CN-121615957-B - Traffic network toughness diagnosis method under flood disasters based on space-time diagram neural network

CN121615957BCN 121615957 BCN121615957 BCN 121615957BCN-121615957-B

Abstract

The invention discloses a traffic network toughness diagnosis method under flood disasters based on a space-time diagram neural network, and relates to the field of intersection of traffic engineering and artificial intelligence. The method comprises the steps of collecting traffic topology, flood monitoring and traffic flow data and carrying out space-time alignment, constructing a dynamic time-space diagram of flood coupling, introducing a water depth-traffic capacity response mechanism, dynamically updating the side weight of a diagram structure according to real-time water depth by utilizing an attenuation function meeting physical monotonicity constraint, realizing real-time mapping of a disaster physical state to a network topology, inputting the dynamic diagram into a pre-trained time-space diagram neural network model, extracting space-time evolution characteristics and outputting toughness diagnosis results, and carrying out supervised learning by adopting toughness labels generated based on a counter-facts baseline in model training and introducing a physical constraint loss function. The method solves the problems of decoupling of disaster characteristics and graph structures and unavailable toughness labels in the prior art, and improves the physical consistency and accuracy of diagnosis.

Inventors

SU XIN
XUAN SHENGWEI
HU JIANWEN
ZHANG YE
CHEN ZHAOYI
WANG SHUANG
LI XINYANG
WANG LEIZHI
LIN FANGCHAO
ZHOU QINGPEI
YANG CHENG
WANG LILI
LI LINGJIE
LIU YONG
XU ZHIYANG

Assignees

水利部交通运输部国家能源局南京水利科学研究院

Dates

Publication Date: 20260508
Application Date: 20260202

Claims (7)

1. A traffic network toughness diagnosis method under flood disasters based on a space-time diagram neural network is characterized by comprising the following steps: collecting and acquiring topology data, flood dynamic monitoring data and traffic flow operation data of a traffic network; Preprocessing traffic network topology data, flood dynamic monitoring data and traffic flow operation data and performing multi-source space-time alignment to obtain space-time sequence data under uniform space-time granularity; constructing a space-time diagram structure reflecting the state of the traffic network, wherein the space-time diagram structure comprises a node set, an edge set and a weight matrix describing the edge connection relation, and mapping space-time sequence data into the space-time diagram structure; Inputting the space-time diagram structure after data mapping into a space-time diagram neural network model which is trained in advance, and extracting space-time evolution characteristics of a traffic network; calculating and outputting a toughness diagnosis result of the traffic network based on the space-time evolution characteristics; Constructing a space-time diagram structure reflecting the state of a traffic network, comprising: Constructing a node set by taking an intersection of a traffic network as a node, and constructing an edge set by taking a physical road section connected with the intersection as an edge; establishing an adjacency weight matrix for describing the connection relation between each edge in the edge set; the space-time diagram structure consists of a node set, an edge set and an adjacent weight matrix; Associating the time-space sequence data as attribute characteristics of a node set or an edge set to finish data mapping; the method for establishing the adjacency weight matrix specifically comprises the following steps of: acquiring the number of lanes and the design speed per hour of each road section in the edge set; Calculating the reference traffic capacity of each road section based on the number of lanes and the design speed per hour, and taking the reference traffic capacity as a static weight value of a corresponding side in the adjacent weight matrix; The adjacency weight matrix is also configured as a dynamic matrix changing with time steps, and the dynamic establishment of the adjacency weight matrix specifically comprises: Introducing a water depth-traffic capacity response mechanism, and determining the real-time water depth of each road section at each time step based on flood dynamic monitoring data; Calculating a traffic capacity attenuation coefficient corresponding to the real-time water depth by utilizing a water depth-traffic capacity response mechanism, and obtaining real-time effective traffic capacity by combining the reference traffic capacity of the road section; according to the real-time effective traffic capacity, dynamically updating an adjacent weight matrix of the space-time diagram structure at each time step; The time space graph neural network model is obtained through supervised learning pre-training, a supervision label used for training is built through a counter facts toughness label generation method, the counter facts toughness label generation method comprises the steps of selecting traffic operation data of a historical non-flood period, building a counter facts base line representing traffic network operation rules in a non-disaster state, obtaining traffic actual measurement data of the flood disaster period, calculating the functional attenuation degree of the traffic actual measurement data relative to the counter facts base line, generating toughness labels of a traffic network based on the accumulated effect of the functional attenuation degree in disaster duration, calculating toughness labels Re of road sections e through the counter facts toughness label generation method, wherein a training loss function of the time space graph neural network model comprises physical constraint items and is used for constraining a water depth-traffic capacity response function phi (·) to accord with the physical rules.
2. The method according to claim 1, characterized in that it comprises: the traffic network topology data comprises node space coordinates and road section physical attributes; the flood dynamic monitoring data comprise rainfall, water level elevation and flooding range; the traffic flow operation data comprises section flow, average speed and passing duration.
3. The method of claim 1, wherein preprocessing and multi-source spatiotemporal alignment of traffic network topology data, flood dynamic monitoring data, and traffic flow operational data comprises: calculating and collecting statistical distribution characteristics of traffic network topology data, flood dynamic monitoring data and traffic flow operation data, and identifying and eliminating abnormal data distributed outside a preset interval by using a 3 sigma criterion; carrying out normalization processing on the data from which the abnormal data are removed, and eliminating dimension differences of different data sources; and interpolating the missing values in the data sequence by using a K nearest neighbor algorithm, and unifying the multi-source data to the same time sampling granularity to obtain the space-time sequence data.
4. The method of claim 1, wherein preprocessing and multi-source spatiotemporal alignment of traffic network topology data, flood dynamic monitoring data, and traffic flow operational data comprises: Mapping the dynamic monitoring data of the floods in the form of grids onto a road section central line based on the space coordinates of the traffic network topology data to obtain a flood state sequence in the road section dimension; resampling the flood state sequence and the traffic flow operation data to unify to a preset time granularity; And identifying the missing time step in the monitoring data, generating a corresponding missing test mask matrix, and combining the aligned multi-source data and the missing test mask matrix into space-time sequence data.
5. Method according to claim 1, characterized in that the effective traffic capacity in real time is calculated using a water depth-traffic capacity response mechanism, in particular based on the following formula: C e (t)=C 0,e *φ(h e (t)); wherein, C e (t) represents the real-time effective traffic capacity of the road section e at the time step t, C 0,e represents the reference traffic capacity of the road section e, h e (t) represents the real-time water depth of the road section e at the time step t in cm, and phi (DEG) represents the water depth-traffic capacity response function; the calculation logic of the water depth-traffic capacity response function phi (h e (t)) satisfies the following: When h e (t)≥h crit , Φ (h e (t))=0; When h e (t)<h crit , φ (h e (t))ε (0, 1), and monotonically decreasing as h e (t) increases; wherein h crit represents a critical water depth threshold for the road section to allow traffic.
6. The method of claim 1, wherein the space-time diagram neural network model comprises a diagram convolutional layer, a time convolutional layer, and a fully-connected layer connected in series; The graph convolution layer is used for extracting space correlation features from features of neighbor nodes in the aggregated space-time graph structure; The time convolution layer is used for extracting a dynamic evolution rule of the traffic state along a time dimension; the full connection layer is used for mapping the extracted features to toughness diagnosis results.
7. The method of claim 1, further comprising, after outputting the toughness diagnostic result of the traffic network: Setting a toughness judgment threshold R th , traversing each evaluation unit in a traffic network, and screening out evaluation units with toughness diagnosis values R i meeting R i <R th as weak links; Calculating the toughness recovery urgency P i of the weak link, wherein the calculation formula is P i =R th -R i ; the weak links are ordered according to the sequence from the big to the small of the toughness recovery urgency degree P i , and a priority recovery list is generated; Wherein i represents the index of the evaluation unit, and R i represents the toughness diagnostic value of the i-th evaluation unit.

Description

Traffic network toughness diagnosis method under flood disasters based on space-time diagram neural network Technical Field The invention belongs to the crossing field of traffic engineering and artificial intelligence, in particular to a traffic network toughness diagnosis method under flood disasters based on a space-time diagram neural network. Background Flood disasters are natural disasters which are high in global climate change background, systematic damage is often caused to an urban traffic network in the forms of storm ponding, river overflow and the like, physical damages such as road section inundation, bridge damage and the like are caused, traffic jams and traffic interruption are caused, emergency rescue and resident evacuation are hindered, and social and economic orders are threatened. The traffic network toughness refers to the comprehensive capacity of the traffic network for resisting damage, adapting to risks and quickly recovering basic traffic functions under flood impact, and the accurate diagnosis is a precondition of making an emergency strategy and optimizing post-disaster reconstruction priority by traffic management departments, and has a key value for improving the urban disaster resistance toughness. The traditional traffic network toughness diagnosis relies on an empirical model and static statistical analysis, and has the limitations that complex spatial association of traffic network node-side-topology cannot be described only through single index evaluation, time characteristics of traffic flow fluctuation and disaster intensity change are difficult to reflect, the traditional traffic network toughness diagnosis relies on historical data offline analysis, dynamic disaster data such as real-time rainfall, water level, road section ponding and the like are underutilized, diagnosis results are lagged, immediate decisions in disaster cannot be supported, and model assumption is disjointed from actual disaster working conditions, so that the diagnosis accuracy and reliability are difficult to meet the requirement of refined management. Artificial intelligence technology development has driven graph neural networks to be widely used in complex network analysis, with the capability of adapting to traffic network characteristics for processing non-euclidean structural data. The space-time diagram neural network is combined with the time sequence modeling of the map convolution network, so that the space-time diagram neural network can accurately extract the space association characteristics of traffic nodes and road sections, dynamically track the time evolution rule of traffic flow and traffic efficiency under the influence of flooding, effectively make up the shortboard of the traditional method and provide a brand new technical path for real-time accurate diagnosis of the toughness of the traffic network. Disclosure of Invention The invention aims to provide a traffic network toughness diagnosis method under flood disasters based on a space-time diagram neural network, which aims to solve one of the problems existing in the prior art. The technical scheme is that the traffic network toughness diagnosis method under the flood disaster based on the space-time diagram neural network comprises the following steps: collecting and acquiring topology data, flood dynamic monitoring data and traffic flow operation data of a traffic network; Preprocessing traffic network topology data, flood dynamic monitoring data and traffic flow operation data and performing multi-source space-time alignment to obtain space-time sequence data under uniform space-time granularity; constructing a space-time diagram structure reflecting the state of the traffic network, wherein the space-time diagram structure comprises a node set, an edge set and a weight matrix describing the edge connection relation, and mapping space-time sequence data into the space-time diagram structure; Inputting the space-time diagram structure after data mapping into a space-time diagram neural network model which is trained in advance, and extracting space-time evolution characteristics of a traffic network; and calculating and outputting a toughness diagnosis result of the traffic network based on the space-time evolution characteristics. The method has the beneficial effects that the problems of decoupling of disaster characteristics and graph structures and unavailable toughness labels in the prior art are solved, and the physical consistency and accuracy of diagnosis are improved. Drawings FIG. 1 is a diagram of an overall logic architecture of a traffic network toughness diagnostic method in an embodiment of the present application. Fig. 2 is a flowchart of steps for constructing a space-time diagram structure reflecting the state of a traffic network in an embodiment of the present application. FIG. 3 is a flowchart illustrating steps for establishing an adjacency weight matrix according to an embodiment of the present a