Bridge group service behavior space-time correlation prediction method based on graph selective state space model

Abstract

The invention provides a bridge group service performance space-time correlation prediction method based on a graph selectivity state space model. According to the method, for a regional bridge group service behavior prediction task, a service behavior data set of bridge group environmental action and traffic flow is constructed, a graph selectivity state space model for service behavior time-space association prediction is established, a time-space graph convolution module is designed to learn time-space association, and the graph selectivity state space module is constructed to learn time-space evolution characteristics. And training the bridge group service behavior space-time correlation prediction model based on gradient descent error back propagation, and after the training is completed, performing space-time correlation prediction on the bridge group environment effect and the traffic flow by adopting the trained model. Through spatial-temporal characteristic decomposition and fusion of service performance data with different complexities, the processing requirements of service performance spatial-temporal data of multi-type, multi-scale and multi-granularity bridge groups are better adapted, and the universality and prediction efficiency of the method are improved.

Description

Technical Field

[0001] The present invention relates to the fields of bridge engineering, structural health monitoring, spatiotemporal data analysis, and deep learning technology, and in particular to a spatiotemporal correlation prediction method for service performance of a bridge group based on a graph selective state space model. The method can be directly applied to the fields of intelligent bridges, intelligent bridge operation and maintenance, intelligent infrastructure, and intelligent disaster prevention and mitigation. Background Art

[0002] As a key component of the urban transportation system, the stability of the service performance of bridges is directly related to the smoothness and safety of urban transportation. As a complex system composed of multiple bridges, bridge groups will inevitably be affected by the coupling of multiple factors such as the natural environment, traffic loads, and material aging during long-term service. The structural performance will continue to degrade, and even extreme situations may occur in which regional traffic safety accidents are caused by the failure of a single bridge. Therefore, accurate and efficient prediction of the service performance of regional bridge groups is of great significance to ensuring urban traffic safety and improving the efficiency of bridge group operation and maintenance management.

[0003] Traditional methods for predicting the service performance of bridge groups mainly rely on the prediction of the service performance of individual bridges. The service performance they focus on is limited to the local components of individual bridges, and they fail to integrate and utilize the shared characteristics of the environment and loads in the urban area where the bridge group is located. They ignore the spatiotemporal correlation between different individual bridges in the bridge group and the individual similarities in terms of structural type, service life, damage evolution, etc., resulting in the correlation and synergy between the individual bridges in the urban area bridge group not being fully explored.

[0004] At present, deep learning technology has achieved rapid development, and domestic and foreign research has been carried out on the use of deep learning methods to predict the service performance of bridges. However, current research mainly focuses on:

[0005] (1) Deep learning-based assessment of single bridge components: For example, the damage location and severity assessment of cable-stayed bridges based on message passing neural networks uses the cable force data of multiple cables as input and outputs the location and cross-sectional area of ​​the damaged cables, which only considers the spatial correlation of different cables on a single bridge.

[0006] (2) Road network performance prediction based on deep learning: For example, the road deterioration modeling based on GraphSAGE takes into account the spatial correlation and functional classification correlation of the road network. However, the actual prediction of road deterioration is often affected by both spatial and temporal factors. Spatial correlation is reflected in the mutual influence between adjacent road sections, while temporal correlation is reflected in the trend of road condition changes over time. The above studies also only consider spatial correlation but lack consideration of spatiotemporal correlation.

[0007] (3) In addition, existing studies often focus on a single type of data, and a general deep learning method for spatiotemporal correlation prediction of multi-type, multi-scale, and multi-granular spatiotemporal data of bridge service performance (including environment, load, traffic flow, etc.) has not yet been established.

[0008] In recent years, spatiotemporal graph structures and state-space models have provided new ideas and advanced algorithms for studying spatiotemporal correlation prediction. Spatiotemporal graph neural networks have shown broad application prospects in predicting the service performance of bridge groups. However, current research still faces huge challenges. The problems that need to be solved urgently include: how to construct a suitable spatiotemporal graph structure to accurately reflect the complex spatiotemporal correlation between bridge groups, how to design an effective spatiotemporal graph neural network model to capture the spatiotemporal dependency of the evolution of the service performance of bridge groups, and how to make full use of the spatiotemporal data of bridge groups to improve prediction accuracy and efficiency. Summary of the invention

[0009] The purpose of the present invention is to solve the problems in the prior art and propose a spatiotemporal correlation prediction method for the service performance of a bridge group based on a graph selective state space model.

[0010] The present invention is implemented by the following technical scheme. The present invention proposes a spatiotemporal correlation prediction method for service performance of a bridge group based on a graph selective state space model. The method comprises the following steps:

[0011] Step 1: Determine the bridge group task object to be studied and establish a bridge group graph structure model that takes into account the effects of urban area environment and traffic flow;

[0012] Step 2: Design a spatiotemporal graph prediction task based on the multi-source perception data of the service performance of bridge groups, and construct the corresponding input and output data sets; design the bridge group service performance prediction task to take the temperature, humidity, wind speed, wind direction and corrosion ion concentration environmental action variables and historical monitoring data of traffic flow as input, and predict the future bridge group environmental action and traffic flow as output; embed the multi-type, multi-scale and multi-granularity bridge group service performance multi-source perception data set into Where N is the number of bridges, T is the input data time step, and D is the feature dimension; the corresponding output dataset embedding is expressed as Where T p is the predicted output data time step;

[0013] Step 3: Establish a graph-selective state-space model for spatiotemporal correlation prediction of bridge group service performance; the graph-selective state-space model for spatiotemporal correlation prediction of bridge group service performance includes two modules: spatiotemporal graph convolution module and graph-selective state-space module. The spatiotemporal graph convolution module is used to learn the spatiotemporal correlation of bridge group service performance, and the graph-selective state-space module is used to learn the spatiotemporal evolution characteristics of bridge group service performance. The outputs of the two modules are weighted fused and then output through the fully connected layer to obtain the final prediction result.

[0014] Step 4: Based on the gradient descent error back propagation training graph selective state space model, the trained model is used to predict the spatiotemporal correlation between the environmental effects of bridge groups and traffic flow.

[0015] Furthermore, in step 1, the graph structure model is established as follows:

[0016] Node definition of bridge group graph structure: each bridge is regarded as a node in the graph, and the node attributes include temperature, humidity, wind speed, wind direction, corrosion ion concentration environmental variables and traffic flow;

[0017] Edge definition of bridge group graph structure: Based on the spatial position relationship between different single bridges in the bridge group, the edges between nodes are preliminarily defined; for temperature, humidity, and corrosion ion prediction, the edges are undirected edges; for wind speed and wind direction prediction, the edges are directed edges; for traffic flow prediction, the edges are directed edges, and the weight of the edges is determined according to the traffic flow size and the distance between bridges;

[0018] Generation of bridge group graph structure: The bridge group graph structure is represented by G = (V; E), where V is the set of bridge nodes and E is the set of edges; based on the definition of nodes and edges, a bridge group graph structure model is generated to determine the adjacency matrix A of the bridge group graph structure.

[0019] Furthermore, in step 2, for the spatiotemporal data of the service performance of multi-type, multi-scale and multi-granular bridge groups, three different time granularities are used as input data, including recent data X r , Periodic data X c and trend data X q ; Recent data is a historical time series directly adjacent to the forecast period, periodic data is composed of segments whose past days in the same time period as the forecast period are integer multiples of α, and trend data is composed of segments whose past days in the same time period as the forecast period are integer multiples of β; the lengths of the three types of input data, recent data, periodic data and trend data, are set according to actual forecast needs to adapt to different forecast objectives and historical data characteristics.

[0020] Furthermore, the step three is specifically as follows:

[0021] Step 31: Establish a spatiotemporal graph convolution module;

[0022] The environmental effects and traffic flow data of the bridge group are passed to the gated temporal convolution layer and the self-attention diffusion graph convolution layer in sequence after 1×1 convolution, and the output is obtained after residual connection with the input and batch normalization.

[0023] Step 32: Establish a graph-selective state space module;

[0024] The recent data on the environmental effects of bridge groups and traffic flow are used to r After layer normalization, they are passed to m dynamic filter graph convolution layers respectively. The outputs of the m graph convolution layers are concatenated and passed through a linear layer. The output of the linear layer is divided into a core sequence and a secondary sequence. The core sequence is passed to the graph state space selection mechanism after one-dimensional convolution and SiLU activation. Its output is multiplied element by element with the secondary sequence activated by SiLU. Finally, it passes through a linear layer and is multiplied with the layer normalized X r Residual connection to get the final output;

[0025] Step 3: Combine the three time-granularity data X of bridge group environmental effects and traffic flow r , X c and X q The fusion result output by N spatiotemporal graph convolution modules is connected to the input X r The outputs of the M graph-selective state-space modules are weighted and summed, and the results are passed through the fully connected layer to obtain the final output result.

[0026] Furthermore, in step 31,

[0027] Construct a gated temporal convolutional layer to learn the complex temporal dependencies of the bridge group service performance sequence data: pass the bridge group environmental effects and traffic flow data input to the gated temporal convolutional layer after passing it through a 1×1 convolution:

[0028] Y=f(w1*X+b)⊙σ(w2*X+c) (4)

[0029] Where Y is the output of the bridge group environmental effect and traffic flow data input after gated time convolution, w1, w2, b and c are the parameters to be optimized; f is the tanh activation function, which transforms and extracts the features of the bridge group environmental effect and traffic flow data input; σ is the sigmoid activation function, which outputs a gating weight between 0 and 1, and the generated gating weight controls the proportion of information flow; ⊙ is element-by-element multiplication, w1*X+b represents convolution layer 1, and w2*X+c represents convolution layer 2;

[0030] Construct a self-attention diffusion graph convolution layer for learning complex spatial dependencies of service state sequence data of bridge groups: Pass the output of the gated temporal convolution layer to the self-attention diffusion graph convolution layer:

[0031]

[0032] Where H is the output result of the bridge group environmental effect and traffic flow data input after gated temporal convolution and self-attention diffusion graph convolution in sequence. and is the parameter to be optimized, and is the attention weight matrix of the bridge group graph structure, K is the number of diffusion steps, which is a hyperparameter and represents the diffusion process of K finite steps to simulate the service performance graph signal of the bridge group.

[0033] Furthermore, the attention weight matrix of the self-attention diffusion graph convolution layer of the spatiotemporal graph convolution module is calculated as follows:

[0034]

[0035] In the formula, a k is the learnable parameter matrix to be optimized; the propagation process of bridge group environmental effect and traffic flow data signal is divided into two directions, and the forward transformation matrix of bridge group environmental effect and traffic flow data signal is A f =A / rowsum(A), the inverse transformation matrix is ​​A b =A T / rowsum(A T ), Adaptive adjacency matrix of bridge group graph structure A k Represents the power level of the transformation matrix and the adaptive adjacency matrix. The columns are normalized during the operation and expressed as (·) / rowsum(·); the adaptive adjacency matrix The calculation of does not require prior knowledge, and end-to-end learning is performed in the gradient descent process to discover the hidden spatial dependencies between the environmental effects of bridge groups and traffic flow data. The learnable parameters to be optimized are U, are two randomly initialized node embeddings; SoftMax represents the exponential normalization operation, and ReLU represents the rectified linear unit activation function.

[0036] Furthermore, in step 33, the output of each spatiotemporal graph convolution block is used as the input of the next spatiotemporal graph convolution block, and a total of N spatiotemporal graph convolution blocks are set; a residual connection is set in each spatiotemporal graph convolution block, and a jump connection is performed with the output layer, specifically: the result of the gated time convolution in each spatiotemporal graph convolution block is processed by a 1×1 convolution, and then added to the result of the gated time convolution in the previous spatiotemporal graph convolution block by a jump connection; finally, the accumulation is passed to the output layer, which is composed of an activation function and a linear layer, and outputs the operation result of the spatiotemporal graph convolution block.

[0037] Furthermore, in step 33, the spatiotemporal graph convolution module outputs of the three time granularities of the bridge group service performance are fused. The specific method is as follows:

[0038]

[0039] In the formula, w r 、w c With w qis a learnable parameter to be optimized, is a scalar, and reflects the influence of the three time granularity components on the spatiotemporal correlation prediction target of the service performance of bridge groups; and They represent the output of the spatiotemporal graph data of the bridge group service state after the spatiotemporal graph convolution block, Represents the fused spatiotemporal graph convolution module output.

[0040] The present invention also proposes an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the method for predicting the spatiotemporal correlation of the service performance of a group of bridges based on a graph selective state space model are implemented.

[0041] The present invention also proposes a computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the steps of the method for predicting the spatiotemporal correlation of the service performance of a group of bridges based on a graph-selective state-space model.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] (1) By embedding perception data of different time granularities (recent data, periodic data, and trend data), the model can capture the dynamic changes of multi-type, multi-scale, and multi-granular bridge service performance data at different time scales, thereby improving the versatility, accuracy, and robustness of the model.

[0044] (2) Design a spatiotemporal graph convolution module that combines graph convolution with a time series prediction network to effectively capture the spatial and temporal dependencies of the service performance data of a group of bridges and achieve efficient spatiotemporal correlation modeling.

[0045] (3) An adaptive adjacency matrix of the bridge group graph structure is introduced into the graph convolution module. This matrix is ​​obtained through end-to-end learning during the training process without any prior knowledge. This enables the model to automatically discover the hidden spatial dependencies in the service performance data of the bridge group, further improving the accuracy of the prediction.

[0046] (4) The attention weight matrix of the bridge group graph structure is introduced into the graph convolution module to determine the contribution of each part of information to the representation of the current node, thereby improving the interpretability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0048] Figure 1 This is a flow chart of the spatiotemporal correlation prediction method for the service performance of a group of bridges based on a graph-selective state-space model.

[0049] Figure 2 Schematic diagram of the bridge group structure model considering environmental effects and traffic flow.

[0050] Figure 3 It is a time series relationship diagram of recent data, periodic data and trend data sampling of the service performance of bridge groups.

[0051] Figure 4 Schematic diagram of the spatiotemporal graph convolution module architecture in the spatiotemporal correlation prediction method for the service performance of a group of bridges.

[0052] Figure 5 Schematic diagram of the gated temporal convolutional layer architecture in the spatiotemporal correlation prediction method for the service performance of a group of bridges.

[0053] Figure 6 Schematic diagram of the self-attention diffusion graph convolutional layer architecture in the spatiotemporal correlation prediction method for the service performance of a group of bridges.

[0054] Figure 7 Schematic diagram of the attention weight matrix calculation process in the spatiotemporal correlation prediction method for the service performance of a group of bridges. DETAILED DESCRIPTION

[0055] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] Combination Figure 1-Figure 7 The present invention proposes a spatiotemporal correlation prediction method for service performance of a bridge group based on a graph selective state space model, the method comprising the following steps:

[0057] Step 1: Determine the bridge group task object to be studied and establish a bridge group graph structure model that takes into account the effects of urban area environment and traffic flow;

[0058] In step 1, the multi-source perception data of the service performance of the bridge group is collected and preprocessed; the multi-source perception data of the service performance of the bridge group is collected: the historical data of environmental effects such as temperature, humidity, wind speed, wind direction, and corrosion ion concentration of each bridge in the bridge group and traffic flow are collected. The multi-source perception data of the service performance of the bridge group is preprocessed: the collected data is cleaned, outliers and missing values ​​are removed, and multi-type data is normalized to ensure the consistency and accuracy of the data.

[0059] In step 1, the graph structure model is established as follows:

[0060] Node definition of bridge group graph structure: each bridge is regarded as a node in the graph, and the node attributes include temperature, humidity, wind speed, wind direction, corrosion ion concentration environmental variables and traffic flow;

[0061] Edge definition of bridge group graph structure: Based on the spatial position relationship between different single bridges in the bridge group, the edges between nodes are preliminarily defined; for temperature, humidity, and corrosion ion prediction, the edges are undirected edges; for wind speed and wind direction prediction, considering that wind direction has a significant impact on wind speed, the edges are directed edges; for traffic flow prediction, considering that the direction of traffic flow has an important impact on the prediction results, the edges are directed edges (if you only focus on the overall flow between bridge groups and do not care about the specific direction of flow, you can set it as an undirected edge), and the weight of the edge is determined according to the traffic flow size and the distance between bridges;

[0062] Generation of bridge group graph structure: The bridge group graph structure is represented by G = (V; E), where V is the set of bridge nodes and E is the set of edges; based on the definition of nodes and edges, a bridge group graph structure model is generated to determine the adjacency matrix A of the bridge group graph structure.

[0063] The schematic diagram of the bridge group structure model considering environmental effects and traffic flow is as follows: Figure 2 Each bridge in a certain urban area (including different types of cable-stayed bridges, suspension bridges, beam bridges, arch bridges, etc.) is modeled as a node v in the graph, and the roads connecting the bridges are modeled as the edges E between the two nodes, forming a regional bridge group graph structure, where node v4 is a cable-stayed bridge, v6 is a beam bridge, v8 is a suspension bridge, and v 10 It is an arch bridge.

[0064] Step 2: Design a spatiotemporal graph prediction task based on the multi-source perception data of the service performance of bridge groups, and construct the corresponding input and output data sets; design the bridge group service performance prediction task to take the temperature, humidity, wind speed, wind direction and corrosion ion concentration environmental action variables and historical monitoring data of traffic flow as input, and predict the future bridge group environmental action and traffic flow as output; embed the multi-type, multi-scale and multi-granularity bridge group service performance multi-source perception data set into Where N is the number of bridges, T is the time step of the input data, and D is the feature dimension (including the total number of environmental variables such as temperature, humidity, wind speed, wind direction, and corrosion ion concentration studied, and the total number of traffic flow types); the corresponding output data set embedding is expressed as Where T p is the predicted output data time step;

[0065] In step 2, for the spatiotemporal data of the service performance of multi-type, multi-scale and multi-granular bridge groups, three different time granularities are used as input data, including recent data X r , Periodic data X c and trend data X q ; Recent data is a historical time series directly adjacent to the forecast period, periodic data is composed of segments whose past days in the same time period as the forecast period are integer multiples of α, and trend data is composed of segments whose past days in the same time period as the forecast period are integer multiples of β; where the period window is expressed by variables α and β, respectively, and the possible values ​​are: 1 for 1 day, 7 for 1 week, 30 for 1 month, 365 for 1 year, etc. Similar settings will not be stated one by one in the future; the length of the three types of input data, recent data, periodic data and trend data (i.e., the number of time steps T contained in each data segment) is set according to actual forecasting needs to adapt to different forecasting objectives and historical data characteristics.

[0066] Recent data on service performance of bridge groups X r , Periodic data X c and trend data X q The embedding representation of is as follows:

[0067] (1) Recent data:

[0068]

[0069] (2) Periodic data:

[0070]

[0071] (3) Trend data:

[0072]

[0073] Where f s is the sampling frequency of a certain type of sensory data analyzed within a day, t0 is the current time, and the prediction window size is T p ; The intercept length along the time axis is T r , T c and T q The three time series segments are respectively used as recent data, periodic data and trend data, where T r , T c and T q All T p An integer multiple of .

[0074] The relationship between the sampling time series of recent data, periodic data and trend data of the service performance of bridge groups is as follows: Figure 3 shown. Figure 3In the example, the sampling frequency of a service performance perception variable (such as temperature) of a bridge group is 48 times a day, and the sampling interval is 30 minutes. p is a time step, T r , T c and T q All T p twice as much.

[0075] Step 3: Establish a graph-selective state-space model for spatiotemporal correlation prediction of bridge group service performance; the graph-selective state-space model for spatiotemporal correlation prediction of bridge group service performance includes two modules: spatiotemporal graph convolution module and graph-selective state-space module. The spatiotemporal graph convolution module is used to learn the spatiotemporal correlation of bridge group service performance, and the graph-selective state-space module is used to learn the spatiotemporal evolution characteristics of bridge group service performance. The outputs of the two modules are weighted fused and then output through the fully connected layer to obtain the final prediction result.

[0076] The step three is specifically as follows:

[0077] Step 31: Establish a spatiotemporal graph convolution module;

[0078] The environmental effects and traffic flow data of the bridge group are passed to the gated temporal convolution layer and the self-attention diffusion graph convolution layer in sequence after 1×1 convolution, and the output is obtained after residual connection with the input and batch normalization.

[0079] In step three one,

[0080] Construct a gated temporal convolutional layer to learn the complex temporal dependencies of the bridge group service performance sequence data: pass the bridge group environmental effects and traffic flow data input to the gated temporal convolutional layer after passing it through a 1×1 convolution:

[0081] Y=f(w1*X+b)⊙σ(w2*X+c) (4)

[0082] Where Y is the output of the bridge group environmental effect and traffic flow data input after gated time convolution, w1, w2, b and c are the parameters to be optimized; f is the tanh activation function, which transforms and extracts the features of the bridge group environmental effect and traffic flow data input; σ is the sigmoid activation function, which outputs a gating weight between 0 and 1, and the generated gating weight controls the proportion of information flow; ⊙ is element-by-element multiplication, w1*X+b represents convolution layer 1, and w2*X+c represents convolution layer 2;

[0083] Construct a self-attention diffusion graph convolution layer for learning complex spatial dependencies of service state sequence data of bridge groups: Pass the output of the gated temporal convolution layer to the self-attention diffusion graph convolution layer:

[0084]

[0085] Where H is the output result of the bridge group environmental effect and traffic flow data input after gated temporal convolution and self-attention diffusion graph convolution in sequence. and is the parameter to be optimized, and is the attention weight matrix of the bridge group graph structure, K is the number of diffusion steps, which is a hyperparameter and represents the diffusion process of K finite steps to simulate the service performance graph signal of the bridge group.

[0086] The attention weight matrix of the self-attention diffusion graph convolution layer of the spatiotemporal graph convolution module is calculated as follows:

[0087]

[0088] In the formula, a k is the learnable parameter matrix to be optimized; the propagation process of bridge group environmental effect and traffic flow data signal is divided into two directions, and the forward transformation matrix of bridge group environmental effect and traffic flow data signal is A f =A / rowsum(A), the inverse transformation matrix is ​​A b =A T / rowsum(A T ), Adaptive adjacency matrix of bridge group graph structure A k Represents the power level of the transformation matrix and the adaptive adjacency matrix. The columns are normalized during the operation and expressed as (·) / rowsum(·); the adaptive adjacency matrix The calculation of does not require prior knowledge, and end-to-end learning is performed in the gradient descent process to discover the hidden spatial dependencies between the environmental effects of bridge groups and traffic flow data. The learnable parameters to be optimized are U, are two randomly initialized node embeddings; SoftMax represents the exponential normalization operation, and ReLU represents the rectified linear unit activation function.

[0089] Step 32: Establish a graph-selective state space module;

[0090] The recent data on the environmental effects of bridge groups and traffic flow are used to r After layer normalization, they are passed to m dynamic filter graph convolution layers respectively. The outputs of the m graph convolution layers are concatenated and passed through a linear layer. The output of the linear layer is divided into a core sequence and a secondary sequence. The core sequence is passed to the graph state space selection mechanism after one-dimensional convolution and SiLU activation. Its output is multiplied element by element with the secondary sequence activated by SiLU. Finally, it passes through a linear layer and is multiplied with the layer normalized X r Residual connection to get the final output;

[0091] Step 3: Combine the three time-granularity data X of bridge group environmental effects and traffic flow r , X c and X q The fusion result output by N spatiotemporal graph convolution modules is connected to the input X r The outputs of the M graph-selective state-space modules are weighted and summed (the sum of the weight coefficients of the two modules is equal to 1), and the result is passed through the fully connected layer to obtain the final output result.

[0092] In step 33, the output of each spatiotemporal graph convolution block is used as the input of the next spatiotemporal graph convolution block, and a total of N spatiotemporal graph convolution blocks are set; a residual connection is set in each spatiotemporal graph convolution block, and a jump connection is made with the output layer, specifically: the result of the gated time convolution in each spatiotemporal graph convolution block is processed by a 1×1 convolution, and then added to the result of the gated time convolution in the previous spatiotemporal graph convolution block by a jump connection; finally, the accumulation is passed to the output layer, which consists of an activation function and a linear layer, and outputs the operation result of the spatiotemporal graph convolution block.

[0093] In step 3, the outputs of the spatiotemporal graph convolution modules of the three time granularities of the service performance of the bridge group are fused. The specific method is as follows:

[0094]

[0095] In the formula, w r 、w c With w q is a learnable parameter to be optimized, is a scalar, and reflects the influence of the three time granularity components on the spatiotemporal correlation prediction target of the service performance of bridge groups; and They represent the output of the spatiotemporal graph data of the bridge group service state after the spatiotemporal graph convolution block, Represents the output of the fused spatiotemporal graph convolution module.

[0096] Step 4: Based on the gradient descent error back propagation training graph selective state space model, the trained model is used to predict the spatiotemporal correlation of bridge group environmental effects and traffic flow. A spatiotemporal correlation prediction system for bridge group service performance is established.

[0097] Based on the multi-source perception data of the service performance of bridge groups, including environmental effects (temperature, humidity, wind speed, wind direction, corrosive ion concentration, etc.) and traffic flow, a graph selective state space model for spatiotemporal correlation prediction of the service performance of bridge groups is trained, and the trained model is used to predict the environmental effects and traffic flow of future bridge groups.

[0098] In step 4, the graph selective state space model for spatiotemporal correlation prediction of bridge group service performance is trained by using a gradient descent-based optimization algorithm (including optimizers such as AdamW and SGD) so that the generated predicted values ​​of bridge group service performance data are as close as possible to the true values ​​of the actual perception data. The graph selective state space model for spatiotemporal correlation prediction is trained in a supervised learning manner, and the mean square error (MSE) is selected as the training objective function:

[0099]

[0100] In the formula, is the service state of the bridge group predicted by the model in the future k steps, X (t+1):(t+k) is the real service performance perception data of the bridge group in the future k steps; N is the number of bridges, k is the time step, D is the feature dimension, including environmental variables such as temperature, humidity, wind speed, wind direction, corrosion ion concentration and traffic flow; B represents the batch size, λ is the regularization coefficient, and R(Θ) is the regularization term.

[0101] For the task of predicting the service performance of regional bridge groups, a service performance dataset of environmental effects (temperature, humidity, wind speed, wind direction, corrosive ion concentration, etc.) and traffic flow of bridge groups is constructed according to the above method. A spatiotemporal correlation prediction model of the service performance of bridge groups is trained based on gradient descent error back propagation. After the training is completed, the trained model is used to predict the spatiotemporal correlation of environmental effects and traffic flow of bridge groups, and a spatiotemporal correlation prediction system for the service performance of bridge groups is established.

[0102] This paper proposes a spatiotemporal correlation prediction method for the service performance of bridge groups based on a graph selective state space model. By capturing the complex spatiotemporal correlation and spatiotemporal evolution laws inherent in the service performance data of bridge groups, it aims to improve the prediction accuracy and versatility of the service performance of bridge groups. The innovation lies in:

[0103] (1) Using the spatiotemporal graph structure, we built a spatiotemporal graph fusion learning module, combined with graph neural networks and deep learning of time series data, and considered the data characteristics of the service performance of bridge groups in both spatial and temporal dimensions to more accurately model and predict the dynamic spatiotemporal evolution of bridge group systems;

[0104] (2) Adopting the state space model, constructing a structured state space sequence, and introducing a selective mechanism to dynamically adjust the model behavior and achieve optimal utilization of computing resources;

[0105] (3) By decomposing and fusing the spatiotemporal characteristics of service performance data of different complexities, the method can better adapt to the processing requirements of spatiotemporal service performance data of multi-type, multi-scale, and multi-granular bridge groups, thereby improving the versatility and prediction efficiency of the method.

[0106] The present invention also proposes an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the method for predicting the spatiotemporal correlation of the service performance of a group of bridges based on a graph selective state space model are implemented.

[0107] The present invention also proposes a computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the steps of the method for predicting the spatiotemporal correlation of the service performance of a group of bridges based on a graph-selective state-space model.

[0108] The memory in the embodiments of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct RAM bus RAM (DRRAM). It should be noted that the memory of the method described in the present invention is intended to include, but is not limited to, these and any other suitable types of memory.

[0109] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (digital subscriber line, DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium (eg, a floppy disk, a hard disk, a magnetic tape), an optical medium (eg, a high-density digital video disc (DVD)), or a semiconductor medium (eg, a solid state disc (SSD)).

[0110] In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in a processor or an instruction in the form of software. The steps of the method disclosed in conjunction with the embodiment of the present application can be directly embodied as a hardware processor for execution, or a combination of hardware and software modules in a processor for execution. The software module can be located in a storage medium mature in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in a memory, and the processor reads the information in the memory and completes the steps of the above method in conjunction with its hardware. To avoid repetition, it is not described in detail here.

[0111] It should be noted that the processor in the embodiment of the present application can be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method embodiment can be completed by an integrated logic circuit of hardware in the processor or an instruction in the form of software. The above processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiment of the present application can be directly embodied as a hardware decoding processor to perform, or the hardware and software modules in the decoding processor can be combined and performed. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in a memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0112] The above is a detailed introduction to the spatiotemporal correlation prediction method for the service performance of a bridge group based on a graph selective state space model proposed in the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A spatiotemporal correlation prediction method for service performance of bridge groups based on a graph selective state space model, characterized in that: The method comprises the following steps: Step 1: Determine the bridge group task object to be studied and establish a bridge group graph structure model that takes into account the effects of urban area environment and traffic flow; Step 2: Design a spatiotemporal graph prediction task based on the multi-source perception data of the service performance of bridge groups, and construct the corresponding input and output data sets; design the bridge group service performance prediction task to take the temperature, humidity, wind speed, wind direction and corrosion ion concentration environmental action variables and historical monitoring data of traffic flow as input, and predict the future bridge group environmental action and traffic flow as output; embed the multi-type, multi-scale and multi-granularity bridge group service performance multi-source perception data set into Where N is the number of bridges, T is the input data time step, and D is the feature dimension; the corresponding output dataset embedding is expressed as Where T p is the predicted output data time step; Step 3: Establish a graph-selective state-space model for spatiotemporal correlation prediction of bridge group service performance; the graph-selective state-space model for spatiotemporal correlation prediction of bridge group service performance includes two modules: spatiotemporal graph convolution module and graph-selective state-space module. The spatiotemporal graph convolution module is used to learn the spatiotemporal correlation of bridge group service performance, and the graph-selective state-space module is used to learn the spatiotemporal evolution characteristics of bridge group service performance. The outputs of the two modules are weighted fused and then output through the fully connected layer to obtain the final prediction result. Step 4: Based on the gradient descent error back propagation training graph selective state space model, the trained model is used to predict the spatiotemporal correlation between the environmental effects of bridge groups and traffic flow.

2. The method according to claim 1, characterized in that In step 1, the graph structure model is established as follows: Node definition of bridge group graph structure: each bridge is regarded as a node in the graph, and the node attributes include temperature, humidity, wind speed, wind direction, corrosion ion concentration environmental variables and traffic flow; Edge definition of bridge group graph structure: Based on the spatial position relationship between different single bridges in the bridge group, the edges between nodes are preliminarily defined; for temperature, humidity, and corrosion ion prediction, the edges are undirected edges; for wind speed and wind direction prediction, the edges are directed edges; for traffic flow prediction, the edges are directed edges, and the weight of the edges is determined according to the traffic flow size and the distance between bridges; Generation of bridge group graph structure: The bridge group graph structure is represented by G = (V; E), where V is the set of bridge nodes and E is the set of edges; based on the definition of nodes and edges, a bridge group graph structure model is generated to determine the adjacency matrix A of the bridge group graph structure.

3. The method according to claim 2, characterized in that In step 2, for the spatiotemporal data of the service performance of multi-type, multi-scale and multi-granular bridge groups, three different time granularities are used as input data, including recent data X r , periodic data X c and trend data X q ; Recent data is a historical time series directly adjacent to the forecast period, periodic data is composed of segments whose past days in the same time period as the forecast period are integer multiples of α, and trend data is composed of segments whose past days in the same time period as the forecast period are integer multiples of β; the lengths of the three types of input data, recent data, periodic data and trend data, are set according to actual forecast needs to adapt to different forecast objectives and historical data characteristics.

4. The method according to claim 3, characterized in that The step three is specifically as follows: Step 31: Establish a spatiotemporal graph convolution module; The environmental effects and traffic flow data of the bridge group are passed to the gated temporal convolution layer and the self-attention diffusion graph convolution layer in sequence after 1×1 convolution, and the output is obtained after residual connection with the input and batch normalization. Step 32: Establish a graph-selective state space module; The recent data on the environmental effects of bridge groups and traffic flow are used to r After layer normalization, they are passed to m dynamic filter graph convolution layers respectively. The outputs of the m graph convolution layers are concatenated and passed through a linear layer. The output of the linear layer is divided into a core sequence and a secondary sequence. The core sequence is passed to the graph state space selection mechanism after one-dimensional convolution and SiLU activation. Its output is multiplied element by element with the secondary sequence activated by SiLU. Finally, it passes through a linear layer and is multiplied with the layer normalized X r Residual connection to get the final output; Step 3: Combine the three time-granularity data X of bridge group environmental effects and traffic flow r , X c and X q The fusion result of the output of N spatiotemporal graph convolution modules is skipped and connected with the input X r The outputs of the M graph-selective state-space modules are weighted and summed, and the results are passed through the fully connected layer to obtain the final output result.

5. The method according to claim 4, characterized in that In step three one, Construct a gated temporal convolutional layer to learn the complex temporal dependencies of the bridge group service performance sequence data: pass the bridge group environmental effects and traffic flow data input to the gated temporal convolutional layer after passing it through a 1×1 convolution: Y=f(w1*X+b)⊙σ(w2*X+c) (4) Where Y is the output of the bridge group environmental effect and traffic flow data input after gated time convolution, w1, w2, b and c are the parameters to be optimized; f is the tanh activation function, which transforms and extracts features from the bridge group environmental effect and traffic flow data input; σ is the sigmoid activation function, which outputs a gating weight between 0 and 1. The generated gating weight controls the proportion of information flow; ⊙ is element-by-element multiplication, w1*X+b represents convolution layer 1, and w2*X+c represents convolution layer 2; Construct a self-attention diffusion graph convolution layer for learning complex spatial dependencies of service state sequence data of bridge groups: Pass the output of the gated temporal convolution layer to the self-attention diffusion graph convolution layer: Where H is the output result of the bridge group environmental effect and traffic flow data input after gated temporal convolution and self-attention diffusion graph convolution in sequence. and is the parameter to be optimized, and is the attention weight matrix of the bridge group graph structure, K is the number of diffusion steps, which is a hyperparameter and represents the diffusion process of K finite steps to simulate the service performance graph signal of the bridge group.

6. The method according to claim 5, characterized in that The attention weight matrix of the self-attention diffusion graph convolution layer of the spatiotemporal graph convolution module is calculated as follows: In the formula, a k is the learnable parameter matrix to be optimized; the propagation process of bridge group environmental effect and traffic flow data signal is divided into two directions, and the forward transformation matrix of bridge group environmental effect and traffic flow data signal is A f =A / rowsum(A), the inverse transformation matrix is ​​A b =A T / rowsum(A T ), Adaptive adjacency matrix of bridge group graph structure A k Represents the power level of the transformation matrix and the adaptive adjacency matrix. The columns are normalized during the operation and expressed as (·) / rowsum(·); the adaptive adjacency matrix The calculation of does not require prior knowledge, and end-to-end learning is performed in the gradient descent process to discover the hidden spatial dependencies between the environmental effects of bridge groups and traffic flow data. The learnable parameters to be optimized are U, are two randomly initialized node embeddings; SoftMax represents the exponential normalization operation, and ReLU represents the rectified linear unit activation function.

7. The method according to claim 6, characterized in that In step 33, the output of each spatiotemporal graph convolution block is used as the input of the next spatiotemporal graph convolution block, and a total of N spatiotemporal graph convolution blocks are set; a residual connection is set in each spatiotemporal graph convolution block, and a jump connection is made with the output layer, specifically: the result of the gated time convolution in each spatiotemporal graph convolution block is processed by a 1×1 convolution, and then added to the result of the gated time convolution in the previous spatiotemporal graph convolution block by a jump connection; Finally, the accumulation is passed to the output layer, which consists of an activation function and a linear layer, and outputs the operation results of the spatiotemporal graph convolution block.

8. The method according to claim 7, characterized in that In step 3, the outputs of the spatiotemporal graph convolution modules of the three time granularities of the service performance of the bridge group are fused. The specific method is as follows: In the formula, w r 、w c With w q is a learnable parameter to be optimized, is a scalar, and reflects the influence of the three time granularity components on the spatiotemporal correlation prediction target of the service performance of bridge groups; and They represent the output of the spatiotemporal graph data of the bridge group service state after the spatiotemporal graph convolution block, Represents the fused spatiotemporal graph convolution module output.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium for storing computer instructions, characterized in that: When the computer instructions are executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images