic state prediction by using the robust and maximized diversity integrated model, enhance the robustness of the model, improve the prediction precision of the model, and build the integrated model with higher flexibility, so that the method and the system are completely suitable for short-time traffic state prediction of normal scenes and complex scenes, and provide powerful guarantee for orderly and efficient traffic of urban road traffic.
Drawings
FIG. 1 is a flow chart of iterative training of a submodel of the present invention.
Description of the embodiments
The invention is further described below with reference to the drawings and examples.
Examples
The embodiment is a traffic state prediction integration method, which specifically comprises the following steps:
acquiring traffic sampling data from different traffic data sampling points as data to be predicted, wherein the data to be predicted is divided into a plurality of corresponding data subsets according to the corresponding traffic data sampling points;
inputting data to be predicted into a trained robust integrated model which maximizes diversity among all components, wherein the integrated model is also divided into a plurality of corresponding sub-models according to corresponding traffic data sampling point positions, each data subset of the data to be predicted is respectively input into the corresponding sub-model to obtain a prediction result of each sub-model, and the prediction result of each sub-model is synthesized (such as by a weighted voting mode) to give a final prediction value;
the objective function of each sub-model is expressed as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,p∈(0,2]，||·|| 2 represents L 2 The distance of the norms,mrepresenting the sample type sequence number of each sub-model data,Cfor each sub-model data sample type total,c m is the first sub-modelmE, obtaining optimal parameters corresponding to the class sample in advance through an optimization algorithm m1 And e m2 Respectively matrix A m And B m Unit column vector of corresponding dimension, matrix a m Representing the first sub-modelmData set consisting of class samples, matrix B m Representing the division of each submodel bymThe remainder outside the class samplem-data set consisting of class 1 samples, matrix a m And B m Having the same dimension numbern，q m Is the first sub-modelmThe relaxation variable corresponding to the class sample is obtained by the constraint condition of the objective function,λ 1 is a preset parameter for balancing the prediction error and diversity of the integrated model,F(w) represents a diversity function, w m Representing the first sub-modelmWeight vector of class sample, b m Representing the first sub-modelmDeviation of class samples, w= [w 1 ,w 2 ,…,w L ] T For the diversity of the weight vectors of the model,Lis the total number of sub-models, w m 、b m 、F(w) andpobtaining an optimal value through iterative optimization; in the objective function, the first term and the second term belong to the prediction error of the integrated model, the third term represents diversity, and when optimization is completed, the first term and the second term are optimal.
In the present embodiment of the present invention,c m obtained by a drosophila optimization algorithm.
Obtaining w by iterative optimization m 、b m 、F(w) andpi.e. iterative training process of each sub-model, as shown in fig. 1, the specific steps are:
1) Setting the iteration timestAnd iteration steppInitial values of 1 and 0.1, respectively, initializing an augmentation vector z m 1 Diagonal matrix D m1 1 And D m2 1 Wherein D is m1 1 Is the first of (2)iDiagonal elements and D m2 1 Is the first of (2)jThe diagonal elements are respectively:
a i representative matrix A m Middle (f)iLine data, b j Representative matrix B m Middle (f)jThe data of the rows are stored in a memory,m 1 representation matrix A m Is a function of the total number of rows of the system,m 2 representation matrix B m E mi1 Represent e m1 Middle (f)iLine data, e mj2 Represent e m2 Middle (f)jLine data;
initializing a diversity weighting vector w;
2) Calculating an augmentation vector z m t(+1) And according to the augmentation vector z m t(+1) Updating diagonal matrixD m1 t(+1) And D m2 t(+1) The method specifically comprises the following steps:
wherein, the liquid crystal display device comprises a liquid crystal display device,εthe regularization coefficient is I, and the I is an identity matrix;
3) When an augmentation vector z is obtained m t(+1) When determining diversity function using alternate direction multiplier methodF(w) specifically:
wherein, the liquid crystal display device comprises a liquid crystal display device,L k andL l respectively represent the firstkSum of alllSub-model, p k Representation ofL k Is p l Representation ofL l Is used for predicting the vector of the vector; matrix D t Is the first of (2)kLine 1lIs arranged as
By passing throughF(w) Lagrangian function, calculating the next iteration value w of the obtained diversity weighting vector t(+1) ：
w t(+1) =(α t e-β t )/( λ 1 D t )
Wherein alpha is t And beta t Representing the number of iterations astTime of dayF(w) non-negative lagrangian multipliers, e is a unit column vector of the corresponding dimension;
4) According to the augmentation vector z obtained in step 2) m t(+1) Obtaining a weight vector w m And deviation b m Combining the diversity function obtained in step 3)F(w) and iteration stepp，According to the sub-model objective function, calculating the sub-model objective function of the iterationA value;
if the number of iterationstEqual to 1, update the iteration numbert=t+1A kind of electronic device with high-pressure air-conditioning systemp=p+0.1Returning to the step 2) to continue iteration;
if the number of iterationstIf the difference is greater than 1, comparing the sub-model objective function value of the current iteration with the sub-model objective function value of the previous iteration, if the difference is less than a preset threshold (such as 0.001), or the number of iterationstIf the weight vector w is larger than the preset upper limit value (such as 20), stopping iteration, and obtaining the weight vector w in the iteration m Deviation b m Function of diversityF(w) iteration steppIs the corresponding optimal value, otherwise the iteration times are updatedt=t+1A kind of electronic device with high-pressure air-conditioning systemp=p+0.1Returning to the step 2) to continue iteration.
The training data set of the sub-model training and the test data set of the test sub-model prediction result are also from collected traffic data, namely, the traffic data obtained from different traffic data sampling points is divided into the training data set and the test data set, wherein the training data set is divided into a plurality of training subsets according to the corresponding traffic data sampling point positions, and each training data subset is input into a corresponding sub-model for training. After training is completed, the test data set is input into the generated sub-models, so that a prediction result of each sub-model is obtained, a final prediction value is given out in a weighted voting mode, and the prediction accuracy of the method is judged according to the prediction result of the test data.
The foregoing is merely a preferred embodiment of the present invention and is not intended to limit the scope of the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of the claims of the present invention.
Claims (7)
1. The traffic state prediction integration method is characterized by comprising the following steps of:
acquiring traffic sampling data from different traffic data sampling points as data to be predicted, wherein the data to be predicted is divided into a plurality of corresponding data subsets according to the corresponding traffic data sampling points;
inputting data to be predicted into a trained robust integrated model which maximizes diversity among all components, wherein the integrated model is also divided into a plurality of corresponding sub-models according to corresponding traffic data sampling point positions, each data subset of the data to be predicted is respectively input into the corresponding sub-model to obtain a prediction result of each sub-model, and the prediction result of each sub-model is synthesized to give a final prediction value;
the objective function of each sub-model is expressed as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,p∈(0,2]，||·|| 2 represents L 2 The distance of the norms,mrepresenting the sample type sequence number of each sub-model data,Cfor each sub-model data sample type total,c m is the first sub-modelmE, obtaining optimal parameters corresponding to the class sample in advance through an optimization algorithm m1 And e m2 Respectively matrix A m And B m Unit column vector of corresponding dimension, matrix a m Representing the first sub-modelmData set consisting of class samples, matrix B m Representing the division of each submodel bymThe remainder outside the class samplem-data set consisting of class 1 samples, matrix a m And B m Having the same dimension numbern，q m Is the first sub-modelmThe relaxation variable corresponding to the class sample is obtained by the constraint condition of the objective function,λ 1 is a preset parameter for balancing the prediction error and diversity of the integrated model,F(w) represents a diversity function, w m Representing the first sub-modelmWeight vector of class sample, b m Representing the first sub-modelmDeviation of class samples, w= [w 1 ,w 2 ,…,w L ] T For the diversity of the weight vectors of the model,Lis the total number of sub-models, w m 、b m 、F(w) andpby passing throughIterative optimization is carried out to obtain an optimal value; in the objective function, the first term and the second term belong to the prediction error of the integrated model, the third term represents diversity, and when optimization is completed, the first term and the second term are optimal;
obtaining w by iterative optimization m 、b m 、F(w) andpthe optimal value of (2) is obtained by iteratively updating the augmentation vector of each sub-model after the iteration is finished, and the specific steps are as follows:
1) Setting the iteration timestAnd iteration steppInitial values of 1 and 0.1, respectively, initializing an augmentation vector z m 1 Diagonal matrix D m1 1 And D m2 1 Wherein D is m1 1 Is the first of (2)iDiagonal elements and D m2 1 Is the first of (2)jThe diagonal elements are respectively:
a i representative matrix A m Middle (f)iLine data, b j Representative matrix B m Middle (f)jThe data of the rows are stored in a memory,m 1 representation matrix A m Is a function of the total number of rows of the system,m 2 representation matrix B m E mi1 Represent e m1 Middle (f)iLine data, e mj2 Represent e m2 Middle (f)jLine data;
initializing a diversity weighting vector w;
2) Calculating an augmentation vector z m t(+1) And according to the augmentation vector z m t(+1) Updating diagonal matrix D m1 t(+1) And D m2 t(+1) The method specifically comprises the following steps:
wherein, the liquid crystal display device comprises a liquid crystal display device,εthe regularization coefficient is I, and the I is an identity matrix;
3) When an augmentation vector z is obtained m t(+1) When determining diversity function using alternate direction multiplier methodF(w) specifically:
wherein, the liquid crystal display device comprises a liquid crystal display device,L k andL l respectively represent the firstkSum of alllSub-model, p k Representation ofL k Is p l Representation ofL l Is used for predicting the vector of the vector; matrix D t Is the first of (2)kLine 1lIs arranged as；
By passing throughF(w) Lagrangian function, calculating the next iteration value w of the obtained diversity weighting vector t(+1) ：
w t(+1) =(α t e-β t )/( λ 1 D t )
Wherein alpha is t And beta t Representing the number of iterations astTime of dayF(w) non-negative lagrangian multipliers, e is a unit column vector of the corresponding dimension;
4) According to the augmentation vector z obtained in step 2) m t(+1) Obtaining a weight vector w m And deviation b m Combining the diversity function obtained in step 3)F(w) and iteration stepp，Calculating a sub-model objective function value of the iteration according to the sub-model objective function;
if the number of iterationstEqual to 1, update the iteration numbert=t+1A kind of electronic device with high-pressure air-conditioning systemp=p+0.1Returning to the step 2) to continue iteration;
if the number of iterationstIf the difference is greater than 1, comparing the sub-model objective function value of the current iteration with the sub-model objective function value of the previous iteration, if the difference is less than the preset threshold, or the number of iterationstGreater than a presetThe upper limit value stops iteration, and the weight vector w obtained in the iteration is obtained m Deviation b m Function of diversityF(w) iteration steppIs the corresponding optimal value, otherwise the iteration times are updatedt=t+1A kind of electronic device with high-pressure air-conditioning systemp=p+0.1Returning to the step 2) to continue iteration.
2. The traffic state prediction integration method according to claim 1, wherein thec m Obtained by a drosophila optimization algorithm.
3. The traffic state prediction integration method according to claim 1, wherein the preset threshold value is 0.001.
4. The traffic state prediction integration method according to claim 1, wherein the preset upper limit value is 30.
5. The traffic state prediction integration method according to claim 1, wherein after each subset of data to be predicted is input into a corresponding sub-model to obtain a prediction result of each sub-model, weighted voting is performed on the prediction result of each sub-model to give a final prediction value.
6. The traffic state prediction integration method according to claim 1, wherein training data used for training the robust and integrated model that maximizes diversity among components is to obtain a part of traffic sample data from different traffic data sample points as a training data set, and divide the traffic sample data into a plurality of training subsets according to the corresponding traffic data sample points.
7. The traffic state prediction integration method according to claim 6, further selecting a part from the acquired traffic sample data as a test data set when training data is adopted, for testing the prediction result of the integrated model that is robust and maximizes diversity among components.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310314503.5A CN116030637B (en) | 2023-03-28 | 2023-03-28 | Traffic state prediction integration method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310314503.5A CN116030637B (en) | 2023-03-28 | 2023-03-28 | Traffic state prediction integration method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116030637A CN116030637A (en) | 2023-04-28 |
| CN116030637B true CN116030637B (en) | 2023-07-21 |
Family
ID=86077928
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310314503.5A Active CN116030637B (en) | 2023-03-28 | 2023-03-28 | Traffic state prediction integration method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116030637B (en) |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108171365B (en) * | 2017-12-15 | 2022-04-08 | 南京理工大学 | Traffic state prediction method based on improved SVM algorithm |
| CN109887279B (en) * | 2019-02-26 | 2021-01-05 | 北京隆普智能科技有限公司 | Traffic jam prediction method and system |
| CN109934274A (en) * | 2019-03-04 | 2019-06-25 | 南京理工大学 | Based on L2,pThe GEPSVM classification method of norm distance measure |
| CN110517494A (en) * | 2019-09-03 | 2019-11-29 | 中国科学院自动化研究所 | Forecasting traffic flow model, prediction technique, system, device based on integrated study |
| CN111462485A (en) * | 2020-03-31 | 2020-07-28 | 电子科技大学 | Traffic intersection congestion prediction method based on machine learning |
| CN114639243B (en) * | 2022-03-31 | 2022-09-27 | 四川九洲视讯科技有限责任公司 | Intelligent traffic prediction and decision method and readable storage medium |
-
2023
- 2023-03-28 CN CN202310314503.5A patent/CN116030637B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN116030637A (en) | 2023-04-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Tekouabou et al. | Improving parking availability prediction in smart cities with IoT and ensemble-based model | |
| CN109658695B (en) | Multi-factor short-term traffic flow prediction method | |
| Wang et al. | A hybrid model for prediction in asphalt pavement performance based on support vector machine and grey relation analysis | |
| CN113487066B (en) | Long-time-sequence freight volume prediction method based on multi-attribute enhanced graph convolution-Informer model | |
| CN111160311A (en) | Yellow river ice semantic segmentation method based on multi-attention machine system double-flow fusion network | |
| CN109117883B (en) | SAR image sea ice classification method and system based on long-time memory network | |
| CN111861013B (en) | Power load prediction method and device | |
| CN106339608A (en) | Traffic accident rate predicting system based on online variational Bayesian support vector regression | |
| CN104899135A (en) | Software defect prediction method and system | |
| CN111382676A (en) | Sand image classification method based on attention mechanism | |
| CN106528417A (en) | Intelligent detection method and system of software defects | |
| Guo et al. | Short‐term passenger flow forecast of urban rail transit based on GPR and KRR | |
| CN113516304B (en) | Regional pollutant space-time joint prediction method and device based on space-time diagram network | |
| CN112991721A (en) | Urban road network traffic speed prediction method based on graph convolution network node association degree | |
| CN101964061B (en) | Binary kernel function support vector machine-based vehicle type recognition method | |
| CN111027662A (en) | SD-LSSVR short-time traffic flow prediction method based on chaos quantum particle swarm optimization | |
| CN114048920A (en) | Site selection layout method, device, equipment and storage medium for charging facility construction | |
| CN113505536A (en) | Optimized traffic flow prediction model based on space-time diagram convolution network | |
| CN116187835A (en) | Data-driven-based method and system for estimating theoretical line loss interval of transformer area | |
| CN115496257A (en) | Short-term vehicle speed prediction based on space-time fusion | |
| CN114565187A (en) | Traffic network data prediction method based on graph space-time self-coding network | |
| Ye et al. | Multi-year ENSO forecasts using parallel convolutional neural networks with heterogeneous architecture | |
| CN115017970A (en) | Migration learning-based gas consumption behavior anomaly detection method and system | |
| CN114971675A (en) | Second-hand car price evaluation method based on deep FM model | |
| CN114742209A (en) | Short-term traffic flow prediction method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |