CN114973665A - Short-term traffic flow prediction method combining data decomposition and deep learning - Google Patents
Short-term traffic flow prediction method combining data decomposition and deep learning Download PDFInfo
- Publication number
- CN114973665A CN114973665A CN202210544580.5A CN202210544580A CN114973665A CN 114973665 A CN114973665 A CN 114973665A CN 202210544580 A CN202210544580 A CN 202210544580A CN 114973665 A CN114973665 A CN 114973665A
- Authority
- CN
- China
- Prior art keywords
- traffic flow
- flow prediction
- short
- data
- decomposition
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0125—Traffic data processing
- G08G1/0129—Traffic data processing for creating historical data or processing based on historical data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Traffic Control Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a short-term traffic flow prediction method combining data decomposition and deep learning, which is based on complementary integrated empirical mode decomposition (CEEMD) and combines a Convolutional Neural Network (CNN) and a long-term short-term memory network (LSTM). The traffic flow prediction model reduces the influence of noise on traffic flow data prediction through CEEMD signal decomposition, and adopts CNN and LSTM to fully mine the space-time characteristics of data, so that the model can make more accurate judgment, thereby improving the learning efficiency of a neural network. The invention combines the advantages of data decomposition and deep learning to provide a short-term traffic flow prediction method, thereby achieving the purposes of more scientific means and improving the short-term traffic flow prediction accuracy.
Description
Technical Field
The invention discloses a short-term traffic flow prediction method combining data decomposition and deep learning, and relates to the field.
Background
With the development of intelligent traffic technology, traffic flow prediction is used as a basis for real-time traffic state judgment and traffic flow induction, and research on the traffic flow prediction has great significance for improving road service and management and control level. The decision can be actively made only by acquiring the state of the road ahead in advance. The effects of traffic control and induction generally take a period of time to accumulate before they can be manifested. According to different time scales, the traffic flow predicts two types of long-term (long-term) and short-term (short-term) to be used for predicting and analyzing the number of vehicles passing through a section of the road within 5-15 min. How to improve the accuracy of traffic flow prediction becomes one of the challenges of traffic flow research at the present stage, and accurate short-time traffic flow prediction is the most fundamental and effective measure for solving the problem of urban road traffic congestion and reducing environmental pollution at present. In recent years, many scholars at home and abroad have successively conducted a great deal of research on how to improve the prediction accuracy of urban short-term traffic flow. With the continuous accumulation of traffic flow data and the continuous development of related technologies, the prediction method is gradually evolving from the original classical statistical method to a data-driven view angle. However, traffic flow data is susceptible to interference from environmental factors such as weather and communication, and external factors, and the collected data has a lot of noise, high nonlinearity and complexity. At present, a short-term traffic flow prediction method with high accuracy and good stability is still lacked.
Disclosure of Invention
Aiming at the defects in the background technology, the invention provides a short-term traffic flow prediction method combining data decomposition and deep learning, and aims to improve the accuracy of short-term traffic flow prediction.
In order to achieve the purpose, the technical scheme adopted by the invention is as follows:
a short-term traffic flow prediction method combining data decomposition and deep learning comprises the following steps:
the acquired traffic prediction stream data is preprocessed,
adopting a traffic flow prediction model trained in advance to obtain a traffic flow prediction result according to the preprocessed traffic prediction flow number;
the traffic flow prediction model comprises: 1 complementary integrated empirical mode decomposition module, 1 convolutional neural network layer, 1 pooling layer, 2 spliced long-term and short-term memory network layers and 1 full connection layer;
the complementary integrated empirical mode decomposition module is used for signal decomposition of traffic flow prediction data and sequentially inputting the signal decomposition into a convolutional neural network layer, a pooling layer, a long-term and short-term memory network layer and a full connection layer;
the convolutional neural network layer is used for extracting the spatial characteristics of traffic flow prediction data;
the pooling layer is used for performing feature dimension reduction on traffic flow prediction data;
the long-short term memory network is used for extracting time characteristics of traffic flow prediction data;
and the full connection layer is used for outputting a traffic flow predicted value.
Further, the preprocessing of the traffic flow prediction data specifically includes: and (3) preprocessing the traffic flow prediction data by adopting a linear function min-max normalization method to form a traffic flow characteristic matrix.
Further, the method for training the traffic flow prediction model trained in advance specifically includes: constructing an initial traffic flow prediction model, inputting historical traffic flow data into the initial traffic flow prediction model, outputting a historical traffic flow predicted value by the initial traffic flow prediction model according to the input historical traffic flow data, comparing the historical traffic flow predicted value with a corresponding historical traffic flow real value, calculating a loss value between the historical traffic flow predicted value and the historical traffic flow real value, performing reverse error propagation according to the loss value, training the initial traffic flow prediction model, and determining a final traffic flow prediction model according to a training result.
Further, the method for training the initial traffic flow prediction model specifically includes: carrying out complementary integration empirical mode decomposition on the traffic flow characteristic matrix after traffic flow historical data preprocessing, inputting the traffic flow characteristic matrix into a convolutional neural network layer, and extracting spatial characteristics in the traffic flow characteristic matrix; taking the output of the convolutional neural network layer as the input of the pooling layer, and performing characteristic dimension reduction on the traffic flow characteristic matrix; taking the output of the pooling layer as the input of a first long-short term memory network layer, and extracting the time characteristics of the traffic flow characteristic matrix; and taking the output of the first long-short term memory network layer as the input of a second long-short term memory network layer, extracting the time characteristic of the traffic flow characteristic matrix, and inputting the extracted space or characteristic into an initial traffic flow prediction model for training.
Further, the performing complementary integration empirical mode decomposition on the through-flow feature matrix specifically includes: adding a group of complementary noises into the traffic flow characteristic matrix data, decomposing the complementary noises into a series of IMF components and a residual component, extracting the highest-frequency component as a random item of a traffic flow sequence, and superposing the residual IMF components to obtain a trend item of the traffic flow sequence:
wherein s (t) represents an input traffic flow characteristic matrix, x a(t) Representing the signal after adding positive noise; x is a radical of a fluorine atom b(t) Representing the signal after the addition of negative noise, n (t) representing a set of complementary noise added;
for x a( t ) 、x b(t) Simultaneously performing Empirical Mode Decomposition (EMD), and recording the decomposed IMF components as IMF sk_a 、IMF sk_b K ═ (1,2, …, n); wherein IMF sk_a Representing the decomposed signal after the addition of the kth positive noise;
IMF sk_b representsAdding the decomposed signal after the kth negative noise;
wherein, IMF sk Representing the lumped average value after adding the kth complementary noise; IMF s A series of IMF components and a residual component finally obtained after decomposition.
Further, extracting the spatial features in the traffic flow feature matrix specifically includes:
wherein the content of the first and second substances,is the output of the convolutional neural network layer;as input to convolutional neural network layers, i.e. IMF s ,Is the weight of the convolutional neural network layer,for the bias of the convolutional neural network layer,is the output of the pooling layer.
Further, extracting time characteristics of the traffic flow characteristic matrix:
f t =σ(W f ·[h t-1 ,x t ]+b f )
i t =σ(W i ·[h t-1 ,x t ]+b i )
o t =σ(W o ·[h t-1 ,x t ]+b o )
C t1 =tanh(W c ·[h t-1 ,x t ]+b c )
C t =C t-1 ·f t +i t ·C t1
h t =o t ·tanh(C t )
wherein f is t To forget the output of the door at the present moment, i t For the output of the input gate at the present moment, o t The output of the output gate at the current moment;
W f to forget the weight of the door, b f To forget the biasing of the door, W i As the weight of the input gate, b i For the biasing of the input gates, W o As weights of output gates, b o Is the offset of the output gate; w is a group of c Is the weight of the memory cell, b c In order to be the bias of the memory cell,
h t-1 input for the previous moment, x t The input of the current moment is the output of the pooling layer; sigma is an activation function sigmoid, and tanh is an activation function;
C t-1 for the state of the memory cell at the previous moment, C t For the state of the memory cell at the current moment, C t1 For the content to be updated in the memory unit, h t Is a predicted traffic flow sequence.
Further, the determining a final traffic flow prediction model according to the training result specifically includes:
selecting a root mean square error RMSE and an average absolute error MAE as evaluation indexes of a traffic flow prediction model; comparing the historical traffic flow predicted value and the historical traffic flow true value output by the model, judging whether the evaluation index meets the specified value range, and determining a final traffic flow prediction model:
wherein, y i Is a historical traffic flow prediction value, x i Is the true value of the historical traffic flow,the average value of the actual values of the historical traffic flow, and N is the number of the input historical data of the traffic flow; and when the root mean square error RMSE is less than 7 and the average absolute error MAE is less than 5, determining a final traffic flow prediction model.
Has the advantages that: according to the traffic flow data reconstruction method, a complementary integrated empirical mode decomposition module (CEEMD) is adopted to decompose traffic flow data according to noise frequency, the CEEMD is added into an original signal by introducing a group of Gaussian white noises with mutually opposite numbers, and EMD decomposition is carried out on two mixed signals at the same time, so that not only is the mode aliasing phenomenon of the EMD overcome, but also the noise residual quantity in the reconstructed data can be almost ignored because the introduced complementary noise is introduced; to a certain extent, CEEMD can also use relatively few integration average times, and the calculation time is saved under the condition of ensuring small residual noise interference;
the invention adopts a convolutional neural network layer (CNN) and a long-short term memory network Layer (LSTM) to establish a prediction model, and the CNN neural network performs characteristic extraction on data through one-dimensional convolution operation and pooling operation. The one-dimensional convolution operation realizes characteristic dimension reduction by controlling the number of convolution kernels, reduces the complexity of a model, further realizes characteristic dimension reduction by pooling operation on the basis of the convolution operation, reduces an overfitting phenomenon, and solves the problems of gradient explosion and long-term data dependence of RNN by LSTM.
The invention provides a short-term traffic flow prediction method combining data decomposition and deep learning; the accurate short-time traffic flow prediction has important practical significance in the field of intelligent traffic, can provide more accurate road information for travelers, improves the road passing efficiency, and solves the problems of traffic jam and the like. The method strengthens the connection between vehicles and roads, and is an important link for building a smart city.
Drawings
FIG. 1 is an overall flow chart of the present invention;
FIG. 2 is a schematic diagram of the CEEMD structure of the present invention;
FIG. 3 is a schematic diagram of the LSTM structure of the present invention.
Detailed Description
The following describes the embodiments in further detail with reference to the accompanying drawings. The following examples are only for illustrating the technical solutions of the present invention more clearly, and the protection scope of the present invention is not limited thereby.
In a Python environment, a Keras neural network framework is used for completing the construction and training of the model. The data set selected here is from the Caltrans Performance Measurement System (PeMS), which is collected in real time from each probe across the highway System in major metropolitan areas, California, and the traffic data for 1 lane point was selected for experimental verification. The traffic flow data of the embodiment are divided into a training set of 90% and a test set of 10%, the time line of the test set data is behind the time line of the training set, the data of the training set is used for a traffic flow prediction model which is trained in advance, and the data of the test set is used for inputting the traffic flow prediction model which is trained in advance to predict the traffic flow in a short time in the future; the specific implementation is as follows:
one embodiment is shown in fig. 1: a short-term traffic flow prediction method combining data decomposition and deep learning comprises the following steps:
the acquired traffic prediction stream data is preprocessed,
adopting a traffic flow prediction model trained in advance to obtain a traffic flow prediction result according to the preprocessed traffic prediction flow number;
the traffic flow prediction model comprises: 1 complementary integrated empirical mode decomposition module, 1 convolutional neural network layer, 1 pooling layer, 2 spliced long-term and short-term memory network layers and 1 full connection layer;
the complementary integrated empirical mode decomposition module is used for signal decomposition of traffic flow prediction data and sequentially inputting the signal decomposition into a convolutional neural network layer, a pooling layer, a long-term and short-term memory network layer and a full connection layer;
the convolutional neural network layer is used for extracting the spatial characteristics of traffic flow prediction data;
the pooling layer is used for performing feature dimension reduction on traffic flow prediction data;
the long-short term memory network is used for extracting time characteristics of traffic flow prediction data;
and the full connection layer is used for outputting a traffic flow predicted value.
Specifically, the layer 1 data decomposition layer decomposes an original signal into a series of IMF components and a residual component; the 2 nd layer is an input layer and determines the dimension of model input; and the 3 rd and 4 th layers are a CNN layer and a pooling layer, and feature extraction is performed on the data through a one-dimensional convolution operation and a pooling operation. And selecting the proper convolution kernel number and the activation function (Relu). The 5 th layer to the 6 th layer are LSTM layers, data characteristics are extracted, and proper neuron Number (Number of nerves), loss function (los), Optimizer (Optimizer), Batch size (Batch _ size), training times (Epochs) and a data ratio value which is randomly discarded by adding a Drapout mechanism are selected. The 7 th layer is a full connection layer Dense, the characteristics fused at the output end of the CNN-LSTM are used as the input of the full connection layer, and the proper number of the neurons is selected.
The preprocessing of the traffic flow prediction data specifically comprises the following steps: and preprocessing the traffic flow prediction data by adopting a min-max normalization method to form a traffic flow characteristic matrix.
The training method of the pre-trained traffic flow prediction model specifically comprises the following steps: constructing an initial traffic flow prediction model, inputting historical traffic flow data into the initial traffic flow prediction model, outputting a historical traffic flow predicted value by the initial traffic flow prediction model according to the input historical traffic flow data, comparing the historical traffic flow predicted value with a corresponding historical traffic flow real value, calculating a loss value between the historical traffic flow predicted value and the historical traffic flow real value, performing reverse error propagation according to the loss value, and training the initial traffic flow prediction model,
the method for training the initial traffic flow prediction model specifically comprises the following steps: carrying out complementary integration empirical mode decomposition on the traffic flow characteristic matrix after traffic flow historical data preprocessing, inputting the traffic flow characteristic matrix into a convolutional neural network layer, and extracting spatial characteristics in the traffic flow characteristic matrix; taking the output of the convolutional neural network layer as the input of the pooling layer, and performing characteristic dimension reduction on the traffic flow characteristic matrix; taking the output of the pooling layer as the input of a first long-short term memory network layer, and extracting the time characteristics of the traffic flow characteristic matrix; and taking the output of the first long-short term memory network layer as the input of a second long-short term memory network layer, extracting the time characteristic of the traffic flow characteristic matrix, and inputting the extracted space or characteristic into an initial traffic flow prediction model for training.
As shown in fig. 2, the performing complementary integrated empirical mode decomposition on the through-flow feature matrix specifically includes: adding a group of complementary noises into the traffic flow characteristic matrix data, decomposing the complementary noises into a series of IMF components and a residual component, extracting the highest-frequency component as a random item of a traffic flow sequence, and superposing the residual IMF components to obtain a trend item of the traffic flow sequence:
wherein s (t) represents a traffic flow characteristic matrix, x a(t) Representing the signal after adding positive noise; x is the number of b(t) Representing the signal after the addition of negative noise, n (t) representing a set of complementary noise added;
for x a( t ) 、x b(t) EMD decomposition is carried out simultaneously, and the decomposed IMF components are respectively recorded as IMF sk_a 、IMF sk_b K ═ (1,2, …, n); wherein IMF sk_a Representing the decomposed signal after the addition of the kth positive noise;
IMF sk_b representsAdding the decomposed signal after the kth negative noise;
wherein, IMF sk Representing the lumped average value after adding the kth complementary noise; IMF s A series of IMF components and a residual component that are finally obtained after decomposition.
The extracting of the spatial features in the traffic flow feature matrix specifically comprises:
wherein the content of the first and second substances,is the output of the convolutional neural network layer;as input to convolutional neural network layers, i.e. IMF s ,Is the weight of the convolutional neural network layer,for the bias of the convolutional neural network layer,is the output of the pooling layer.
As shown in fig. 3, the time characteristic of the traffic flow characteristic matrix is extracted:
f t =σ(W f ·[h t-1 ,x t ]+b f )
i t =σ(W i ·[h t-1 ,x t ]+b i )
o t =σ(W o ·[h t-1 ,x t ]+b o )
C t1 =tanh(W c ·[h t-1 ,x t ]+b c )
C t =C t-1 ·f t +i t ·C t1
h t =o t ·tanh(C t )
wherein f is t To forget the output of the door at the present moment, i t For the output of the input gate at the present moment, o t The output of the output gate at the current moment;
W f weight of forgetting gate, b f To forget the biasing of the door, W i As the weight of the input gate, b i For the biasing of the input gates, W o As weights of output gates, b o Is the offset of the output gate; w c Is the weight of the memory cell, b c In order to be the bias of the memory cell,
h t-1 input for the previous moment, x t The input of the current moment is the output of the pooling layer; sigma is an activation function sigmoid, and tanh is an activation function;
C t-1 for the previous moment of memory cell state, C t For the state of the memory cell at the present moment, C t1 For the content to be updated in the memory unit, h t Is a predicted traffic flow sequence.
The model of the invention transmits IMF signal components after CEEMD decomposition to the CNN layer to extract the space-time characteristics of data, and adopts LSTM to enhance the memory capacity of the neural network and improve the learning efficiency. In the present invention, the detailed hyper-parameter settings are given by table 1.
TABLE 1
Determining a final traffic flow prediction model according to the training result: selecting the root mean square error RMSE, the mean absolute error MAE and the decision coefficient R 2 As an evaluation index of a traffic flow prediction model; comparing the historical traffic flow predicted value and the historical traffic flow true value output by the model, judging whether the evaluation index meets the specified value range, and determining a final traffic flow prediction model:
wherein, y i Is a historical traffic flow prediction value, x i Is the true value of the historical traffic flow,is the average value of the actual values of the historical traffic flow; and when the root mean square error RMSE is less than 7 and the average absolute error MAE is less than 5, determining a final traffic flow prediction model.
The model of the invention was experimentally compared on the PeMS public data set with 5 other models on 3 indices, with the results shown in Table 2.
Model structure | MAE | RMSE | R 2 score |
LSTM | 8.7467 | 11.3662 | 0.9137 |
CNN-LSTM | 7.9690 | 10.4812 | 0.9266 |
EMD-LSTM | 6.9989 | 9.1929 | 0.9436 |
CEEMD-LSTM | 5.9453 | 7.5417 | 0.9620 |
CEEMD-CNN-LSTM | 4.0592 | 5.6985 | 0.9783 |
TABLE 2
Experimental results show that the traffic flow time sequence prediction model provided by the invention is used for evaluating indexes RMSE, MAE and R 2 Can achieve better effect. Compared with the existing model, the prediction precision is obviously improved, and the change of the road traffic flow can be reflected more accuratelyCharacteristic; the problem that a short-time traffic flow prediction method with high accuracy rate is lacked at present is solved.
The above description is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, several modifications and variations can be made without departing from the technical principle of the present invention, and these modifications and variations should also be regarded as the protection scope of the present invention.
Claims (8)
1. A short-term traffic flow prediction method combining data decomposition and deep learning is characterized by comprising the following steps:
the acquired traffic prediction stream data is preprocessed,
adopting a traffic flow prediction model trained in advance to obtain a traffic flow prediction result according to the preprocessed traffic prediction flow number;
the traffic flow prediction model comprises: complementary integration empirical mode decomposition module, convolution neural network layer, pooling layer, spliced long-short term memory network layer and full connection layer;
the complementary integrated empirical mode decomposition module is used for signal decomposition of traffic flow prediction data and sequentially inputting the signal decomposition into a convolutional neural network layer, a pooling layer, a long-term and short-term memory network layer and a full connection layer;
the convolutional neural network layer is used for extracting the spatial characteristics of traffic flow prediction data;
the pooling layer is used for performing feature dimension reduction on traffic flow prediction data;
the long-short term memory network is used for extracting time characteristics of traffic flow prediction data;
and the full connection layer is used for outputting a traffic flow predicted value.
2. The short-term traffic flow prediction method combining data decomposition and deep learning according to claim 1, wherein the preprocessing of the traffic flow prediction data specifically comprises: and preprocessing the traffic flow prediction data by adopting a min-max normalization method to form a traffic flow characteristic matrix.
3. The short-term traffic flow prediction method combining data decomposition and deep learning according to claim 1, wherein the training method of the pre-trained traffic flow prediction model specifically comprises: constructing an initial traffic flow prediction model, inputting historical traffic flow data into the initial traffic flow prediction model, outputting a historical traffic flow predicted value by the initial traffic flow prediction model according to the input historical traffic flow data, comparing the historical traffic flow predicted value with a corresponding historical traffic flow real value, calculating a loss value between the historical traffic flow predicted value and the historical traffic flow real value, performing reverse error propagation according to the loss value, training the initial traffic flow prediction model, and determining a final traffic flow prediction model according to a training result.
4. The short-term traffic flow prediction method combining data decomposition and deep learning according to claim 3, wherein the method for training the initial traffic flow prediction model specifically comprises the following steps: carrying out complementary integration empirical mode decomposition on the traffic flow characteristic matrix after traffic flow historical data preprocessing, inputting the traffic flow characteristic matrix into a convolutional neural network layer, and extracting spatial characteristics in the traffic flow characteristic matrix; taking the output of the convolutional neural network layer as the input of the pooling layer, and performing characteristic dimension reduction on the traffic flow characteristic matrix; taking the output of the pooling layer as the input of a first long-short term memory network layer, and extracting the time characteristics of the traffic flow characteristic matrix; and taking the output of the first long-short term memory network layer as the input of a second long-short term memory network layer, extracting the time characteristic of the traffic flow characteristic matrix, and inputting the extracted space or characteristic into an initial traffic flow prediction model for training.
5. The short-term traffic flow prediction method combining data decomposition and deep learning according to claim 4,
the complementary integration empirical mode decomposition of the through-flow feature matrix specifically comprises the following steps: adding a group of complementary noises into the traffic flow characteristic matrix data, decomposing the complementary noises into a series of IMF components and a residual component, extracting the highest-frequency component as a random item of a traffic flow sequence, and superposing the residual IMF components to obtain a trend item of the traffic flow sequence:
wherein s (t) represents an input traffic flow characteristic matrix, x a(t) Representing the signal after adding positive noise; x is the number of b(t) Representing the signal after the addition of negative noise, n (t) representing a set of complementary noise added;
for x a(t) 、x b(t) EMD decomposition is carried out simultaneously, and the decomposed IMF components are respectively recorded as IMF sk_a 、IMF sk_b K ═ (1,2, …, n); wherein IMF sk_a Representing the decomposed signal after the addition of the kth positive noise;
IMF sk_b represents the decomposed signal after the addition of the kth negative noise;
wherein, IMF sk Representing the lumped average value after adding the kth complementary noise; IMF s A series of IMF components and a residual component are finally obtained after decomposition; n is the number of complementary noise added.
6. The short-term traffic flow prediction method combining data decomposition and deep learning according to claim 4, wherein extracting spatial features in a traffic flow feature matrix specifically comprises:
wherein the content of the first and second substances,is the output of the convolutional neural network layer;as input to convolutional neural network layers, i.e. IMF s ,Is the weight of the convolutional neural network layer,for the bias of the convolutional neural network layer,is the output of the pooling layer.
7. The short-term traffic flow prediction method combining data decomposition and deep learning according to claim 4, characterized in that the time features of the traffic flow feature matrix are extracted:
f t =σ(W f ·[h t-1 ,x t ]+b f )
i t =σ(W i ·[h t-1 ,x t ]+b i )
o t =σ(W o ·[h t-1 ,x t ]+b o )
C t1 =tanh(W c ·[h t-1 ,x t ]+b c )
C t =C t-1 ·f t +i t ·C t1
h t =o t ·tanh(C t )
wherein f is t To forget the output of the door at the present moment, i t For the output of the input gate at the present moment, o t The output of the output gate at the current moment;
W f weight of forgetting gate, b f To forget the biasing of the door, W i As the weight of the input gate, b i For the biasing of the input gates, W o As weights of output gates, b o Is the offset of the output gate; w is a group of c Is the weight of the memory cell, b c In order to be the bias of the memory cell,
h t-1 input for the previous moment, x t The input of the current moment is the output of the pooling layer; sigma is an activation function sigmoid, and tanh is an activation function;
C t-1 for the state of the memory cell at the previous moment, C t For the state of the memory cell at the present moment, C t1 For the content to be updated in the memory unit, h t Is a predicted traffic flow sequence.
8. The short-term traffic flow prediction method combining data decomposition and deep learning according to claim 1, wherein the determining a final traffic flow prediction model according to the training result specifically comprises:
selecting a root mean square error RMSE and an average absolute error MAE as evaluation indexes of a traffic flow prediction model; comparing the historical traffic flow predicted value and the historical traffic flow true value output by the model, judging whether the evaluation index meets the specified value range, and determining a final traffic flow prediction model:
wherein, y i Is a historical traffic flow prediction value, x i Is the true value of the historical traffic flow,is the average value of the actual values of the historical traffic flow; n is the number of the input traffic flow historical data.
And when the root mean square error RMSE is less than 7 and the average absolute error MAE is less than 5, determining a final traffic flow prediction model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210544580.5A CN114973665A (en) | 2022-05-19 | 2022-05-19 | Short-term traffic flow prediction method combining data decomposition and deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210544580.5A CN114973665A (en) | 2022-05-19 | 2022-05-19 | Short-term traffic flow prediction method combining data decomposition and deep learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114973665A true CN114973665A (en) | 2022-08-30 |
Family
ID=82985434
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210544580.5A Pending CN114973665A (en) | 2022-05-19 | 2022-05-19 | Short-term traffic flow prediction method combining data decomposition and deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114973665A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115514439A (en) * | 2022-09-26 | 2022-12-23 | 华工未来科技(江苏)有限公司 | Channel air interface utilization rate prediction method, system, electronic equipment and medium |
CN116504076A (en) * | 2023-06-19 | 2023-07-28 | 贵州宏信达高新科技有限责任公司 | Expressway traffic flow prediction method based on ETC portal data |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111241466A (en) * | 2020-01-15 | 2020-06-05 | 上海海事大学 | Ship flow prediction method based on deep learning |
CN112488415A (en) * | 2020-12-14 | 2021-03-12 | 国网江苏省电力有限公司经济技术研究院 | Power load prediction method based on empirical mode decomposition and long-and-short-term memory network |
CN112801386A (en) * | 2021-02-04 | 2021-05-14 | 北京工业大学 | Short-term traffic flow prediction method based on three-stage model |
CN113487855A (en) * | 2021-05-25 | 2021-10-08 | 浙江工业大学 | Traffic flow prediction method based on EMD-GAN neural network structure |
CN113705864A (en) * | 2021-08-13 | 2021-11-26 | 华北水利水电大学 | Weather drought prediction method and device based on VMD-CNN-BilSTM-ATT mixed model |
CN113722308A (en) * | 2021-09-02 | 2021-11-30 | 重庆大学 | Acceleration response data completion method and device based on EEMD-MultiCNN-LSTM |
CN113988357A (en) * | 2021-09-03 | 2022-01-28 | 重庆大学 | High-rise building wind-induced response prediction method and device based on deep learning |
CN114091766A (en) * | 2021-11-24 | 2022-02-25 | 东北电力大学 | Space load prediction method based on CEEMDAN-LSTM |
-
2022
- 2022-05-19 CN CN202210544580.5A patent/CN114973665A/en active Pending
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111241466A (en) * | 2020-01-15 | 2020-06-05 | 上海海事大学 | Ship flow prediction method based on deep learning |
CN112488415A (en) * | 2020-12-14 | 2021-03-12 | 国网江苏省电力有限公司经济技术研究院 | Power load prediction method based on empirical mode decomposition and long-and-short-term memory network |
CN112801386A (en) * | 2021-02-04 | 2021-05-14 | 北京工业大学 | Short-term traffic flow prediction method based on three-stage model |
CN113487855A (en) * | 2021-05-25 | 2021-10-08 | 浙江工业大学 | Traffic flow prediction method based on EMD-GAN neural network structure |
CN113705864A (en) * | 2021-08-13 | 2021-11-26 | 华北水利水电大学 | Weather drought prediction method and device based on VMD-CNN-BilSTM-ATT mixed model |
CN113722308A (en) * | 2021-09-02 | 2021-11-30 | 重庆大学 | Acceleration response data completion method and device based on EEMD-MultiCNN-LSTM |
CN113988357A (en) * | 2021-09-03 | 2022-01-28 | 重庆大学 | High-rise building wind-induced response prediction method and device based on deep learning |
CN114091766A (en) * | 2021-11-24 | 2022-02-25 | 东北电力大学 | Space load prediction method based on CEEMDAN-LSTM |
Non-Patent Citations (1)
Title |
---|
卢海鹏;: "基于VMD-BiLSTM-BLS 模型的短时交通流 预测", 计算机系统应用 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115514439A (en) * | 2022-09-26 | 2022-12-23 | 华工未来科技(江苏)有限公司 | Channel air interface utilization rate prediction method, system, electronic equipment and medium |
CN116504076A (en) * | 2023-06-19 | 2023-07-28 | 贵州宏信达高新科技有限责任公司 | Expressway traffic flow prediction method based on ETC portal data |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112418547B (en) | Bus stop passenger flow prediction method based on GCN-LSTM combination model | |
CN110889546B (en) | Attention mechanism-based traffic flow model training method | |
CN111915059B (en) | Attention mechanism-based Seq2Seq berth occupancy prediction method | |
Tian | Approach for short-term traffic flow prediction based on empirical mode decomposition and combination model fusion | |
CN111613054B (en) | Multi-step traffic speed prediction method cooperatively considering space-time correlation and contribution difference | |
CN110220725B (en) | Subway wheel health state prediction method based on deep learning and BP integration | |
CN114973665A (en) | Short-term traffic flow prediction method combining data decomposition and deep learning | |
CN111709549A (en) | Navigation reminding method for short-time traffic flow prediction based on SVD-PSO-LSTM | |
CN111862592B (en) | Traffic flow prediction method based on RGCN | |
CN114664091A (en) | Early warning method and system based on holiday traffic prediction algorithm | |
CN115148019A (en) | Early warning method and system based on holiday congestion prediction algorithm | |
CN114781538A (en) | Air quality prediction method and system of GA-BP neural network coupling decision tree | |
CN112966871A (en) | Traffic jam prediction method and system based on convolution long-short term memory neural network | |
CN113205698A (en) | Navigation reminding method based on IGWO-LSTM short-time traffic flow prediction | |
Li et al. | Short term traffic flow prediction based on LSTM | |
Liu et al. | Explanatory prediction of traffic congestion propagation mode: A self-attention based approach | |
CN115496257A (en) | Short-term vehicle speed prediction based on space-time fusion | |
CN111141879B (en) | Deep learning air quality monitoring method, device and equipment | |
CN114970815A (en) | Traffic flow prediction method and device based on improved PSO algorithm optimized LSTM | |
Li et al. | Parking demand forecasting based on improved complete ensemble empirical mode decomposition and GRU model | |
Haputhanthri et al. | Short-term traffic forecasting using LSTM-based deep learning models | |
Feng et al. | A dual-staged attention based conversion-gated long short term memory for multivariable time series prediction | |
CN114596726A (en) | Parking position prediction method based on interpretable space-time attention mechanism | |
Chen et al. | Improved LSTM based on attention mechanism for short-term traffic flow prediction | |
CN116946183A (en) | Commercial vehicle driving behavior prediction method considering driving capability and vehicle equipment |
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 |