Simulation system and method for predicting road transportation capacity in rockfill dam field based on vehicle speed

Abstract

The invention provides a simulation system and method for predicting road transportation capacity in a rock-fill dam field based on vehicle speed. The simulation system comprises a vehicle speed monitoring module, a vehicle speed data set, a vehicle speed space-time prediction module and a road transportation evaluation module. The vehicle speed monitoring module is used for deleting data of an abnormal track of a transport vehicle, and deleting repeated data and isolated data to generate first standard vehicle speed data; the vehicle speed data set adopts a regional range to divide first standard vehicle speed data to obtain a vehicle speed training set, a vehicle speed verification set and a vehicle speed test set; the vehicle speed space-time prediction module is used for establishing time and space relevance of each node of a vehicle speed training set, a vehicle speed verification set and a vehicle speed test set to obtain second standard vehicle speed data; and the road transportation evaluation module outputs the road transportation capacity according to the second standard vehicle speed data and a road segment saturation flow method. The construction road transportation capacity evaluation system can accurately evaluate the construction road transportation capacity.

Description

Technical Field

[0001] The present invention belongs to the field of construction simulation of rockfill dams, and particularly relates to a simulation system and method for the transportation capacity of in-site roads of a rockfill dam based on vehicle speed prediction. Background Art

[0002] Rockfill dams have the characteristics of being able to make full use of local materials, adapting to various different topographical, geological and climatic conditions, having strong seismic resistance, low cost, and simple structure, and have developed into one of the most widely used dam types in the construction of water conservancy and hydropower projects at home and abroad. [1] During the construction process of a rockfill dam, it is necessary to continuously and repeatedly excavate, transport and fill a large amount of materials, such as soil and rock; this determines that the transportation capacity of the construction roads of the rockfill dam is the key factor restricting the construction progress of the rockfill dam project. After loading at the material yard, the vehicle travels along a specific route, passes through several roads and several intersections, and then arrives at the unloading point for unloading; the unloaded vehicle returns to the original working face and waits for the loader to load again. [2] In this cycle, if the road transportation capacity is too low, it will lead to the inability to supply dam materials in time, resulting in delays in the project progress. On the contrary, when the road transportation capacity is too high and the filling intensity cannot keep up with the feeding intensity, it will lead to the retention of dam materials at the filling site, which will affect the normal construction, and at the same time, it will also lead to too high investment in building roads, causing economic losses. However, the traditional transportation capacity is often evaluated based on empirical formulas, and it is impossible to consider the impacts of extremely bad weather, time differences, and uncertain factors on the road transportation capacity during the actual construction process. Existing research mostly uses the method of model simulation to predict the transportation capacity of the construction roads of the rockfill dam, so as to obtain transportation simulation results such as vehicle driving density.

[0003] In the field of in-site traffic simulation of water conservancy and hydropower project construction, the research work started earliest at Tianjin University. In 1987, Zhu Guangxi, Sun Xiheng, etc. [3] used the system simulation method to simulate the construction process of the Longtan concrete face rockfill dam located on the Hongshui River of the Nanpan River in Guangxi, and at the same time simulated the in-site traffic conditions, so as to obtain the driving density of the access road during each construction period, and scientifically demonstrated the arrangement of construction machinery and the transportation scheduling plan. Liu Ning [4] et al. established a transportation model for a high core wall rockfill dam by globally considering the overall construction process, real-time statistically calculated the driving density and queuing situation, and optimized the machinery matching based on the utilization rate of transportation machinery. Shen Mingliang [5] et al. linked the earth-rock allocation and the transportation intensity, and carried out multi-objective joint decision optimization to achieve the optimization of earth-rock allocation and machinery matching. Liu Xu [6]Integrate the earthwork allocation, transportation, and dam surface operations, simulate the entire transportation process in real-time, and conduct comprehensive analysis by comparing the simulation results with the monitoring results. Hu Chao, Dong Jingyan [7-8] Based on the allocation results, use the effective construction days to establish a transportation model and optimize the transportation machinery. Cao Jiayun [9] etc. analyzed the traffic layout of the Lianghekou Hydropower Project, established a transportation model using the cyclic network simulation technology, and analyzed the transportation plan and mechanical matching plan. Ma Xiaohang

[10] etc. simulated the transportation situation through simulation technology, obtained information such as road transport queuing and mechanical utilization rate, and verified the rationality of the earthwork allocation. Zhong Denghua etc.

[11] Establish a traffic simulation model through digital monitoring theory, and be able to obtain real-time construction information. Zhang Ping obtained information such as the position and speed of dump trucks through real-time monitoring of the transportation system. Li Zexin

[12] Through the analysis of the overall layout of hydraulic structures, adopt the method of combining roads, traffic tunnels, and bridges to realize the traffic design of the project construction and serve the construction site. Zhao Yu

[13] etc. analyzed the transportation module system of the concrete face rockfill dam, based on the earthwork allocation results, adopted the cyclic queuing theory, and used Anylogic software to establish a transportation system simulation model to realize agent simulation.

[0004] The vehicle driving speed is a key parameter in the road transport simulation model. Its change may cause significant changes in the cycle time and construction progress, so it has an important impact on the simulation results. The probability distributions used by most simulation programs are taken from the historical databases of previous projects or expert experience, and may not be reasonably used when facing changing and dynamic scenarios

[14] . Because the vehicle speed is highly affected by time and space, there will be different speed parameters during day and night, on straight roads and curves.

[0005] Wang Wei et al.

[15] Collected the point speeds at 1-minute intervals, and used the grey system theory to establish a GM(1,1) grey prediction model for predicting the spot speeds of expressways to predict the point speeds. M.H. Hajimiri

[16] etc. combined predictive control with fuzzy control, and used simple mathematical methods to predict the road surface slope and vehicle speed on the future driving route of the vehicle through the data collected by the vehicle GPS. Corrado de Fabritiis

[17] etc. based on the real-time floating car data (FCD), respectively based on two algorithms of artificial neural network and pattern matching, and predicted the average vehicle speed of the target section for 5-10 steps, that is, 15-30 minutes, through the average vehicle speeds of the current and adjacent sections every 3 minutes. Jungme Park

[18] et al. mainly introduced a traffic model for vehicle speed prediction based on neural networks. Ruoqian

[19] et al. proposed a speed prediction algorithm KTM-SP (Kinetic Traffic Modeling-Speed Prediction) based on an aerodynamic traffic model.

[0006] Existing research on vehicle speed prediction mostly focuses on single-point speed or average speed, etc., and does not consider the impact of the correlation between time and space on speed prediction too much, resulting in insufficient prediction accuracy of existing vehicle speed prediction models. Therefore, this paper proposes a vehicle speed prediction method based on spatio-temporal graph neural networks, and uses the vehicle speed predicted considering spatio-temporal relationships as simulation parameters to input into the simulation model of the transportation capacity of the construction road of the rockfill dam for the simulation of the maximum transportation capacity of the construction road.

[0007] [1] Wang Bole, Liu Yingzhen, Wu Hehe. New progress in the construction of earth-rock dams in China [J]. Hydropower Generation, 2005(01): 63-65.

[0008] [2] B. Ozdemir, M. Kumral, "Stochastic assessment of the transportation efficiency of earthwork industrial materials", J. Constr. Eng. Manage. 143(8)(2017)05017013,

[0009] [3] D. Zhong, N. Liu, B. Cui. Research on the construction traffic simulation of high core-wall rockfill dams based on digital monitoring, Journal of Water Resources and Hydropower Engineering / Journal of Hydropower Engineering 31(2012)223-230+236.

[0010] [4] Zhu Guangxi, Sun Xiheng. Special report on the simulation research of the construction system of the Longtan concrete face rockfill dam [J]. Department of Water Resources and Harbor Engineering, Tianjin University, 1988.

[0011] [5] Liu Ning, Zhong Denghua, Zhang Ping. Research on the visual simulation of the construction site traffic of high core-wall rockfill dams [J]. Water Resources and Hydropower Engineering, 2012, 43(4): 73-78.

[0012] [6] Shen Mingliang, Liu Xingang, Chen Gang, et al. Joint optimization model of earth-rock allocation and road transportation intensity of rockfill dams [J]. Journal of Wuhan University (Engineering Edition), 2006(5): 14-18.

[0013] [7] Hu Chao. Research on the dynamic simulation and resource optimization of the dam surface operation of concrete face rockfill dams [D]. Yichang: China Three Gorges University, 2012.

[0014] [8] Dong Jingyan. Research on the visual simulation of the construction process of concrete face rockfill dams [D]. Yichang: China Three Gorges University, 2010.

[0015] [9] Cao Jiayun, Liu Ning, Song Qitao. Simulation Analysis of In-site Traffic for Lianghekou Hydropower Project [J]. Design of Hydroelectric Power Station, 2013, 29(2): 61-65.

[0016]

[10] Ma Xiaohang, Dong Suo, Jiang Shaoyang, et al. Simulation Study on Earthwork Allocation Road Transportation System for Rockfill Dam Project [J]. Water Resources and Power, 2012, 31(31): 142-144, 184.

[0017]

[11] Zhong Denghua, Liu Ning, Cui Bo. In-site Traffic Simulation for Construction of High Core Rockfill Dam Based on Digital Monitoring

[0018] Research [J]. Journal of Hydroelectric Engineering, 2012, 31(6): 223-230, 236.

[0019]

[12] Zhang Ping. Research on Construction Simulation and Schedule Control of High Core Rockfill Dam Based on Real-time Monitoring [D]. Tianjin: Tianjin University, 2011.

[0021]

[13] Zhao Yu, Wang Gaowei, Sun Kai, et al. Research on Traffic Simulation System of Concrete Face Rockfill Dam Based on Anylogic [J]. Guangdong Water Resources and Hydropower, 2021, 1: 1-6.

[0023]

[14] Fei Lv, Jiajun Wang, Bo Cui, et.al. An improved extreme gradient boosting approach to vehicle speed prediction for construction simulation of earthwork [J]. Automation in Construction 119(2020)103351.

[0025]

[15] Wang Wei, Dong Decun. Analysis of Point Speed Prediction Model Based on Grey Theory [J]. Transportation Science & Technology and Economy, 2010, 12(2): 1-4.

[0027]

[16] M.H. Hajimiri, F.R. Salmasi. A Fuzzy Energy Management Strategy for Series Hybrid Electric Vehicle with Predictive Control and Durability Extension of the Battery[C]. Proceedings of 2006 IEEE Conference on Electric and Hybrid Vehicles (ICEHV). Pune, Dec, 10 - 20, 2006: 1 - 5.

[0028]

[17] Corrado de Fabritiis, Roberto Ragona GV. Traffic Estimation and Prediction Based on Real Time Floating Car Date[C]. Proceedings of the 11th International IEEE Conference on Intelligent Transportation Systems. Beijing, China, October 12 - 15, 2008: 197 - 203.

[18] Jungme Park, Dai Li, Yi L. Murphey, et.al. Real Time Vehicle Speed Prediction using a Neural Network Traffic Model[C]. Proceedings of 2011 International Joint Conference on Neural Networks. San Jose, California, USA, July 31–August 5, 2011: 2991 - 2996.

[0030]

[19] Ruoqian Liu, Shen Xu, Jungme Park, et.al. Real Time Vehicle Speed

[0031] Predition using Gas - Kinetic. Summary of the Invention

[0032] The present invention aims to overcome the deficiencies of existing construction road transportation capacity evaluation methods, consider the impact of the correlation between time and space on the prediction of the vehicle speed of transport vehicles, establish a simulation model for the transportation of construction roads of rockfill dams, simulate the maximum transportation capacity of the roads, and provide theoretical guidance for the design and construction of construction roads. The present invention obtains the driving speed of the transport vehicles of the rockfill dam through a real-time monitoring system, and preprocesses the vehicle speed data to enhance its accuracy. Then, a vehicle speed prediction model based on a spatio-temporal graph neural network is established based on the preprocessed vehicle speed data, fully considering the impact of the correlation between time and space on the vehicle speed prediction, and predicting the vehicle speed of the transport vehicles in different spatio-temporal distributions. Finally, a simulation model of the transportation road of the rockfill dam is established through VISSIM, and the predicted vehicle speed is used as the input of the simulation parameters; and the maximum input method is used for the simulation to obtain the maximum transportation capacity of the construction road of the rockfill dam.

[0033] To solve the problems of the existing technology, the present invention adopts the following technical solutions:

[0034] A simulation system for the transportation capacity of the in-site roads of a rockfill dam based on vehicle speed prediction, the simulation system includes a vehicle speed monitoring module, a vehicle speed data set, a vehicle speed spatio-temporal prediction module, and a road transportation evaluation module; the vehicle speed spatio-temporal prediction module includes a first fully connected layer, a plurality of spatio-temporal fusion graph neural network layers, a second fully connected layer, and a third fully connected layer; wherein: each spatio-temporal fusion graph neural network layer is composed of a plurality of parallel spatio-temporal fusion graph neural units and gated convolutional units; the spatio-temporal fusion graph unit is composed of a stacked multi-layer graph multiplication block and a pooling layer; the gated unit is composed of a first extended convolutional part and a second extended convolutional part; wherein:

[0035] The vehicle speed monitoring module is used to delete the data of abnormal trajectories of transport vehicles, duplicate data, and isolated data to generate first standard vehicle speed data;

[0036] The vehicle speed data set is obtained by dividing the first standard vehicle speed data by a regional range to obtain a vehicle speed training

[0037] set, a vehicle speed verification set, and a vehicle speed test set;

[0038] The vehicle speed spatio-temporal prediction module is used to establish the time and space correlation of each node of the vehicle speed training set, the vehicle speed verification set, and the vehicle speed test set to obtain second standard vehicle speed data;

[0039] The road transportation evaluation module outputs the road transportation capacity according to the second standard vehicle speed data according to the section saturation flow method.

[0040] Further, the process of the vehicle speed monitoring module for deleting the data of abnormal trajectories of transport vehicles, duplicate data, and isolated data to generate first standard vehicle speed data includes:

[0041] Traverse the satellite positioning trajectory records in the order of sample collection time, and sequentially determine whether the instantaneous speed value of the dump truck is within the set reasonable range. If not, it is regarded as an abnormal satellite positioning trajectory record and deleted;

[0042] Traverse and determine whether the difference in longitude and latitude between two adjacent satellite positioning trajectory records is less than the set threshold, identify duplicate satellite positioning trajectory records and delete them;

[0043] Sequentially determine whether the positioning time difference between the current record row and the adjacent record row is equal to the sampling time interval, identify isolated satellite positioning trajectory records and delete them.

[0044] Furthermore, the process of dividing the first standard vehicle speed data into a vehicle speed training set, a vehicle speed verification set, and a vehicle speed test set by using a regional range for the vehicle speed data set includes:

[0045] Use Z-score standardization to process the first standard vehicle speed data to obtain the number of nodes, the number of edges, the time interval, and the missing value rate of each data set;

[0046]

[0047] Split the data of the four data sets into a training set, a verification set, and a test set according to a ratio of 6:2:2.

[0048] Furthermore, the process of the vehicle speed spatio-temporal prediction module establishing the time and space correlation of each node in the vehicle speed training set, the vehicle speed verification set, and the vehicle speed test set to obtain the second standard vehicle speed data includes:

[0049] Represent the road network as a topological graph G=(V, E, A SG ), where: V is the set of nodes |V| = N 2 , corresponding to N sensor observations; E is the set of edges, and A SG ∈R N×N is the spatial graph of the spatial adjacency matrix, representing the proximity or distance of nodes;

[0050] According to the observation graph signal Construct the temporal graph A of the spatial graph information G at time step t TC ;

[0051] The spatio-temporal fusion graph neuron can obtain spatial correlation from A SG for each node, obtain temporal correlation from A TG , and obtain self-correlation from A TC , and calculate the relevant feature data of each node time series according to the following formula:

[0052] h l+1 =(A * h​l W 1 +b 1 )⊙σ(A * h l W 2 +b 2 )

[0053] Among them: h l is the hidden state of the STFGN module; A * is the abbreviation of the spatio-temporal fusion graph; A STFG ∈R KN×KN ,W 1 ,W 2 ∈R C ×C ,b 1 ,b 2 ∈R C are the model parameters of GLU respectively; ⊙ represents the Hadamard product, and σ represents the sigmoid function;

[0054] The gating convolutional unit obtains the dependent feature data of each node time series according to the following formula:

[0055] Y = φ(Θ 1 *X + a)⊙σ(Θ 2 *X + b)

[0056] Among them: φ(·) and σ(·) are activation functions, and Θ1 and Θ2 are two independent 1D convolutional operations with a dilation rate of K - 1;

[0057] The second standard vehicle speed data is obtained by training the relevant feature data and dependent feature data of each node time series through the following loss function;

[0058]

[0059] Among them: δ is a hyperparameter used to control the sensitivity of the mean squared error loss.

[0060] The present invention can also adopt a simulation method for predicting the transportation capacity of the in-site road of a rockfill dam, including the following steps:

[0061] Delete the abnormal trajectory data, duplicate data, and isolated data of the transport vehicles to generate the first standard vehicle speed data;

[0062] Divide the first standard vehicle speed data using a regional range with a ratio of 6:2:2 to obtain a vehicle speed training set, a vehicle speed verification set, and a vehicle speed test set;

[0063] Establish the time and space correlation of each node of the vehicle speed training set, the vehicle speed verification set, and the vehicle speed test set to obtain the second standard vehicle speed data; Among them:

[0064] The road network is represented as a topological graph G = (V, E, A SG ), where: V is the set of nodes, |V| = N 2 , corresponding to N sensor observations; E is the set of edges, and A SG ∈R N×N is the spatial graph of the spatial adjacency matrix, representing the proximity or distance of nodes;

[0065] According to the observed graph signal construct the temporal graph A of the spatial graph information G at time step t TC ;

[0066] The spatio-temporal fusion graph neural unit can obtain spatial correlation from A SG for each node, obtain temporal correlation from A TG , and obtain self-correlation from A TC . The relevant feature data of each node's time series is calculated according to the following formula:

[0067] h l+1 =(A * h l W 1 +b 1 )⊙σ(A * h l W 2 +b 2 )

[0068] where: h l is the hidden state of the STFGN module; A * is the abbreviation of the spatio-temporal fusion graph; A STFG ∈R KN×KN , W 1 , W 2 ∈R C ×C , b 1 , b 2 ∈R C are the model parameters of GLU respectively; ⊙ represents the Hadamard product, and σ represents the sigmoid function;

[0069] The gated convolutional unit obtains the dependent feature data of each node's time series according to the following formula:

[0071] Y = φ(Θ 1 *X + a)⊙σ(Θ 2 *X + b)

[0072] where: φ(·) and σ(·) are activation functions, Θ1 and Θ2 are two independent 1D convolutional operations, and the dilation rate is K - 1;

[0073] The relevant feature data and dependent feature data of each node time series are trained through the following loss function to obtain the second standard vehicle speed data;

[0074]

[0075] where: δ is a hyperparameter used to control the sensitivity of the squared error loss;

[0076] According to the second standard vehicle speed data, the road transportation capacity is output according to the section saturation flow method.

[0077] Beneficial effects

[0078] Advantages of the present invention:

[0079] 1. By obtaining real vehicle data through a real-time monitoring system and performing multiple preprocessings on the obtained data, the accuracy and authenticity of the vehicle speed data can be enhanced more.

[0080] 2. A vehicle speed prediction model based on a spatio-temporal graph neural network is adopted. The dynamic time warping (DTW) algorithm is used to calculate the similarity of the historical time series of nodes, and a time graph is dynamically constructed according to the results. The time graph is fused with other graphs to help the model extract hidden spatio-temporal features, which can effectively solve the problem that the graph construction process lacks rich information as support; at the same time, a threshold diffusion convolution module is also used to comprehensively consider the relationship between global and local correlations. It can comprehensively consider the influence of time and space on the vehicle speed, making the predicted vehicle speed more accurate.

[0081] 3. Based on VISSIM, a simulation model is established, comprehensively considering the vehicle speed distribution of different sections, characteristics of the rockfill dam construction road (such as many curves and large elevation differences, etc.), and the most realistic rockfill dam construction road transportation capacity is obtained through simulation. Description of the drawings

[0082] Figure 1 It is a schematic diagram of a simulation system for the transportation capacity of the rockfill dam site road based on vehicle speed prediction of the present invention.

[0083] Figure 2 It is a schematic diagram of the spatio-temporal fusion graph module related to the present invention. Detailed implementation manners

[0084] The following is an explanation of the present invention in conjunction with the attached Figure 1 ~Attached Figure 2 The following is an explanation of the present invention:

[0085] As Figure 1As shown in the figure, the present invention provides a simulation system for the transportation capacity of the in-site roads of a rockfill dam based on vehicle speed prediction. The simulation system includes a vehicle speed monitoring module, a vehicle speed data set, a vehicle speed spatio-temporal prediction module, and a road transportation evaluation module. The vehicle speed spatio-temporal prediction module consists of a first fully connected layer, multiple spatio-temporal fusion graph neural network layers, a second fully connected layer, and a third fully connected layer. Among them: Each spatio-temporal fusion graph neural network layer is composed of a spatio-temporal fusion graph, multiple parallel spatio-temporal fusion graph neural units, and a gated convolutional unit. The spatio-temporal fusion graph unit is composed of a stacked multi-layer graph multiplication block and a pooling layer. The gated unit is composed of a first extended convolutional part and a second extended convolutional part. Among them:

[0086] The vehicle speed monitoring module is used to delete the data of abnormal trajectories of transport vehicles, duplicate data, and isolated data to generate first standard vehicle speed data. The vehicle speed monitoring module is mainly divided into the processing of abnormal trajectory data, duplicate data, and isolated data. Among them:

[0087] First, obtain the original dump truck trajectory data through the on-dam transportation monitoring system, and screen and delete abnormal records that do not meet the requirements. Traverse the satellite positioning trajectory records in the order of sample collection time, and sequentially judge whether the instantaneous speed value of the dump truck is within the set reasonable range. If not, it is regarded as an abnormal satellite positioning trajectory record and deleted. Then traverse again and judge whether the difference in longitude and latitude between two adjacent satellite positioning trajectory records is less than the set threshold, identify duplicate satellite positioning trajectory records and delete them. Finally, sequentially judge whether the positioning time difference between the current record row and the adjacent record row is equal to the sampling time interval, identify isolated satellite positioning trajectory records and delete them.

[0088] (1) Vehicle speed acquisition and data preprocessing based on the real-time monitoring system. During the construction process of rockfill dam filling, dump trucks continuously transport filling materials that meet the quality requirements from the material yard transportation point to the dam body partition filling surface. However, the rockfill dam project has the characteristics of a large construction workload, and the dam materials sources are often scattered in multiple locations. Therefore, the travel time of dam material transportation to the dam has a direct impact on the overall progress of the project. And the vehicle speed of the dump truck is the most important factor affecting the transportation travel time. To realize the simulation of the dam material transportation process, the main construction simulation parameters to be determined refer to the driving speed of the dump truck on any transportation path of the dam area road network.

[0089] By installing positioning devices on dump trucks, the dynamic tracking of the dump truck transportation process is realized by using satellite positioning technology. Through the wireless communication network deployed in the dam area, the positioning terminal on the dump truck sends its real-time positioning information to the central control center database at a certain sampling time interval to realize real-time traffic information collection. When a large number of dump trucks move on the dam area road network, the dump truck trajectory data collected based on the on-dam transportation monitoring system can reflect the in-site real-time traffic conditions.

[0090] Define the dump truck driving trajectory data as a sequence of dump truck positioning points arranged in chronological order under a certain sampling time interval. A single dump truck trajectory record mainly includes the license plate number of the dump truck, the positioning date, the positioning time, the longitude, the latitude, and the instantaneous speed, etc.

[0091] Due to the influence of factors such as signal occlusion, traffic congestion, and network transmission errors, the original satellite positioning driving trajectory data of the dump trucks collected by the on-dam transportation monitoring system may contain abnormal, isolated, and duplicate records. To ensure the reliability of the subsequent path travel time distribution estimation results, before performing map matching calculations, first preprocess the original satellite positioning trajectory data of the dump trucks and delete the trajectory records that do not meet the requirements. The specific steps include the following sub-steps:

[0092] (1.1) Delete the abnormal satellite positioning trajectory records of the dump trucks. The driving speed of the dump trucks in the construction site should be within a reasonable range. When it exceeds a certain threshold, it can be determined as an abnormal satellite positioning trajectory record and deleted.

[0093] (1.2) Delete the isolated satellite positioning trajectory records of the dump trucks. The satellite positioning trajectory records that are missing both the adjacent previous moment and the adjacent next moment records are called isolated satellite positioning trajectory records and cannot be used to calculate the non-zero value section travel time samples and are deleted during preprocessing. During the on-dam transportation monitoring process, satellite positioning equipment failures and network transmission errors may occur, resulting in partial moment missing measurements in the dump truck trajectory records. Deleting abnormal trajectory records may also lead to the appearance of isolated trajectory records. By looping through all the satellite positioning trajectory records and determining whether the positioning time difference between the current record row and the adjacent record row is equal to the sampling time interval, the isolated satellite positioning trajectory records can be identified.

[0094] (1.3) Delete duplicate satellite positioning track records of dump trucks. When the vehicle is waiting or avoiding at intersections, the dump truck may pause and stay at a fixed position during transportation, resulting in duplicate satellite positioning track records. Specifically, in multiple adjacent satellite positioning track records, the longitude and latitude field data of the dump truck are exactly the same, and the instantaneous speed of the dump truck is 0 km / h. In addition, due to the influence of satellite positioning static drift, when the driving speed of the dump truck is slow, the positioning position of the dump truck may randomly fluctuate slightly around a certain longitude and latitude position. The above two types of track records cannot be used to calculate effective non-zero value section travel time samples and are deleted during preprocessing. For duplicate satellite positioning track record sequences, only the first record is retained. By looping through all satellite positioning track records and determining whether the difference in longitude and latitude between adjacent two records is less than the set threshold, duplicate satellite positioning track records can be identified.

[0095] The vehicle speed dataset is divided into a vehicle speed training set, a vehicle speed validation set, and a vehicle speed test set by using a regional range for the first standard vehicle speed data; the production of the vehicle speed dataset mainly divides the obtained preprocessed data into four datasets according to the regional range, performs standardized processing on the input of the preprocessed data, and then splits the data of the four datasets into a training set, a validation set, and a test set according to a ratio of 6:2:2; the processed vehicle operation data is more accurate and reliable.

[0096] For the production of the dataset, the obtained preprocessed data is divided into four datasets according to the regional range, and Z-score standardization is used to perform standardized processing on the input of the preprocessed data to obtain the number of nodes, the number of edges, the time interval, and the missing value rate of each dataset. Then, the data of the four datasets are split into a training set, a validation set, and a test set according to a ratio of 6:2:2. The historical data of 12 consecutive time steps in one hour is used to predict the data of 12 consecutive time steps in the next hour. STFGNN is evaluated more than 10 times in each common dataset.

[0097] The vehicle speed spatio-temporal prediction module is used to establish the time and space correlation of each node in the vehicle speed training set, the vehicle speed validation set, and the vehicle speed test set to obtain the second standard vehicle speed data; the vehicle speed spatio-temporal prediction module based on the spatio-temporal graph neural network is composed of a first fully connected layer, multiple spatio-temporal fusion graph neural network layers, a second fully connected layer, and a third fully connected layer; each spatio-temporal fusion graph neural network layer consists of a spatio-temporal fusion graph, spatio-temporal fusion graph neural units, and a gated convolutional unit.

[0098] Establishment of a spatio-temporal vehicle speed prediction model based on graph neural networks. The vehicle speed prediction model structure adopted in the present invention includes an input layer composed of a single fully connected layer, a plurality of stacked spatio-temporal fusion graph neural network layers (SpatioTemporalfusion graph neural network, STFGN layers), and an output layer composed of two fully connected layers. Each STFGN layer is composed of a plurality of parallel spatio-temporal fusion graph neural modules (STFGN modules) and a gated convolutional (CNN) module, and the latter includes two parallel one-dimensional dilated convolutional modules. Specifically, as Figure 1 shown.

[0099] The road network is represented as a graph G=(V, E, A SG ), where V is a finite set of nodes |V| = N 2 , corresponding to N sensor observations; E is the set of edges, and A SG ∈R N×N is the spatial adjacency matrix, representing the proximity or distance between nodes. Denote the observed graph signal as the observation of the spatial graph information G at time step t, and its elements are d traffic features (e.g., speed, volume) observed by each sensor. The purpose of traffic prediction is to learn a function f from the first T speed observations in the road network, so as to predict the next T' traffic speeds from N relevant sensors.

[0100] The input data is independently and parallelly processed by multiple STFGN modules, which is time-saving and can capture more complex correlations. Then, the concatenation of the outputs of each STFGN module is added to the output of the gated convolutional (CNN) to form the input of the next STFGN layer. Note that each STFGN layer will clip the input. The specific construction of the STFGN layer includes the following sub-steps:

[0101] Among them:

[0102] (3.1) Construction of the Speed Spatiotemporal Fusion Graph (STFG). To consider as much information as possible during the graph construction process, a method is proposed to dynamically construct a temporal graph based on the traditional method of given spatial graphs and then fuse this graph with the spatial graph. The purpose of dynamically constructing the temporal graph is to obtain a specific graph structure with more accurate dependency and real relationships than using only the given spatial graph alone. This method can make the deep learning model lighter because the fusion graph already has the relevant information of each node with its spatial neighbor nodes, the relevant information of each node with nodes of similar temporal patterns, and the state information of the node itself along the time axis. Here, the DTW algorithm is also used to calculate the correlation of each node's time series. The DTW algorithm is a classic algorithm for calculating sequence correlation, but it has a time complexity. To reduce the complexity of DTW, its "search length" T is limited to 12 time steps.

[0103] (3.2) Spatiotemporal Fusion Graph Neural (STFGN) Unit. The basic graph convolution operator uses a spatial method based on matrix multiplication instead of the graph convolution method based on spectral analysis. Residual connections and max pooling operations are also introduced for each layer. Finally, the features corresponding to the intermediate time steps are saved as the output. Combining Figure 2 It can be seen that the basic graph convolution operator uses a spatial method based on matrix multiplication instead of the graph convolution method based on spectral analysis, thus eliminating the need to calculate the Laplacian matrix of the adjacency matrix. By performing multiple matrix multiplications on A STFG , each node in the graph can obtain spatial correlation from A SG , temporal correlation from A TG , and self-correlation from A TC . The gating mechanism in LSTM is also used in the graph multiplication block. The calculation formula of the graph multiplication block is as follows, where h l represents the hidden state of the l-th layer of STFGN, and both W and h are learnable parameters.

[0104] h l+1 = (A * h l W 1 + b 1 ) ⊙ σ(A * h l W 2 + b 2 ) (1)

[0105] In the formula: h l is the hidden state of the STFGN module. A * is the abbreviation of the speed spatiotemporal fusion graph. A STFG ∈ R KN×KN , W1 , W 2 ∈R C×C , b 1 , b 2 ∈R C are the model parameters of GLU respectively. ⊙ represents the Hadamard product, and σ represents the sigmoid function.

[0106] Stacking multiple graph multiplication blocks can aggregate more complex non-local spatial correlations. Residual connections and max pooling operations are also introduced for each layer. Finally, the features corresponding to the intermediate time steps are saved as the output. The output features contain complex heterogeneities. In each matrix multiplication, the A in the middle of the diagonal SG transmits information through spatial neighbor nodes, and A TG provides its own self-connection information for each node along the time axis in the horizontal and vertical directions. The A at the corner TC can enhance the information from nodes with similar time patterns. The input vehicle speed data will independently and in parallel pass through multiple STFGN modules, and their outputs will be concatenated and added to the output of the gated convolutional module and used as the input for the next STFGN layer.

[0107] (3.3) Gated convolutional unit. Gated convolution with a large dilation rate is introduced to consider the long-term spatio-temporal dependence of the node itself. Additionally, Huber Loss is used as the loss function for the entire model training.

[0108] (3.3) Gated convolutional module. Although A STFG can extract global spatio-temporal correlations by integrating A TG , the correlations it contains come more from other nodes, and the long-term spatio-temporal dependence of the node itself is also very important. Therefore, gated convolution with a large dilation rate is also introduced, and its formula is as follows:

[0109] Y = φ(Θ 1 *X + a) ⊙ σ(Θ 2 *X + b) (2)

[0110] where φ(·) and σ(·) are activation functions, Θ1 and Θ2 are two independent 1D convolutional operations with a dilation rate of K - 1. It can expand the receptive field along the time axis, thereby enhancing the model performance to extract the dependencies of the vehicle speed time series.

[0111] Additionally, Huber Loss is used as the loss function for the entire model training. As follows:

[0112]

[0113] δ is a hyperparameter used to control the sensitivity of the squared error loss.

[0114] The road transportation evaluation module outputs the road transportation capacity according to the second standard vehicle speed data by the road section saturation flow method. The vehicle speed predicted by the road transportation evaluation module simulates the road to obtain the maximum traffic capacity of the road.

[0115] (4.1) Establish a road model based on the import of CAD drawings. Import the CAD road design drawings of the actual project into the VISSIM software, and initially establish the model of the road according to the actual design shape and scale.

[0116] (4.2) Parameter selection. Mainly include setting the weight, width, power and type of the vehicle; setting the length, width, number of lanes and shape characteristics of the road; setting the safety distance parameter.

[0117] (4.3) Simulation based on the maximum input method. Use the maximum input method to simulate and predict the maximum traffic capacity of the road. Input a large traffic flow to form a long queue of vehicles at the entrance, and the output flow reaches stability at this time, which is the maximum traffic capacity.

[0118] The present invention first obtains information such as the driving speed, position and time of vehicles on the rockfill dam construction road through a real-time monitoring system, and preprocesses this information to make the data more real and accurate. Then, a vehicle speed prediction model is established based on the spatio-temporal graph neural network using the obtained vehicle speed data, fully considering the influence of the correlation between time and space on vehicle speed prediction, and the vehicle speed is determined through this model for vehicle speed prediction. Finally, the predicted vehicle speed is used to establish a simulation model of the rockfill dam construction road through VISSIM, and the maximum input method is used for simulation to obtain the maximum transportation capacity of the road during the construction of the rockfill dam.

[0119] (4) Conduct road transportation capacity simulation based on VISSIM. VISSIM is a microscopic simulation modeling tool based on time intervals and driving behavior. It can analyze various traffic conditions, such as lane settings, traffic composition, traffic signals, bus stops, etc., and is an effective tool for evaluating traffic engineering designs and urban planning schemes. It mainly adopts a microscopic traffic flow simulation model including car-following and lane-changing logics. The construction of the simulation model includes the following sub-steps:

[0120] (4.1) Establish a road model based on the import of CAD drawings. First, import the road design CAD drawings into VISSIM. Determine the shape characteristics and length of the road according to the CAD drawings. Initially establish the road model. Then, set parameters such as the number of roads, road width, and slope according to the design requirements to refine the road model.

[0121] (4.2) Parameter selection. Since there are significant differences in the transportation speeds of dump trucks at different times on different road sections, and there are many other parameters that will have some impact on the simulation results in addition to speed. Therefore, the selection of these parameters is crucial.

[0122] Firstly, the vehicle type is set. For the transport vehicles of the rockfill dam, generally larger truck types are selected, and according to the actual situation, the width, weight, and power of the vehicle are set. Weight and power will have some impact on the vehicle speed on sloped roads, but through comparison of the simulation results, the impact is not very significant.

[0123] Then, the speed is selected. On different road sections, such as straight roads and curves, there will be different speeds; for different vehicle types, there are also corresponding different speeds. Therefore, the speed parameter is set to the vehicle speed predicted in 2. The vehicle speed parameter is a range parameter, not an exact value, and is randomly selected according to a random distribution, which is more practical and increases randomness.

[0124] Finally, the safety distance is selected. VISSIM uses the Wiedemann 74 model to calculate the safety distance. The safety distance calculation formula is:

[0125] d_safe = ax+(bx_add+bx_mult*z)*v 1 / 2 (3)

[0126] Where: ax is the average stopping distance, bx_add is the additional part, bx_mult is the multiple part, v is the vehicle driving speed, and z is a random number in (0, 1). Appropriate parameter sizes are selected through result comparison and national regulations requirements.

[0127] (4.3) Simulation of road transport capacity based on the output flow method. In the real traffic flow, the arrival numbers of vehicles are random and discrete, and the number of traffic-generated vehicles per unit time is also random. Since there is no mutual influence and external interference among vehicles at the traffic flow generation position on the road section, the number of vehicle arrivals per unit time conforms to the Poisson distribution probability.

[0128] The output flow method is a method for simulating the saturated flow of a road section. This method uses the output file to determine the saturated flow of the road section. When the input flow exceeds the saturated flow of the road section, after a period of simulation operation, at the road section entrance, due to the reflection of the blocking wave, a large number of vehicles queue outside the road section, and at the same time, the number of vehicle generations per unit time is very large, resulting in the queue of vehicles getting longer and longer. This means that even if the input flow is increased further, the traffic flow passing through the road section per unit time cannot increase due to the existence of the blocking wave. Then, the number of vehicles passing through the road section per hour at this time is the saturated flow of the road section. However, it is necessary to run the simulation for several hours to calculate the average hourly flow for several hours to accurately obtain the saturated value of this road section.

[0129] Based on the output flow method, vehicle inputs are set on the road. The larger input flow is selected, and a longer simulation time is set to make the vehicle transportation capacity on the road reach stability. Observe the results of the simulation. If the traffic capacity of the road is less than the input flow and a long vehicle queue forms at the road entrance, then the traffic capacity is the transportation capacity of the road. Otherwise, increase the vehicle input flow and repeat the above process until the requirements are met.

[0130] Although the present invention has been described above, the present invention is not limited to the above specific embodiments. The above specific embodiments are merely illustrative and not restrictive. Under the inspiration of the present invention, those of ordinary skill in the art can also make many variations without departing from the purpose of the present invention, and these all fall within the protection scope of the present invention.

Claims

1. A simulation system for predicting road transport capacity within a rockfill dam based on vehicle speed, the simulation system comprising a vehicle speed monitoring module, a vehicle speed data set, a vehicle speed spatiotemporal prediction module and a road transport evaluation module; characterized in that: The vehicle speed spatiotemporal prediction module comprises a first fully connected layer, multiple spatiotemporal fusion graph neural network layers, a second fully connected layer and a third fully connected layer; wherein: each spatiotemporal fusion graph neural network layer is composed of multiple parallel spatiotemporal fusion graph neural units and gated convolution units; the spatiotemporal fusion graph unit is composed of stacked multi-layer graph multiplication blocks and pooling layers; the gated unit is composed of a first extended convolution unit and a second extended convolution unit; wherein: The vehicle speed monitoring module is used to delete data of abnormal tracks of transport vehicles, duplicate data and isolated data to generate first standard vehicle speed data; The vehicle speed data set uses a regional range to divide the first standard vehicle speed data to obtain a vehicle speed training set, a vehicle speed verification set and a vehicle speed test set; The vehicle speed spatiotemporal prediction module is used to establish the time and space correlation of each node of the vehicle speed training set, the vehicle speed verification set and the vehicle speed test set to obtain the second standard vehicle speed data; The road transport evaluation module outputs the road transport capacity according to the second standard vehicle speed data and the road section saturation flow method.

2. The system for simulating the transport capacity of roads within a rockfill dam based on vehicle speed prediction according to claim 1 is characterized by: The vehicle speed monitoring module is used to delete data of abnormal tracks of transport vehicles, duplicate data and isolated data to generate first standard vehicle speed data, including: Traverse the satellite positioning trajectory records in the order of sample collection time, and determine whether the instantaneous speed value of the dump truck is within the set reasonable range. If not, it is regarded as an abnormal satellite positioning trajectory record and deleted; Traverse and determine whether the difference between the longitude and latitude of two adjacent satellite positioning trajectory records is less than a set threshold, identify duplicate satellite positioning trajectory records and delete them; It is determined in turn whether the positioning time difference between the current record row and the adjacent record row is equal to the sampling time interval, and isolated satellite positioning trajectory records are identified and deleted.

3. The system for simulating the transport capacity of roads within a rockfill dam based on vehicle speed prediction according to claim 1 is characterized by: The vehicle speed data set uses a regional range to divide the first standard vehicle speed data to obtain a vehicle speed training set, a vehicle speed verification set, and a vehicle speed test set, including: The first standard vehicle speed data was processed using Z-score standardization to obtain the number of nodes, edges, time intervals, and missing value rate of each data set; The data of the four datasets are split into training set, validation set and test set in a ratio of 6:2:

2.

4. The system for simulating the transport capacity of roads within a rockfill dam based on vehicle speed prediction according to claim 1 is characterized by: The vehicle speed spatiotemporal prediction module establishes the time and space correlation of each node of the vehicle speed training set, the vehicle speed verification set and the vehicle speed test set to obtain the second standard vehicle speed data, including: The road network is represented as a topological graph G = (V, E, A SG ), where: V is the node set |V| = N 2 , corresponding to N sensor observations; E is the set of edges, A SG ∈R N×N is a spatial graph of spatial adjacency matrix, representing the proximity or distance of nodes; According to the observation signal Construct a time graph A of the spatial graph information G at time step t TC ; The spatiotemporal fusion graph neural unit can be obtained from A according to each node SG Get spatial correlation from A TG Get the time correlation from A Tc Obtain autocorrelation from the equation, and calculate the relevant feature data of each node time series according to the following formula: h l+1 =(A * h l W1+b1)⊙σ(A * h l W2+b2) Where: h l is the hidden state of the STFGN module; A * It is the abbreviation of space-time fusion map; A STFG ∈R KN×KN , W1, W2∈R C×C , b1, b2∈R C are the model parameters of GLU respectively; ⊙ represents the Hadamard product, σ represents the sigmoid function; The gated convolution unit obtains the dependent feature data of each node time series according to the following formula: Y=φ(Θ1*X+a)⊙σ(Θ2*X+b) Where: φ(·) and σ(·) are activation functions, Θ1 and Θ2 are two independent 1D convolution operations with a dilation rate of K-1; The second standard vehicle speed data is obtained by training the relevant feature data and dependent feature data of each node time series through the following loss function; Where: δ is a hyperparameter used to control the sensitivity of the squared error loss.

5. The method for simulating the system prediction of road transport capacity within a rockfill dam according to claim 1, characterized in that: The steps include: Deleting data of abnormal trajectories of transport vehicles, deleting duplicate data and deleting isolated data to generate first standard vehicle speed data; The first standard vehicle speed data is divided into a vehicle speed training set, a vehicle speed verification set, and a vehicle speed test set using a regional range with a ratio of 6:2:2; Establishing the time and space correlation of each node of the vehicle speed training set, the vehicle speed verification set and the vehicle speed test set to obtain the second standard vehicle speed data; wherein: The road network is represented as a topological graph G = (V, E, A SG ), where: V is the node set |V| = N 2 , corresponding to N sensor observations; E is the set of edges, A SG ∈R N×N is a spatial graph of spatial adjacency matrix, representing the proximity or distance of nodes; According to the observation signal Construct a time graph A of the spatial graph information G at time step t TC ; The spatiotemporal fusion graph neural unit can be obtained from A according to each node SG Get spatial correlation from A TG Get the time correlation from A Tc Obtain autocorrelation from the equation, and calculate the relevant feature data of each node time series according to the following formula: h l+1 =(A * h l W1+b1)⊙σ(A * h l W2+b2) Where: h l is the hidden state of the STFGN module; A * It is the abbreviation of space-time fusion map; A STFG ∈R KN×KN , W1, W2∈R C×C , b1, b2∈R C are the model parameters of GLU respectively; ⊙ represents the Hadamard product, σ represents the sigmoid function; The gated convolution unit obtains the dependent feature data of each node time series according to the following formula: Y=φ(Θ1*X+a)⊙σ(Θ2*X+b) Where: φ(·) and σ(·) are activation functions, Θ1 and Θ2 are two independent 1D convolution operations with a dilation rate of K-1; The second standard vehicle speed data is obtained by training the relevant feature data and dependent feature data of each node time series through the following loss function; Where: δ is a hyperparameter used to control the sensitivity of the squared error loss; The road transport capacity is output according to the second standard vehicle speed data in accordance with the road section saturation flow method.

Images