CN114925994A - Urban village risk assessment and risk factor positioning method based on deep learning - Google Patents
Urban village risk assessment and risk factor positioning method based on deep learning Download PDFInfo
- Publication number
- CN114925994A CN114925994A CN202210493040.9A CN202210493040A CN114925994A CN 114925994 A CN114925994 A CN 114925994A CN 202210493040 A CN202210493040 A CN 202210493040A CN 114925994 A CN114925994 A CN 114925994A
- Authority
- CN
- China
- Prior art keywords
- risk
- scene
- urban
- level
- village
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 34
- 238000012502 risk assessment Methods 0.000 title claims abstract description 24
- 238000013135 deep learning Methods 0.000 title claims abstract description 20
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 20
- 238000005516 engineering process Methods 0.000 claims abstract description 16
- 238000012549 training Methods 0.000 claims description 30
- 238000012163 sequencing technique Methods 0.000 claims description 3
- 239000002023 wood Substances 0.000 claims description 3
- 230000004927 fusion Effects 0.000 abstract description 3
- 238000003745 diagnosis Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 6
- 238000013528 artificial neural network Methods 0.000 description 5
- 238000013527 convolutional neural network Methods 0.000 description 5
- 239000013598 vector Substances 0.000 description 5
- 238000011156 evaluation Methods 0.000 description 3
- 208000027418 Wounds and injury Diseases 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000011835 investigation Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000013058 risk prediction model Methods 0.000 description 2
- 238000012800 visualization Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 206010039203 Road traffic accident Diseases 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000004931 aggregating effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 230000007958 sleep Effects 0.000 description 1
- 238000012358 sourcing Methods 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0635—Risk analysis of enterprise or organisation activities
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A30/00—Adapting or protecting infrastructure or their operation
- Y02A30/60—Planning or developing urban green infrastructure
Abstract
The invention provides a method for risk assessment and risk factor positioning of urban villages based on deep learning, which comprises the steps of searching various risk accidents by using a crowd sensing technology, clustering each risk accident, taking the risk accident exceeding a threshold value as a risk hotspot, and acquiring risk hotspot situation data according to the urban village risk hotspot, wherein the risk hotspot situation data comprises urban village level situation data and scene level situation data; extracting the characteristics of the urban village level scenes from the urban village level scene data, and extracting the characteristics of the scene level scenes from the scene level scene data; inputting the urban village level scene characteristics and the scene level scene characteristics into the constructed trained residual error network model, evaluating the urban village risk, and positioning risk factors by adopting a layer-by-layer correlation propagation algorithm; the heterogeneous data fusion is realized, the advantages of high efficiency and low consumption are achieved, and meanwhile, higher accuracy is achieved.
Description
Technical Field
The invention relates to the fields of artificial intelligence, big data and urban calculation, in particular to a method for urban village risk assessment and risk factor positioning based on deep learning.
Background
A rural village refers to a laggard village located in a developed city of developing countries. Generally, rural areas in cities are often characterized by overcrowded buildings, high population density, obsolescence of infrastructure, poor public safety, which often results in potential safety risks such as fires, traffic accidents and epidemics. Such security incidents often result in significant loss of life and property. Subsequently, reducing these risks has become one of the main tasks of the city authorities.
Traditionally, diagnosing the security risk of rural areas in cities generally relies on-site investigation, where researchers need to collect events occurring in rural areas and perform expert on-site investigation at rural areas to draw relevant conclusions, which is time consuming and labor intensive. In recent years, data-driven methods have been used to investigate and analyze security events in rural areas. However, the existing methods are highly dependent on expensive domain knowledge provided by experts, making it difficult to extend the analysis to heterogeneous urban big data, and thus, it is impossible to perform systematic diagnosis of various security risks in rural-urban regions.
Disclosure of Invention
The invention mainly aims to overcome the defects in the prior art, provides a method for risk assessment and risk factor positioning of the urban village based on deep learning, adopts a method for risk assessment of the urban village based on heterogeneous data fusion and a method for deducing potential risk factors of the urban village based on deep learning interpretability technology, has the advantages of high efficiency and low consumption, and simultaneously achieves higher estimation accuracy.
The invention adopts the following technical scheme:
a method for urban village risk assessment and risk factor positioning based on deep learning is characterized by comprising the following steps:
searching various risk accidents by using a crowd sensing technology, clustering the risk accidents, taking the risk accidents exceeding a threshold value as a risk hotspot, and acquiring risk hotspot scene data according to the risk hotspot of the urban village, wherein the risk hotspot scene data comprises urban village level scene data and scene level scene data;
extracting the characteristics of the urban village level scenes from the urban village level scene data, and extracting the characteristics of the scene level scenes from the scene level scene data;
inputting the urban village level situation characteristics and the scene level situation characteristics into the constructed trained residual network model, evaluating the urban village risk, and positioning risk factors by adopting a layer-by-layer correlation propagation algorithm.
Specifically, the method searches for various risk accidents by using the crowd sensing technology, clusters each risk accident, and uses those exceeding a threshold as risk hotspots, specifically including:
searching various risk accidents including but not limited to the occurrence time, the occurrence place, the longitude and latitude and the accident type information of the risk accidents by using a crowd sensing technology;
clustering each type of risk accident, clustering the risk accidents from the space, clustering the risk accidents from the time to obtain a clustering result, and taking the record exceeding the threshold value as a risk hotspot.
Specifically, the method includes the steps of obtaining risk hotspot situation data according to the urban village risk hotspots, wherein the risk hotspot situation data comprise urban village level situation data and scene level situation data, and specifically includes the following steps:
searching situation data according to the risk hotspots of the villages in the city, wherein the risk hotspot situation data comprise city village level situation data and scene level situation data;
for the situation data of the urban village level, collecting population density maps of risk hotspots and satellite remote sensing images of the risk hotspots;
for scene level contextual data, collecting scene information inside the risk hotspot, including but not limited to scene longitude and latitude, street view information, type information.
Specifically, the method for extracting the urban village level scene features from the urban village level scene data specifically comprises the following steps:
obtaining the population number of each urban village risk hotspot according to the longitude and latitude of the urban village risk hotspot and the grid data of the population density map;
counting the number of pixel points of the remote sensing image according to the remote sensing image of the urban village risk hotspot, and calculating the area of the current hotspot by using the size and the number of the pixel points;
collecting remote sensing images of non-hot urban villages, and marking various building targets in each satellite remote sensing image as a training set; training a Mask R-CNN model by using the training set until convergence; counting the number of houses of different types by using a Mask R-CNN model according to the remote sensing image of the urban village risk hotspot;
counting the number of urban-rural level points of interest inside the risk hotspots according to the boundary information of the urban-rural risk hotspots, wherein the points of interest include but are not limited to shops, tourist attractions, companies, living services, restaurants, self-service services, entertainment facilities, hospitals and hotels;
and splicing the population number, the area, the number of houses of different types and the number of the attention points together to obtain the urban-rural area level scene characteristics of the current risk hot spots.
Specifically, scene level scene features are extracted from scene level scene data, specifically:
searching scenes inside the urban village risk hotspots according to the boundary information of the urban village risk hotspots to obtain a scene information list;
acquiring longitude and latitude, type and street view data information of a scene according to the scene information list;
obtaining a street view inside a non-hotspot city village, marking targets in each picture, wherein the targets include but are not limited to gas tanks, advertising boards, wound wires, electric vehicles, mobile board houses, transformers and wood piles, and training a YoloV3 model to be convergent by taking the targets in the picture as a training set; extracting the number of targets and the highest confidence coefficient of a street view corresponding to each scene of the urban village risk hot spot by using a YoloV3 model as the characteristics of each scene;
calculating the distance between scenes in each hot spot according to the longitude and latitude of the scenes in each village-in-city risk hot spot;
establishing a map of each urban village risk hotspot by taking the scenes as nodes and the geographic distance between the scenes as the weight of the edges;
and realizing graph embedding by using a graph convolution network GCN, and fusing the characteristics of the corresponding edges into corresponding nodes to obtain scene level scene characteristics.
Specifically, the method further includes training a residual error network model, specifically:
constructing a training set positive sample according to the urban village level situation characteristics and the scene level situation characteristics of each urban village risk hotspot;
constructing a training set negative sample according to the urban village level situation characteristics and the scene level situation characteristics of each non-urban village risk hotspot;
constructing a residual error network model, and training on a training set until convergence to obtain a trained residual error network model;
and on the trained residual error network model, taking the urban village level scene characteristics and the scene level scene characteristics of each urban village risk hotspot as input, and analyzing the risk of the urban village risk hotspots.
Specifically, the locating of the risk factors by using the layer-by-layer correlation propagation algorithm specifically includes:
inputting the risk result of the urban village risk hotspot into a layer-by-layer correlation propagation algorithm module;
outputting a specific contribution degree value of each feature position of the urban village level contextual characteristics and the scene level contextual characteristics by a layer-by-layer correlation propagation algorithm module, and sequencing according to the feature contribution degree;
and outputting the characteristic position and the name of the set ranking.
In another aspect, the present invention provides a system for risk assessment and risk factor location of urban villages based on deep learning, including:
risk hotspot situation data acquisition unit: searching various risk accidents by using a crowd sensing technology, clustering the risk accidents, taking the risk accidents exceeding a threshold value as a risk hotspot, and acquiring risk hotspot scene data according to the risk hotspot of the urban village, wherein the risk hotspot scene data comprises urban village level scene data and scene level scene data;
a scenario feature acquisition unit: extracting the urban village level scene features from the urban village level scene data, and extracting the scene level scene features from the scene level scene data;
risk assessment and risk factor location unit: inputting the urban village level scene characteristics and the scene level scene characteristics into the constructed trained residual error network model, evaluating the urban village risk, and positioning risk factors by adopting a layer-by-layer correlation propagation algorithm.
Yet another aspect of the embodiments of the present invention provides a non-transitory computer readable storage medium storing computer instructions, wherein the computer instructions are configured to execute the method for urban village risk assessment and risk factor location based on deep learning.
As can be seen from the above description of the present invention, compared with the prior art, the present invention has the following advantages:
(1) the invention provides a method for risk assessment and risk factor positioning of a village in a city based on deep learning, which comprises the steps of searching various risk accidents by using a crowd sensing technology, clustering the various risk accidents, taking the risk accidents exceeding a threshold value as a risk hotspot, and acquiring risk hotspot situation data according to the risk hotspot of the village in the city, wherein the risk hotspot situation data comprises the situation data of the level of the village in the city and the situation data of the level of the scene; extracting the characteristics of the urban village level scenes from the urban village level scene data, and extracting the characteristics of the scene level scenes from the scene level scene data; inputting the urban village level scene characteristics and the scene level scene characteristics into the constructed trained residual error network model, evaluating the urban village risk, and positioning risk factors by adopting a layer-by-layer correlation propagation algorithm; the heterogeneous data fusion is realized, the advantages of high efficiency and low consumption are achieved, and meanwhile, higher accuracy is achieved.
(2) The method divides the characteristics into urban-rural area level situation characteristics and scene level situation characteristics, comprehensively extracts and evaluates the characteristics related to urban-rural area risks, adds each non-urban-rural area risk hotspot during risk evaluation, provides the accuracy of risk evaluation, adopts a layer-by-layer correlation propagation algorithm to position the risks, displays the characteristics and the characteristic positions in the display factors, accurately and comprehensively analyzes the urban-rural area risks, and provides a brand new and effective method for various safety risk diagnosis systems of the urban-rural areas.
Drawings
FIG. 1 is a block diagram of an overall flow scheme of an embodiment of the present invention;
FIG. 2 is a schematic diagram of a visualization of part of historical security events provided by an embodiment of the present invention;
FIG. 3 is a schematic diagram of the building identification of a village-in-city risk hotspot provided by an embodiment of the invention;
FIG. 4 is a schematic diagram of a multi-view neural network provided by an embodiment of the present invention;
the invention is described in further detail below with reference to the figures and specific examples.
Detailed Description
Fig. 1 is an overall flow framework diagram of the technical solution of the embodiment of the present invention, which is divided into three stages, namely, collecting urban village risk hotspot scenario data, extracting hotspot scenario features, and constructing an interpretable diagnosis model.
In the urban village risk hotspot scenario data collection stage, firstly, defining the urban rural risk hotspots as a concept combining time and space, wherein each risk hotspot needs to be comprehensively determined according to the urban rural location and time. And then, collecting accident occurrence related data by adopting a crowd sensing technology, obtaining urban village-to-rural risk hotspots through a clustering algorithm, collecting corresponding hotspot situation data, including satellite images, scene data and the like, and dividing the data into scene level data and urban village-to-rural level data. In the hot spot scene feature extraction stage, for scene level data, a graph is established for each risk hot spot, wherein scenes are used as nodes, and geographic distances among different scenes are used as edges. And then, inputting a city village risk hotspot diagram by adopting a Graph constraint Network model with a good effect on Graph embedding, and outputting the characteristics of the current risk hotspot. For the data of the grade of the urban village, the number of houses of the risk hotspots of the urban village is calculated by using Mask R-CNN, the number of the population of the risk hotspots of the urban village is calculated by using a population density table, the area of the risk hotspots of the urban village is calculated by using an empirical formula, and the data and the quantity of each type of scenes of the risk hotspots of the urban village are spliced by using features to obtain the grade features of the urban village of each risk hotspot. In the construction stage of an interpretable diagnosis model, a multi-view residual error network fused with a layer-by-layer correlation propagation algorithm is established, coarse-grained features and fine-grained features extracted in the previous stage are sent into the residual error network, and the residual error network is trained, so that the level of each urban village risk is evaluated, and factors of the evaluation result are explained to play a main role.
The detailed implementation steps are as follows:
a method for urban village risk assessment and risk factor positioning based on deep learning is characterized by comprising the following steps:
s1: searching various risk accidents by using a crowd sensing technology, clustering the risk accidents, taking the risk accidents exceeding a threshold value as risk hot spots, and acquiring risk hot spot scene data according to the risk hot spots of the villages in the city, wherein the risk hot spot scene data comprises the situation data of the village level in the city and the situation data of the scene level;
specifically, the method for searching various risk accidents by using crowd sensing technology, clustering each risk accident, and taking the risk accident exceeding a threshold as a risk hotspot specifically comprises the following steps:
searching various risk accidents including but not limited to the occurrence time, the occurrence place, the longitude and latitude and the accident type information of the risk accidents by using a crowd sensing technology;
clustering is carried out on each type of risk accidents, the risk accidents are clustered from the space, then the risk accidents are clustered from the time, a clustering result is obtained, and records exceeding a threshold value are used as risk hot spots.
Traditionally, researchers have defined hotspots as places where certain events occur frequently, and such definitions ignore temporal information in the event. Furthermore, in rural areas, the occurrence of high-risk events is mainly concentrated on specific time periods. Therefore, we define it not only as geospatial information, but also as high risk spatiotemporal information. Therefore, we divide a day into six time periods: morning (5:00-9:00), noon (9:00-13:00), afternoon (13:00-17:00), dusk (17:00-21:00), evening (21:00-1:00), dawn (1:00-5: 00). We represent each hotspot with the following format: city village name-time.
Then, we collect a series of historical security risk events for the building. Because of the lack of an open platform for security history events, collecting them is a labor intensive and time consuming task, we use a crowd-sourcing system to collect events over the past decade. Fig. 2 shows a visualization of some historical security risk events and their distribution on a building island.
And finally, clustering all risk events according to positions, and clustering the risk events according to the periodic labels to obtain a group of hot spots.
Specifically, the method includes the steps of obtaining risk hotspot situation data according to the urban village risk hotspots, wherein the risk hotspot situation data comprise urban village level situation data and scene level situation data, and specifically includes the following steps:
searching situation data according to the urban village risk hotspots, wherein the risk hotspot situation data comprise urban village level situation data and scene level situation data;
for the situation data of the urban village level, collecting population density maps of risk hotspots and satellite remote sensing images of the risk hotspots;
for scene level contextual data, scene information inside the risk hot spot is collected, including but not limited to scene longitude and latitude, street view information, and type information.
Although the rural-in-town risk hotspots are discovered by the event by aggregating a large number of events, there is not enough data to represent each rural-in-town risk hotspot. Therefore, it is necessary to collect the situation data of the urban village risk hotspots from a plurality of aspects. We collect contextual data from two levels to fully describe the urban village risk hotspots: the town village level and the scene level.
At the city village level, we describe each city village as a whole. We first obtained a map of the population density in the country over the last decade to calculate the population for each village in the city. And then collecting remote sensing satellite images of each urban village for calculating the number of buildings and the area size of each urban village. In addition, relevant indexes which change along with time, including temperature, wind speed, humidity and wind speed, are collected to describe the time series characteristics of the urban village risk hot spots.
At the scene level, we collected all scenes of each town village and divided them into ten kinds of scenes of stores, tourist attractions, companies, living services, restaurants, self-service services, entertainment facilities, hospitals, hotels, etc. For each scene, we also collect its longitude and latitude and street view images, and synthesize to describe this scene.
S2: extracting the urban village level scene features from the urban village level scene data, and extracting the scene level scene features from the scene level scene data;
specifically, the method for extracting the urban village level scene features from the urban village level scene data specifically comprises the following steps:
obtaining the population number of each urban village risk hotspot according to the longitude and latitude of the urban village risk hotspot and the grid data of the population density map;
counting the number of pixel points of the remote sensing image according to the remote sensing image of the urban village risk hotspot, and calculating the area of the current hotspot by using the size and the number of the pixel points;
collecting remote sensing images of non-hot urban villages, and marking various building targets in each satellite remote sensing image as a training set; training a Mask R-CNN model by using the training set until convergence; according to the remote sensing image of the urban village-in-village risk hot spots, counting the number of houses of different types by using a Mask R-CNN model, as shown in FIG. 3;
counting the number of urban-rural level points of interest inside the risk hotspots according to the boundary information of the urban-rural risk hotspots, wherein the points of interest include but are not limited to shops, tourist attractions, companies, living services, restaurants, self-service services, entertainment facilities, hospitals and hotels;
and splicing the population number, the area, the number of different types of houses and the number of the attention points together to obtain the situation characteristics of the current risk hotspots at the urban-rural area level.
Specifically, scene level scene features are extracted from scene level scene data, specifically:
searching scenes inside the urban village risk hotspots according to the boundary information of the urban village risk hotspots to obtain a scene information list;
acquiring longitude and latitude, type and street view data information of a scene according to the scene information list;
obtaining a street view inside a non-hotspot city village, marking targets in each picture, wherein the targets include but are not limited to gas tanks, advertising boards, wound wires, electric vehicles, mobile board houses, transformers and wood piles, and training a YoloV3 model to be convergent by taking the targets in the picture as a training set; extracting the number of targets and the highest confidence coefficient of a street view corresponding to each scene of the urban village risk hot spot by using a YoloV3 model as the characteristics of each scene;
calculating the distance between scenes in each hot spot according to the longitude and latitude of the scenes in each village-in-city risk hot spot;
establishing a map of each urban village risk hotspot by taking the scenes as nodes and the geographic distance between the scenes as the weight of the edges;
graph embedding is realized by using a graph convolution network GCN, and the characteristics of the corresponding edges are fused into corresponding nodes to obtain scene level scene characteristics.
S3: inputting the urban village level scene characteristics and the scene level scene characteristics into the constructed trained residual error network model, evaluating the urban village risk, and positioning risk factors by adopting a layer-by-layer correlation propagation algorithm.
Specifically, the method further comprises training a residual error network model, specifically:
constructing a training set positive sample according to the urban village level scene characteristics and the scene level scene characteristics of each urban village risk hotspot;
constructing a training set negative sample according to the urban village level situation characteristics and the scene level situation characteristics of each non-urban village risk hotspot;
constructing a residual error network model, and training on a training set until convergence to obtain a trained residual error network model;
and on the trained residual error network model, taking the urban village level scene characteristics and the scene level scene characteristics of each urban village risk hotspot as input, and analyzing the risk of the urban village risk hotspots.
Specifically, the positioning of the risk factors by using the layer-by-layer correlation propagation algorithm specifically includes:
inputting the risk result of the urban village risk hotspot into a layer-by-layer correlation propagation algorithm module;
outputting a specific contribution degree value of each feature position of the urban village level contextual characteristics and the scene level contextual characteristics by a layer-by-layer correlation propagation algorithm module, and sequencing according to the feature contribution degree;
and outputting the characteristic position and the name of the set ranking. After each risk hotspot is represented in a multi-view manner, an interpretable neural network needs to be constructed next to fit the features. The model may not only predict the types of potential risk events, but may also explain the reasons behind these risk events. Specifically, a multi-layer neural network is first established to predict the risk of rural areas in the city. Next, layer-by-layer correlation propagation (LRP) is used to analyze which factors are primarily affecting the risk leading up to the occurrence.
In this step, the risk of potential hotspots is assessed using a classification model based on the features of the two layers. Intuitively, we can directly connect these features as vectors to train a classification neural network to classify them. However, this approach ignores the different roles of these features in the underlying risk prediction task. In particular, the spatial features describe the geographical distribution of the venue; for example, a distribution of crowded places would be at higher risk than elsewhere in an urban village. The time profile describes the correlation of external environmental factors over different periods of time, for example, the weather of the previous hours may have an effect on the current weather, resulting in a risk event in the current environment. The basic features describe various basic information for the entire urban village, e.g., the smaller the area, the more crowded the urban village, the greater the risk. Therefore, we constructed a multi-view neural network (MvNN) to predict potential risks, as shown in fig. 4.
In MvNN, the site level and the village-in-town level are considered view a and view B, respectively. For each view, two fully-connected layers are used to generate vectors of the same size. The two prediction vectors are then added and connected to a fully connected layer to predict the potential risk of villages in the city. In addition, a DropOut layer is added after each fully-connected layer of the MvNN, which sleeps part of the neurons in the fully-connected layer to avoid MvNN overfitting. At the same time, we used ReLU as the activation function in each fully-connected layer and trained the model with Adam optimizer.
To this end, the risk potential prediction model may efficiently use multi-view features to predict risk potential. However, it can only output the final predicted result, and cannot interpret which feature contributes to the generation of the result. It is important for the government to know each predicted impact characteristic in order to implement the corresponding precautionary measure. Therefore, layer-by-Layer Relevance Propagation (LRP) is combined in a model with a risk prediction model to enable diagnosis of potentially interpretable risk factors. In particular, layer-by-layer correlation propagation is used to monitor and analyze inputs to a risk potential prediction model. When the prediction model outputs the prediction labels, the layer-by-layer correlation propagation outputs which position in the input vector contributes most to the result, and highlights the feature position in the feature and significant factors.
layer-by-Layer Relevance Propagation (LRP) interprets classifier decisions by decomposition. Mathematically, it reassigns the predictions f (x) backwards using a local reassignment rule until a correlation score Ri is assigned to each input variable. The key to this reallocation process is called dependency protection and can be summarized as
∑ i R i ＝…＝Σ j R j＝ ∑ k R k＝ f(x)
This attribute indicates that the relevant total amount is saved at each step of the reassignment process. In the diagnosis of high risk factors, because the input of the risk prediction model is the feature representation of multiple views, the LRP can be directly used for monitoring and calculating the contribution of each dimension of an input vector under different views, and the positions with great contribution in the heat map of the features and significant factors are highlighted, so that the feature importance can be analyzed from the urban village level and the scene level, the factors which are more likely to cause different risk events are found, and the more comprehensive management of urban villages by urban authorities is promoted.
In another aspect, the present invention provides a system for risk assessment and risk factor location of urban villages based on deep learning, including:
risk hotspot situation data acquisition unit: searching various risk accidents by using a crowd sensing technology, clustering the risk accidents, taking the risk accidents exceeding a threshold value as a risk hotspot, and acquiring risk hotspot scene data according to the risk hotspot of the urban village, wherein the risk hotspot scene data comprises urban village level scene data and scene level scene data;
a scenario feature acquisition unit: extracting the characteristics of the urban village level scenes from the urban village level scene data, and extracting the characteristics of the scene level scenes from the scene level scene data;
risk assessment and risk factor location unit: inputting the urban village level situation characteristics and the scene level situation characteristics into the constructed trained residual network model, evaluating the urban village risk, and positioning risk factors by adopting a layer-by-layer correlation propagation algorithm.
Yet another aspect of embodiments of the present invention provides a non-transitory computer-readable storage medium storing computer instructions for executing the above-mentioned road obstacle risk assessment method based on urban big data.
Computer instructions (also known as programs, software applications or code) include machine instructions for a programmable processor, and may be implemented using high-level procedural and/or object-oriented programming languages, and/or assembly/machine languages. The terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, apparatus, and/or device (e.g., magnetic discs, optical disks, memory, Programmable Logic Devices (PLDs)) used to provide machine instructions and/or data to a programmable processor, including a machine-readable medium that receives machine instructions as a machine-readable signal.
The above description is only an embodiment of the present invention, but the design concept of the present invention is not limited thereto, and any insubstantial modifications made by using the design concept should fall within the scope of infringing the present invention.
Claims (9)
1. A method for urban village risk assessment and risk factor positioning based on deep learning is characterized by comprising the following steps:
searching various risk accidents by using a crowd sensing technology, clustering the risk accidents, taking the risk accidents exceeding a threshold value as a risk hotspot, and acquiring risk hotspot scene data according to the risk hotspot of the urban village, wherein the risk hotspot scene data comprises urban village level scene data and scene level scene data;
extracting the urban village level scene features from the urban village level scene data, and extracting the scene level scene features from the scene level scene data;
inputting the urban village level situation characteristics and the scene level situation characteristics into the constructed trained residual network model, evaluating the urban village risk, and positioning risk factors by adopting a layer-by-layer correlation propagation algorithm.
2. The method for urban village risk assessment and risk factor positioning based on deep learning according to claim 1, wherein the crowd sensing technology is used for searching various types of risk accidents, clustering each type of risk accidents, and using the risk accidents exceeding a threshold as risk hotspots, specifically:
searching various risk accidents including but not limited to the occurrence time, the occurrence place, the longitude and latitude and the accident type information of the risk accidents by using a crowd sensing technology;
clustering each type of risk accident, clustering the risk accidents from the space, clustering the risk accidents from the time to obtain a clustering result, and taking the record exceeding the threshold value as a risk hotspot.
3. The method for risk assessment and risk factor location of urban villages based on deep learning according to claim 1, wherein risk hotspot scenario data is obtained according to urban village risk hotspots, and the risk hotspot scenario data comprises urban village level scenario data and scene level scenario data, specifically:
searching situation data according to the urban village risk hotspots, wherein the risk hotspot situation data comprise urban village level situation data and scene level situation data;
for the situation data of the urban village level, collecting population density maps of risk hotspots and satellite remote sensing images of the risk hotspots;
for scene level contextual data, collecting scene information inside the risk hotspot, including but not limited to scene longitude and latitude, street view information, type information.
4. The method for risk assessment and risk factor location of urban villages based on deep learning as claimed in claim 1, wherein the urban village level contextual features are extracted from the urban village level contextual data, and specifically are:
obtaining the population number of each urban village risk hotspot according to the longitude and latitude of the urban village risk hotspot and the grid data of the population density map;
counting the number of pixel points of the remote sensing image according to the remote sensing image of the urban village risk hot spot, and calculating the area of the current hot spot by using the size and the number of the pixel points;
collecting remote sensing images of non-hot urban villages, and marking various building targets in each satellite remote sensing image as a training set; training a Mask R-CNN model by using the training set until convergence; according to the remote sensing image of the urban village risk hot spot, the number of different types of houses is counted by using a MaskR-CNN model;
counting the number of urban-rural level points of interest inside the risk hotspots according to boundary information of the urban-rural risk hotspots, wherein the points of interest include but are not limited to shops, tourist attractions, companies, living services, restaurants, self-service services, entertainment facilities, hospitals and hotels;
and splicing the population number, the area, the number of houses of different types and the number of the attention points together to obtain the urban-rural area level scene characteristics of the current risk hot spots.
5. The urban village risk assessment and risk factor positioning method based on deep learning according to claim 1, characterized in that scene-level contextual features are extracted from scene-level contextual data, specifically:
searching scenes inside the urban village risk hotspots according to the boundary information of the urban village risk hotspots to obtain a scene information list;
acquiring longitude and latitude, type and street view data information of a scene according to the scene information list;
obtaining a street view inside a non-hotspot city village, marking targets in each picture, wherein the targets include but are not limited to gas tanks, advertising boards, wound wires, electric vehicles, mobile board houses, transformers and wood piles, and training a YoloV3 model to be convergent by taking the targets in the picture as a training set; extracting the number of targets and the highest confidence coefficient of a street view corresponding to each scene of the urban village risk hot spot by using a YoloV3 model as the characteristics of each scene;
calculating the distance between scenes in each hot spot according to the longitude and latitude of the scenes in each village-in-city risk hot spot;
establishing a map of each urban village risk hotspot by taking the scenes as nodes and the geographic distance between the scenes as the weight of the edges;
graph embedding is realized by using a graph convolution network GCN, and the characteristics of the corresponding edges are fused into corresponding nodes to obtain scene level scene characteristics.
6. The urban village risk assessment and risk factor positioning method based on deep learning according to claim 1, characterized by further comprising training a residual error network model, specifically:
constructing a training set positive sample according to the urban village level situation characteristics and the scene level situation characteristics of each urban village risk hotspot;
constructing a training set negative sample according to the urban village level situation characteristics and the scene level situation characteristics of each non-urban village risk hotspot;
constructing a residual error network model, and training on a training set until convergence to obtain a trained residual error network model;
and on the trained residual error network model, taking the urban village level scene characteristics and the scene level scene characteristics of each urban village risk hotspot as input, and analyzing the risk of the urban village risk hotspots.
7. The urban village risk assessment and risk factor positioning method based on deep learning according to claim 6, wherein the positioning of the risk factors by using a layer-by-layer correlation propagation algorithm specifically comprises:
inputting the risk result of the urban village risk hotspot into a layer-by-layer correlation propagation algorithm module;
outputting specific contribution degree values of each feature position of the urban village level contextual characteristics and the scene level contextual characteristics by a layer-by-layer correlation propagation algorithm module, and sequencing according to the feature contribution degrees;
and outputting the characteristic position and the name for setting the ranking.
8. A system for urban village risk assessment and risk factor positioning based on deep learning is characterized by comprising:
risk hotspot situation data acquisition unit: searching various risk accidents by using a crowd sensing technology, clustering the risk accidents, taking the risk accidents exceeding a threshold value as a risk hotspot, and acquiring risk hotspot scene data according to the risk hotspot of the urban village, wherein the risk hotspot scene data comprises urban village level scene data and scene level scene data;
a scene feature acquisition unit: extracting the urban village level scene features from the urban village level scene data, and extracting the scene level scene features from the scene level scene data;
risk assessment and risk factor location unit: inputting the urban village level scene characteristics and the scene level scene characteristics into the constructed trained residual error network model, evaluating the urban village risk, and positioning risk factors by adopting a layer-by-layer correlation propagation algorithm.
9. A non-transitory computer readable storage medium having computer instructions stored thereon for performing the method of any one of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210493040.9A CN114925994A (en) | 2022-05-07 | 2022-05-07 | Urban village risk assessment and risk factor positioning method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210493040.9A CN114925994A (en) | 2022-05-07 | 2022-05-07 | Urban village risk assessment and risk factor positioning method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114925994A true CN114925994A (en) | 2022-08-19 |
Family
ID=82807829
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210493040.9A Pending CN114925994A (en) | 2022-05-07 | 2022-05-07 | Urban village risk assessment and risk factor positioning method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114925994A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116894585A (en) * | 2023-07-25 | 2023-10-17 | 浙江三网科技股份有限公司 | Intelligent analysis method and system applied to future community comprehensive data |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108198415A (en) * | 2017-12-28 | 2018-06-22 | 同济大学 | A kind of city expressway accident forecast method based on deep learning |
| BR102019008102A2 (en) * | 2019-04-22 | 2020-10-27 | Vale S.A | method and system for computing railway incident risk indices |
| CN112862240A (en) * | 2020-12-30 | 2021-05-28 | 厦门大学 | Road obstacle risk assessment method and device based on urban big data and readable storage medium |
| KR20210067861A (en) * | 2019-11-29 | 2021-06-08 | 울산과학기술원 | Apparatus and method for evaluating walking safety risk, and recording media recorded program realizing the same |
| CN114358462A (en) * | 2021-11-10 | 2022-04-15 | 华北科技学院(中国煤矿安全技术培训中心) | City safety risk assessment method and system and electronic equipment |
| CN114445624A (en) * | 2022-01-24 | 2022-05-06 | 吉林大学 | Fine-grained traffic accident risk identification method in urban geographic space |
-
2022
- 2022-05-07 CN CN202210493040.9A patent/CN114925994A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108198415A (en) * | 2017-12-28 | 2018-06-22 | 同济大学 | A kind of city expressway accident forecast method based on deep learning |
| BR102019008102A2 (en) * | 2019-04-22 | 2020-10-27 | Vale S.A | method and system for computing railway incident risk indices |
| KR20210067861A (en) * | 2019-11-29 | 2021-06-08 | 울산과학기술원 | Apparatus and method for evaluating walking safety risk, and recording media recorded program realizing the same |
| CN112862240A (en) * | 2020-12-30 | 2021-05-28 | 厦门大学 | Road obstacle risk assessment method and device based on urban big data and readable storage medium |
| CN114358462A (en) * | 2021-11-10 | 2022-04-15 | 华北科技学院(中国煤矿安全技术培训中心) | City safety risk assessment method and system and electronic equipment |
| CN114445624A (en) * | 2022-01-24 | 2022-05-06 | 吉林大学 | Fine-grained traffic accident risk identification method in urban geographic space |
Non-Patent Citations (4)
| Title |
|---|
| HONGLEI REN 等: "A Deep Learning Approach to the Citywide Traffic Accident Risk Prediction", 《2018 21ST INTERNATIONAL CONFERENCE ON INTELLIGENT TRANSPORTATION SYSTEMS (ITSC)》, 9 December 2018 (2018-12-09) * |
| 刘兴旺;朱大明;: "GIS在城市社区子区域火灾风险评估中的应用", 地理空间信息, no. 05, 28 October 2009 (2009-10-28) * |
| 唐飞;胡隆华;霍然;徐勇;祝实;姚斌;: "基于层次分析法的城中村区域火灾风险评估模型", 消防科学与技术, no. 06, 15 June 2010 (2010-06-15) * |
| 闫旭;范晓亮;郑传潘;臧彧;王程;程明;陈龙彪;: "基于图卷积神经网络的城市交通态势预测算法", 浙江大学学报(工学版), no. 06, 6 July 2020 (2020-07-06) * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116894585A (en) * | 2023-07-25 | 2023-10-17 | 浙江三网科技股份有限公司 | Intelligent analysis method and system applied to future community comprehensive data |
| CN116894585B (en) * | 2023-07-25 | 2024-02-20 | 浙江三网科技股份有限公司 | Intelligent analysis method and system applied to future community comprehensive data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Yuan et al. | A survey of traffic prediction: from spatio-temporal data to intelligent transportation | |
| Lee et al. | A visual analytics system for exploring, monitoring, and forecasting road traffic congestion | |
| Wang et al. | Machine learning-based regional scale intelligent modeling of building information for natural hazard risk management | |
| Souza et al. | City Information Modelling as a support decision tool for planning and management of cities: A systematic literature review and bibliometric analysis | |
| Feng et al. | A survey on trajectory data mining: Techniques and applications | |
| Sun et al. | DxNAT—Deep neural networks for explaining non-recurring traffic congestion | |
| US7801842B2 (en) | Method and system for spatial behavior modification based on geospatial modeling | |
| US8341110B2 (en) | Temporal-influenced geospatial modeling system and method | |
| US7917458B2 (en) | Temporal-influenced geospatial modeling system and method | |
| US7571146B2 (en) | Event, threat and result change detection system and method | |
| Liu et al. | A spatio‐temporal ensemble method for large‐scale traffic state prediction | |
| Zheng | Urban computing | |
| Shang et al. | Estimating building-scale population using multi-source spatial data | |
| KR101937940B1 (en) | Method of deciding cpted cctv position by big data | |
| Yuan et al. | Smart flood resilience: harnessing community-scale big data for predictive flood risk monitoring, rapid impact assessment, and situational awareness | |
| CN116227834A (en) | Intelligent scenic spot digital platform based on three-dimensional point cloud model | |
| CN112862240B (en) | Road obstacle risk assessment method and device based on urban big data and readable storage medium | |
| Deng et al. | The pulse of urban transport: Exploring the co-evolving pattern for spatio-temporal forecasting | |
| Liu et al. | Towards human-centric digital twins: Leveraging computer vision and graph models to predict outdoor comfort | |
| Zhang et al. | Emergency-oriented spatiotemporal trajectory pattern recognition by intelligent sensor devices | |
| CN114925994A (en) | Urban village risk assessment and risk factor positioning method based on deep learning | |
| Wang et al. | Simulating the effect of urban light rail transit on urban development by coupling cellular automata and conjugate gradients | |
| Ueda et al. | Spatio-temporal multidimensional collective data analysis for providing comfortable living anytime and anywhere | |
| Cerqueira et al. | Integrative analysis of traffic and situational context data to support urban mobility planning | |
| CN113971496A (en) | Urban traffic network state evolution trend prediction method and system under influence of activities |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |