GCN-LSTM 预测出租车速度 英文 Taxi Speed Prediction Using GCN-LSTM

GCN-LSTM模型预测出租车速度

GCN：又称GNN，图神经网络    LSTM：长短时记忆网络

可做学习参考

Summary

One of the most valuable findings in engineering is the determination of taxi speed. Since the GCN-LSTM program software can readily detect and calculate the taxi speed even when it is some distance before reaching the traffic system, implementing this type of engineering invention is a significant contribution to the world because improves people’s safety [1]. The anticipation of traffic is critical for an intelligent transportation board framework in a metropolitan zone. Traffic congestion may increase fuel consumption, pollute the air, and make travel plans challenging. It was created to predict taxi speeds. GCN-LSTM is a noteworthy development since it has resulted in revolutionary traffic behavior in cities and their outskirts, as automobiles are tracked without controlling their physical port. The GCN-LSTM model combines GCN and LSTM. Consequently, it combines the best features of both system paradigms. The GCN-LSTM displays data structure and material expectations based on time series data.

Taxi Speed Prediction Using GCN-LSTM

1. Introduction

       Determination of taxi speed is one of the most compelling discoveries in the engineering world. The application of this kind of engineering invention is a significant contribution to the world because it improves the safety of people. The GCN-LSTM program software can easily detect and determine the taxi speed even when it is some distance before reaching the traffic system [1]. One of the main reasons for improving intelligence is to enhance transportation. The board framework in a metropolitan region is traffic expectation. Traffic jams can raise fuel utilization, contaminate the air, and unveil it harder to carry out travel plans. Step by step instructions to precisely assess future street velocities and give partners demonstrating and dynamic devices are a significant problem for the board office. In the meantime, dependable and convenient street speed data is critical for individual explorers. It can decrease gridlock and help drivers in making informed travel decisions. Further developing traffic conditions are essential for improving civil proficiency, supporting the economy, and making life simpler for residents. Wise Transportation Systems (ITS), wherein traffic forecast assumes a crucial part, are one potential answer to reducing urban gridlock. None of the accurate anticipating stays a test because of the complex spatial-worldly reliance among traffic streams.

1. Research Background

       The applications of GCN-LSTM have been in place for quite some time. However, it has not been utilized in the specific and exact aspect of determining taxi speed because the research was demonstrated [2]. However, it has been shown in the design and application of the graphical convolution, the quality assessment systems, and the action recognition operations due to its high degree of sensitivity to action response [3]. Diagram Neural Networks (CNN) is a kind of profound gaining calculation that is utilized to induce information from charts. GNNs are brain networks that can be applied straightforwardly to diagrams, making hub-level, edge-level, and chart-level expectation occupations fundamental. It handles the test of ordering hubs (like records) in a chart (like a reference organization) where just a minor extent of corners has names; semi-administered learning. On diagrams, an illustration of semi-directed education, A few seats are not marked; they are obscure hubs [4]. A Graph Convolutional Network (GCN) is a semi-regulated learning technique for diagram-organized information. It is based on a quick variety of convolutional brain networks that straightforwardly work as far as weight sharing. The word 'convolution' in Graph Convolutional Networks is practically identical to Convolutional Neural Networks [5]. The essential qualification is in the information structure since GCNs are a summed-up variation of CNN that can interact information with non-normal geographies [6]. To prepare chart brain networks on diagrams too immense to even think about fitting in GPU memory, it is crucial to use the CPU to construct mini-batches of arbitrarily chosen chart hubs and edges, which communicates to the GPU alongside information portraying every corner - the hub highlights.

1. Significance of using GCN-LSTM

The invention of Taxi Speed Prediction Using GCN-LSTM is significant because it has transformed traffic behavior in the cities and their outskirts since the vehicles are monitored without necessarily operating the physical aspects [7]. The GCN-LSTM application has also translated to better change in transport and road ethics because Taxi Speed Prediction Using GCN-LSTM equips the taxi operators with a more disciplined behavioral framework that triggers them to remain ethically obedient in the use of road rules [8]. Apart from improving traffic and road safety's welfare and ethical conduct, the application of taxi speed prediction using GCN-LSTM increases the security, confidentiality, and moral standards of traffic and road behavior because the vehicle operators develop the required level of ethical behavior. One of the most valuable findings in engineering is the determination of taxi speed [9]. Since the GCN-LSTM program software can quickly detect and calculate the taxi speed even when it is some distance before reaching the traffic system, implementing this type of engineering invention is a significant contribution to the world because it improves people’s safety. The anticipation of traffic is critical for improving an intelligent transportation board framework in a metropolitan zone. Traffic congestion can increase fuel consumption, pollute the air, and make it more challenging to travel plans. Step-by-step guidelines for accurately estimating future roadway speeds and providing partners with exhibiting and dynamic gadgets are also important.

1. Detailed Description of the Model

The GCN Model

       The GCN model technology is an accurate and real-time traffic detection system that focuses on the application framework of the intelligent traffic system. The accuracy and the reliability of the GCN model also define its applicability in the design and application of traffic intelligence because it exhibits the real-time speed forecasting framework [10]. It offers an exceptional brain network-based traffic determining strategy, the convolutional diagram network (GCN) model, which is joined with the chart convolutional network (GCN) and the gated repetitive unit to catch spatial and fleeting reliance simultaneously (GRU) [11]. Its operating function depends on the variables of this function:

The operating function for GNC and LSTM models

 The gated repetitive unit is utilized to learn dynamic changes in rush hour gridlock information to catch fleeting reliance. At the same time, the GCN is used to understand complex topological designs to capture spatial dependence. The GCN model is then used to expect traffic utilizing the metropolitan street organization [12]. Tests show that the GCN model can separate spatio-fleeting connections from traffic information and that the forecasts beat state-of-the-art baselines on real-world datasets.

The LSTM Model

       LSTM model is an emerging technology used to classify information and predict materials based on time-series data. The reason and the working principle behind this aspect are that there are some lags of unfamiliar durations between the actual events and the time series being demonstrated. The LSTM model eliminates the vanishing gradient complexity in traditional RNNs [13]. It is a short-term memory whose components are designed as in the figure below.

Figure 1. LSTM design structure

The specialists fostered a spatio-worldly lengthy momentary memory network that went before by map-coordinating to gauge fine-grained traffic conditions. These models calculate the future given authentic information without considering the design of the street organization [14]. In the interim, a few investigations are seeing combining traffic diagrams for traffic learning and determination. To comprehend the associations between streets in the rush hour gridlock organization and expect the organization’s comprehensive traffic status, it is crucial to consider a remarkable profound learning structure called Traffic Graph Convolutional Long Short-Term Memory Neural Network (TGC-LSTM) [15]. Regardless of the way specific investigations have been led on the development of a street organization and it is wasteful to catch primary street characteristics without a diagram while thinking about spatial relations of the street, it operates on the forget and input gate, as shown below.

Model 2. LSTM model for forgetting and input gate

Besides, it has a super performance, based on the graphical representation below:

Figure 3. LSTM performance

      

GCN-LSTM Model

       The GCN-LSTM model is a combination of GCN and LSTM. For this reason, it comprises both aspects of the two system models. The GCN-LSTM shows data arrangement and material expectation considering time-series information. The justification for this component and the working hypothesis is certain deferrals of obscure lengths between the urgent occasions and the time series being shown [16]. Standard RNNs have decreasing inclination intricacy, which the GCN-LSTM model evades. To decide fine-grained traffic conditions, the specialists fostered a spacious-common extended passing memory network that was gone before by-map planning. The GCN-LSTM model innovation is an exact and constant traffic discovery framework that spotlights the insightful traffic framework's application design [17]. Since it shows a continuous speed anticipating structure, the GCN-LSTM model's precision and reliability decide its value in the plan and sending of traffic insight. It gives a great mind network-based traffic choosing technique, the outline GCN-LSTM model, joined with the graph GCN-LSTM and gated tedious unit to simultaneously catch spatial and short-lived reliance (GRU). It operates based on the Bayesian Spatio-temporal graph, as shown below:

Figure 4. GCN-LSTM Model considering Bayesian Spatial-temporal operation

1. Experimental Results and Analysis

Data Description

       The type of data, as indicated in the excel file “speed data set,” is a numerical statistic exhibited by the traffic taxi and other vehicles. On the left column is the taxi number plate. In contrast, the proceeding columns 0 to 2015 represent the speed prediction rationale of the cab across the specified geographical location where the GCN-LSTM system observer is installed [18]. The sensor distance data expresses the degree of the vicinity of the GCN-LSTM system observer. The longer the sensor distance of the GCN-LSTM system observer, the more accurate it becomes.

Data Processing Process

       The GCN-LSTM data is processed and expressed in an excel file, as demonstrated in the files named “speed data” and the “sensor distance data.” The speed data says the speed at which the GCN-LSTM detects the taxi vehicles while navigating around the city [19]. The nearer the taxi is to the GCN-LSTM observer model, the easier the ability of the system to detect and analyze the information for further ethical and legal action in case the taxi was traveling at a prohibited speed.

Training Process

       The training process in the GCN-LSTM is based on the analytical data present in the two data sets of the GCN-LSTM model “speed data” and the “sensor distance data.” [13] The training entails speed detection in determining and regulating traffic congestion [20]. In this regard, the application of the GCN-LSTM model “speed data” and the “sensor distance data” will draw from the type and quality of the training process since the proximity and the speed of the taxi will be examined and taken by the system for a determination whether the rate was within the required standards

Evaluation indicators and evaluation process

       The evaluation indicators of the GCN-LSTM model, as revealed in the data set “speed data” and the “sensor distance data.” From the data set, the evaluation critique validates the results because there is a high range and the number of vehicles that the system can be analyzed in terms of speeds at once [14] as the GCN-LSTM model “speed data” and the “sensor distance data” indicate, a total of 2015 vehicles can be evaluated.

Prediction result analysis

       The GCN-LSTM model “speed data” and the “sensor distance data” sample a few data sets” can predict the viability, as demonstrated by the graphical representation of the sample data taken from the “speed data.” Taking a sample to represent the visual analysis can help to predict the average framework of how the taxi is moving in the city, as shown in the sample graph below:

Graph 1. Graphical representation of the GCN-LSTM model “speed data” and the “sensor distance data.”

1. Conclusion and Outlook

       Taxi Speed Prediction was invented and the usage of GCN-LSTM is a notable invention since it has resulted in revolutionary traffic behavior in cities and their outskirts, as cars are tracked without operating the physical parts. GCN and LSTM are combined in the GCN-LSTM model. As a result, it incorporates both characteristics of the two system models. Considering time-series information, the GCN-LSTM displays data organization and material expectation.

References

1. Cui, K. Henrickson, R. Ke, and Y. Wang, “High-order graph convolutional recurrent neural network: A deep learning framework for network-scale traffic learning and forecasting,” arXiv preprint arXiv:1802.07007, 2018.
2. Lin, J. Li, F., J. Ye, and J. Huai, “Road traffic speed prediction: a probabilistic model is fusing     multisource data,” IEEE Transactions on Knowledge and Data Engineering, vol. 30, no.7, pp. 1310–1323, 2018.
3. Wang, Q. Gu, J. Wu, G. Liu, and Z. Xiong, “Traffic speed prediction and congestion source exploration: A deep learning method,” in 2016 IEEE 16th International Conference on

Data Mining (ICDM). IEEE, 2016, pp. 499–508.

1. Kim, P. Wang, and L. Mihaylova, “Structural recurrent neural network for traffic speed

prediction,” arXiv preprint arXiv:1902.06506, 2019.

1. Kim, P. Wang, Y. Zhu, and L. Mihaylova, “A capsule network for traffic speed prediction in complex road networks,” in 2018 Sensor Data Fusion: Trends, Solutions, Applications (SDF). IEEE, 2018, pp. 1–6.
2. Cui, R. Ke, and Y. Wang, “Deep bidirectional and unidirectional lstm recurrent neural network

for network-wide traffic speed prediction,” arXiv preprint arXiv:1801.02143, 2018.

1. Karim, S. Majumdar, H. Darabi, and S. Chen, “Lstm fully convolutional networks for time series classification,” IEEE Access, vol. 6, pp. 1662–1669, 2018.
2. Liao, J. Zhang, C. Wu, D. McIlwraith, T. Chen, S. Yang, Y. Guo, and F. Wu, “Deep sequence learning with auxiliary information for traffic prediction,” in Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. ACM, 2018, pp. 537–546.
3. Wei, J. Li, Q. Yuan, K. Chen, A. Zhou, and F. Yang, “Predicting fine-grained traffic conditions via spatiotemporal lstm,” Wireless Communications and Mobile Computing, vol. 2019, 2019.
4. Cui, K. Henrickson, R. Ke, and Y. Wang, “Traffic graph convolutional recurrent neural

network: A deep learning framework for network-scale traffic learning and forecasting,” 2018

1. Lv, Y. Duan, W. Kang, Z. Li, and F.-Y. Wang, “Traffic flow prediction with big data: a deep learning approach,” IEEE Transactions on Intelligent Transportation Systems, vol. 16, no. 2, pp. 865–873, 2015. Zhang, Y. Zheng, and D. Qi, “Deep Spatio-temporal residual networks for citywide crowd flow prediction,” in Thirty-First AAAI Conference on Artificial Intelligence, 2017.
2. Ma, Z. Dai, Z. He, J. Ma, Y. Wang, and Y. Wang, “Learning traffic as images: a deep

convolutional neural network for large-scale transportation network speed prediction,” Sensors, vol. 17, no. 4, p. 818, 2017. Fu, Z. Zhang, and L. Li, “Using lstm and GRU neural network methods for traffic flow prediction,” in 2016 31st Youth Academic Annual Conference of Chinese Association of Automation (YAC). IEEE, 2016, pp. 324– 328.

1.  Zhao, W. Chen, X. Wu, P. C. Chen, and J. Liu, “Lstm network: a deep learning approach for short-term traffic forecast,” IET Intelligent Transport Systems, vol. 11, no. 2, pp. 68–75, 2017.
2. Yu, Y. Li, C. Shahabi, U. Demiryurek, and Y. Liu, “Deep learning: A generic approach for extreme condition traffic forecasting,” in Proceedings of the 2017 SIAM International Conference on Data Mining. SIAM, 2017, pp. 777–785.
3. Zhang, X. Shi, J. Xie, H. Ma, I. King, and D.- Y. Yeung, “Gaan: Gated attention networks for learning on large and spatiotemporal graphs,” arXiv preprint arXiv:1803.07294, 2018.
4. Lu, L. Liu, J. Panneerselvam, B. Yuan, J. Gu, and N. Antonopoulos, “A gru-based prediction framework for intelligent resource management at cloud data centers in the age of 5g,” IEEE Transactions on Cognitive Communications and Networking, 2019.
5. Kipf and M. Welling, “Variational graph autoencoders,” arXiv preprint arXiv:1611.07308, 2016.
6. Kalofolias and N. Perraudin, “Large scale graph learning from smooth signals,” arXiv preprint arXiv:1710.05654, 2017.
7. Zhang, Y. Zheng, and D. Qi, “Deep Spatio-temporal residual networks for citywide crowd flow prediction,” in Thirty-First AAAI Conference on Artificial Intelligence, 2017.
8. Yu, Z. Wu, S. Wang, Y. Wang, and X. Ma, “Spatiotemporal recurrent convolutional networks for traffic prediction in transportation networks,” Sensors, vol. 17, no. 7, p. 1501, 2017.