Regression prediction | MATLAB implements CNN-BiLSTM (convolutional bidirectional long short-term memory neural network) with multiple input and single output
Table of contents
List of effects
basic introduction
A convolutional-bidirectional long short-term memory (CNN-BiLSTM) model that considers both temporal and spatial factors is proposed, which combines a convolutional neural network (CNN) with spatial local feature extraction ability and a convolutional neural network (CNN) with long-term information in both forward and backward directions. Combined with bidirectional long short-term memory (BiLSTM), it is used to predict traffic flow that is more responsive to changes in time and space.
Research Review
- At present, there have been a lot of research results on traffic flow forecasting. The early methods include traditional linear forecasting methods. The linear forecasting method is simple and easy to operate, but cannot reflect the irregular traffic flow state well. In order to adapt to the constantly fluctuating traffic flow, and Traditional nonlinear forecasting methods have emerged. Later, with the development of computer technology and people's mastery of accurate real-time traffic, modern intelligent prediction methods, such as prediction methods based on machine learning and deep learning, emerged. Due to the more or less defects of a single method, a combined forecasting method has appeared in recent years, which is to combine different forecasting methods to make up for the shortcomings of a single forecasting method to achieve a better forecasting effect.
- A estimates the traffic flow of all road segments in the transportation network for which traffic data is not available, but can only predict short-term traffic flow.
- D transforms the time series analysis problem into an image analysis task. The proposed model has the ability to predict the incomplete traffic data of the road network, but these deep networks have complex architecture and weak interpretability.
- H Hierarchical temporal memory has potential as an effective tool for short-term traffic flow prediction, its effect is comparable to LSTM, and it improves when the traffic flow distribution changes, but it is not good at detecting abnormal flow from the model output and applying it to LSTM. to infer the existence of anomalous events.
- C connects the machine learning algorithm with the statistical model one after another, and post-processes it through ARIMA analysis, thereby significantly improving the accuracy of the prediction. The limitation is that the machine learning algorithm only considers the simplest traditional neural network.
- Combining the complementary functions of neural networks and fuzzy logic, M et al. achieved satisfactory results in short-term traffic flow prediction.
- L et al. proposed a method to extract the abstract and latent features of traffic flow based on the SAE model, but when the traffic flow is small, the small difference between the observed flow and the predicted flow will lead to a large relative error.
Model structure
The model consists of an input layer, a CNN layer, a BiLSTM layer, a fully connected layer and an output layer. The CNN layer consists of a stack of convolutional layers and maximum pooling layers. A Dropout layer is added at the end to randomly drop nodes to prevent overfitting. The LSTM model solves the long-term dependence problem of the recurrent neural network RNN. The unique "gate" structure can avoid gradient explosion and gradient disappearance, and has the advantage of strong long-term memory ability. While BiLSTM has the advantages of LSTM, it can also consider forward and backward bidirectional time series information in the time dimension, and the prediction is more comprehensive and accurate. CNN is suitable for extracting local spatial features, and BiLSTM takes into account the combination of two-way time series information, which can analyze traffic flow data more comprehensively from the spatial and temporal characteristics, so that the prediction results have a higher degree of fit.
summarize
Aiming at the characteristics of the spatiotemporal correlation of traffic flow sequences, a CNN-BiLSTM model combining the respective advantages of CNN and BiLSTM is proposed in this paper. The CNN-BiLSTM model extracts spatial and temporal features through CNN and LSTM, respectively, and through example verification analysis shows that compared with other benchmark models, the model in this paper can better adapt to the constantly fluctuating traffic flow data. Stability and accuracy are high.
References
[1] https://blog.csdn.net/kjm13182345320/article/details/127281969?spm=1001.2014.3001.5502
[2] https://blog.csdn.net/kjm13182345320/article/details/127261869?spm=1001.2014 .3001.5502
[3] https://blog.csdn.net/kjm13182345320/article/details/127179100?spm=1001.2014.3001.5502
[4] https://download.csdn.net/download/kjm13182345320/85476987?spm=1001.20 .3001.5503