Research method for smooth relation between obstacles in evacuation road and walking evacuation

Abstract

The invention discloses a method for researching a smooth relation between obstacles in an evacuation road and walking evacuation, and belongs to the technical field of computer simulation. The research method comprises the following specific steps: (1) collecting real-time data of an evacuation scene and preprocessing each group of data; (2) predicting behaviors of pedestrians and vehicles through a motion prediction model; (3) simulating an evacuation scene according to the real-time data and the prediction result; (4) simulating different evacuation strategies and evaluating evacuation smoothness; (5) optimizing an evacuation strategy and performing simulation optimization through actual evacuation drilling; according to the method, future actions of people and vehicles can be predicted more accurately, meanwhile, prediction is more credible, different responses of different people in the same evacuation scene can be better understood, the execution efficiency and safety of evacuation decision making are improved, researchers can better understand the mutual influence between scene objects, and the research efficiency is improved. More comprehensive information can be provided, and a better evacuation strategy can be formulated.

Description

Technical Field

[0001] The invention relates to the technical field of computer simulation, and in particular to a method for studying the relationship between obstacles in an evacuation road and the smoothness of pedestrian evacuation. Background Art

[0002] Emergency evacuation is an important topic in the field of urban planning and safety management, and its success or failure is directly related to people's life safety. In the process of emergency evacuation, obstacles in the evacuation road and the smoothness of the evacuation passage become key factors, affecting the rapid evacuation of personnel. However, in actual evacuation scenarios, obstacles can include various types of obstacles, such as people, vehicles, furniture, building structures, etc., and their location, number and movement mode have an important impact on the smoothness of evacuation. With the growth of urban population and the advancement of urbanization, emergency evacuation events in cities are becoming more and more frequent, such as fire, earthquake, severe weather, etc. In these emergency situations, the life safety of people depends on effective evacuation plans and smooth evacuation roads. However, obstacles and congestion on evacuation roads often affect the rapid evacuation of people, which may lead to serious consequences; therefore, it is particularly important to invent a research method on the relationship between obstacles in evacuation roads and the smoothness of pedestrian evacuation.

[0003] The existing research methods on the relationship between obstacles in evacuation roads and the smoothness of pedestrian evacuation have low prediction credibility, which reduces the execution efficiency and safety of evacuation decisions; in addition, the existing research methods on the relationship between obstacles in evacuation roads and the smoothness of pedestrian evacuation do not facilitate researchers to understand the mutual influence between scene objects and cannot provide more comprehensive information; therefore, we propose a research method on the relationship between obstacles in evacuation roads and the smoothness of pedestrian evacuation. Summary of the invention

[0004] The purpose of the present invention is to solve the defects in the prior art and to propose a research method for the relationship between obstacles in evacuation roads and the smoothness of pedestrian evacuation.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] The research method of the relationship between obstacles in the evacuation road and the smoothness of pedestrian evacuation, the specific steps of the research method are as follows:

[0007] (1) Collect real-time data of evacuation scenarios and preprocess each set of data;

[0008] (2) Predict pedestrian and vehicle behavior through motion prediction models;

[0009] (3) Simulate evacuation scenarios based on real-time data and prediction results;

[0010] (4) Simulate different evacuation strategies and evaluate evacuation smoothness;

[0011] (5) Optimize evacuation strategies and conduct simulation optimization through actual evacuation drills.

[0012] It is characterized in that the specific steps of real-time data preprocessing in step (1) are as follows:

[0013] Step 1: Collect real-time data of the evacuation scene through sensors and cameras, then use Gaussian filtering to remove noise in each group of real-time signal data and smooth the data, then calculate the standard deviation of the signal data set, and detect and filter out abnormal data based on the calculated standard deviation, unify the data format, and then detect whether there are duplicate data records. If there are duplicate data, delete them;

[0014] Step 2: Detect the missing values ​​in each group of data, mark the location of each missing value in the corresponding data, perform statistical and visual analysis on the missing values ​​in each group of data to obtain the distribution and impact range of the missing values, and replace the abnormal values ​​or missing values ​​with the average or median of the corresponding K groups of data points found by the KNN algorithm;

[0015] Step 3: Extract a certain number of image frames from the real-time video data every second to cover the entire video, obtain the pixel values ​​of each group in each group of image frames, and then for each group of pixels, replace the value of the central pixel with the average value of the pixels around the pixel. Repeat the above process to process the entire image to remove the existing noise, watermark or distortion, and adjust the image size and resolution so that each group of images has the same size.

[0016] It is characterized in that the specific steps of constructing the motion prediction model in step (2) are as follows:

[0017] Step 1: Obtain historical data of evacuation scenarios, and extract feature information of each group of data after preprocessing each group of historical data. Then, create new features by randomly replacing or sampling the acquired feature data. Then, through feature scaling and standardization, make different features on the same scale, and randomly divide each group of collected feature data into training set and test set;

[0018] Step 2: Create a motion prediction model and define the model loss function. Then select the stochastic gradient descent algorithm to adjust the model parameters to minimize the loss function. Divide the training set into small batches and use each batch to update the model weights. Repeat the forward propagation to calculate the output of the model, and then perform backpropagation to calculate the gradient and update the model parameters until all training sets are used up.

[0019] Step 3: Use an independent test set to evaluate the performance of the trained model. Based on the model evaluation results, if the evaluation results do not meet the preset threshold, for each set of data, the loss value of the motion prediction model training results is calculated, and then the test set is replaced with another subset. The remaining subset is taken as the training set, and the loss value is calculated again until all data are predicted once. The combined parameters corresponding to the minimum loss value are selected as the optimal parameters in the data interval and the original parameters of the motion prediction model are replaced.

[0020] It is characterized in that the specific calculation formula of the gradient descent algorithm in step 2 is as follows:

[0021] θ=θ-α▽L(θ) (1)

[0023] In the formula, θ represents the model parameters that need to be adjusted; α represents the learning rate;

[0024] The specific calculation formula of the loss function in step 2 is as follows:

[0025]

[0026] In the formula, n represents the number of data; represents the actual observed value; Represents the predicted value of the model.

[0027] It is characterized in that the specific steps of the motion prediction model behavior prediction in step (2) are as follows:

[0028] Step I: Linearly transform the preprocessed real-time data from different angles through multiple sets of attention heads, extract features that match the features used in model training, and fuse the outputs of multiple sets of attention heads to obtain the corresponding multimodal data set;

[0029] Step II: Input each group of data in the multimodal dataset into the trained motion prediction model. The input data starts from the input layer of the motion prediction model and passes through each hidden layer of the model. Each hidden layer performs a linear transformation on the input data and transmits it layer by layer through the weights and activation functions between each layer. Then, the output layer outputs the probability distribution of one or more groups of behaviors.

[0030] Step III: Select the behavior data with the highest probability as the final prediction result, and then explain and visualize the prediction results of the model for users to understand and analyze.

[0031] It is characterized in that the specific steps of evacuation scene simulation in step (3) are as follows:

[0032] Step 1: Use computer graphics technology to simulate the actual evacuation scenario, integrate the collected real-time data into the virtual environment, and ensure the time synchronization of each set of data. Then identify the pedestrian and obstacle information in the real-time data, and build a personnel behavior model and an obstacle model.

[0033] Step ②: Integrate the prediction data generated by the motion prediction model into the virtual environment, and then start the simulation in the virtual environment. At the same time, the personnel and obstacle models will move in the virtual environment, and monitor the simulation results in real time. After the simulation is completed, record the data during the simulation.

[0034] It is characterized in that the specific steps of the evacuation smoothness assessment in step (4) are as follows:

[0035] Step 1: Collect the paths, personnel behaviors, obstacles and evacuation strategy data in the simulation data, then extract the entity information in each set of recorded data and unify the format information of each set of data;

[0036] Step 2: Collect various knowledge and information related to evacuation smoothness from the Internet and historical evacuation databases, classify, remove duplicates and filter the collected evacuation knowledge, and use NLP technology to identify and extract entities in the processed evacuation knowledge;

[0037] Step 3: Extract the corresponding attributes of each entity from the relevant knowledge information, establish the relationship between entities, form the connection of the evacuation knowledge graph, process the entities, attributes and relationships into the corresponding graph structure in the form of triples, select a suitable graph database to store and manage the evacuation knowledge graph, and continuously update and maintain the evacuation knowledge graph;

[0038] Step 4: Match the extracted entities with the entities in the evacuation knowledge graph, output the evaluation scores of each group of data with consistent matching results, and feed them back to the researchers for review.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] 1. The research method of the relationship between obstacles in the evacuation road and the smoothness of pedestrian evacuation obtains the historical data of the evacuation scene, and extracts the feature information of each group of data after preprocessing each group of historical data. Then, the acquired feature data is used to create new features through random replacement or sampling, and then different features are made on the same scale through feature scaling and standardization. The collected groups of feature data are randomly divided into training sets and test sets, and then a motion prediction model is created and trained. Then, the preprocessed real-time data is linearly transformed from different angles through multiple groups of attention heads to extract features that match the features used in model training, and the outputs of multiple groups of attention heads are fused to obtain the corresponding multi-mode The multimodal dataset is used to input each group of data in the trained motion prediction model. The input data starts from the input layer of the motion prediction model and is processed by each hidden layer of the model. Then, it is transmitted layer by layer through the weights and activation functions between each layer. Then, the output layer outputs the probability distribution of one or more groups of behaviors, and the behavior data with the highest probability is selected as the final prediction result. Then, the prediction results of the model are explained and visualized for users to understand and analyze. This can more accurately predict the future actions of people and vehicles, and make the prediction more credible, which is conducive to better understanding the different reactions of different people in the same evacuation scenario and improving the execution efficiency and safety of evacuation decisions.

[0041] 2. The research method of the relationship between obstacles in the evacuation road and the smoothness of pedestrian evacuation collects the path, personnel behavior, obstacles and evacuation strategy data in the simulation data, and then extracts the entity information in each set of recorded data, and unifies the format information of each set of data, collects various knowledge and information related to evacuation smoothness from the Internet and historical evacuation databases, and classifies, removes duplicates and screens the collected evacuation knowledge, identifies and extracts entities in the processed evacuation knowledge through NLP technology, extracts the corresponding attributes of each entity from the relevant knowledge information, and establishes the relationship between entities to form a connection of the evacuation knowledge graph, processes entities, attributes and relationships into corresponding graph structures in the form of triples, selects a suitable graph database to store and manage the evacuation knowledge graph, and continuously updates and maintains the evacuation knowledge graph, matches the extracted entities with the entities in the evacuation knowledge graph, and outputs the evaluation scores of each set of data with consistent matching results, and feeds them back to researchers for review, which is conducive to researchers to better understand the mutual influence between scene objects, provide more comprehensive information, and formulate better evacuation strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0043] Figure 1A flowchart of a method for studying the relationship between obstacles in evacuation roads and smoothness of pedestrian evacuation proposed by the present invention;

[0044] Figure 2 This is a schematic diagram of an experimental simulation scenario for the research method on the relationship between obstacles in evacuation roads and smoothness of pedestrian evacuation proposed in the present invention. DETAILED DESCRIPTION

[0045] Example 1

[0046] Reference Figure 1 , a research method for the relationship between obstacles in evacuation roads and smoothness of pedestrian evacuation. The specific steps of this research method are as follows:

[0047] Collect real-time data of evacuation scenarios and preprocess each set of data.

[0048] Specifically, real-time data of the evacuation scene is collected through sensors and cameras, and then the noise in each group of real-time signal data is removed and the data is smoothed through Gaussian filtering. Then the standard deviation of the signal data set is calculated, and the abnormal data is detected and screened out according to the calculated standard deviation, and the data format is unified. Then it is detected whether there are duplicate data records. If there are duplicate data, they are deleted, and the missing values ​​in each group of data are detected, and the location of each missing value in the corresponding data is marked. The missing values ​​in each group of data are statistically and visually analyzed to obtain the distribution and influence range of the missing values, and the average value or median of the corresponding K groups of data points found by the KNN algorithm are used to replace the abnormal value or missing value. A certain number of image frames are extracted from the real-time video data per second to cover the entire video, and the pixel values ​​of each group in each group of image frames are obtained. Then, for each group of pixels, the value of the central pixel is replaced by the average value of the pixels around the pixel. The above process is repeated to process the entire image to remove the existing noise, watermark or distortion, and the image size and resolution are adjusted so that each group of images has the same size.

[0049] Predict pedestrian and vehicle behavior through motion prediction models.

[0050] Specifically, historical data of evacuation scenes are obtained, and feature information of each group of data is extracted after preprocessing each group of historical data. Then, new features are created by randomly replacing or sampling the acquired feature data. Then, different features are made on the same scale through feature scaling and standardization. Each group of collected feature data is randomly divided into a training set and a test set, a motion prediction model is created and the loss function of the model is defined. Then, a stochastic gradient descent algorithm is selected to adjust the model parameters to minimize the loss function, the training set is divided into small batches, and each batch is used to update the weights of the model. The forward propagation is repeatedly performed to calculate the output of the model, and then back propagation is performed to calculate the gradient and update the parameters of the model until all training sets are used up and then stopped. An independent test set is used to evaluate the performance of the trained model. According to the model evaluation result, if the evaluation result does not meet the preset threshold, for each group of data, the loss value of the motion prediction model training result is statistically calculated, and then the test set is replaced with another subset, and the remaining subset is taken as the training set, and the loss value is calculated again until all data are predicted once, and the combined parameters corresponding to the minimum loss value are selected as the optimal parameters in the data interval and the original parameters of the motion prediction model are replaced.

[0051] Specifically, the preprocessed real-time data is linearly transformed from different angles through multiple groups of attention heads to extract features that match the features used in model training, and the outputs of the multiple groups of attention heads are fused to obtain the corresponding multimodal data set. Each group of data in the multimodal data set is input into the trained motion prediction model, and the input data starts from the input layer of the motion prediction model and passes through the hidden layers of the model. After each hidden layer performs a linear transformation on the input data, it is passed layer by layer through the weights and activation functions between the layers. Then, the output layer outputs the probability distribution of one or more groups of behaviors, and the behavior data with the highest probability is selected as the final prediction result. Then, the prediction result of the model is interpreted and visualized for users to understand and analyze.

[0052] It should be further explained that the specific calculation formula of the gradient descent algorithm is as follows:

[0053]

[0054] In the formula, θ represents the model parameters that need to be adjusted; α represents the learning rate;

[0055] The specific calculation formula of the loss function is as follows:

[0056]

[0057] In the formula, n represents the number of data; represents the actual observed value; Represents the predicted value of the model.

[0058] Example 2

[0059] Reference Figure 1-2 , a research method for the relationship between obstacles in evacuation roads and smoothness of pedestrian evacuation. The specific steps of this research method are as follows:

[0060] Simulate evacuation scenarios based on real-time data and forecast results.

[0061] Specifically, Figure 2 As shown, computer graphics technology is used to simulate the actual evacuation scenario, the collected real-time data is integrated into the virtual environment, and the time synchronization of each set of data is ensured. Then, the pedestrian and obstacle information in the real-time data is identified, and the personnel behavior model and obstacle model are constructed. The predicted data generated by the motion prediction model is integrated into the virtual environment, and then the simulation in the virtual environment is started. At the same time, the personnel and obstacle models will move in the virtual environment, and the simulation results will be monitored in real time. After the simulation, the data during the simulation period will be recorded.

[0062] Simulate different evacuation strategies and evaluate evacuation efficiency.

[0063] Specifically, the paths, personnel behaviors, obstacles and evacuation strategy data in the simulation data are collected, and then the entity information in each set of recorded data is extracted, and the format information of each set of data is unified. Various knowledge and information related to evacuation smoothness are collected from the Internet and historical evacuation databases, and the collected evacuation knowledge is classified, deduplicated and screened. The entities in the processed evacuation knowledge are identified and extracted through NLP technology, and the corresponding attributes of each entity are extracted from the relevant knowledge information. The relationship between the entities is established to form a connection of the evacuation knowledge graph, and the entities, attributes and relationships are processed into corresponding graph structures in the form of triples. A suitable graph database is selected to store and manage the evacuation knowledge graph, and the evacuation knowledge graph is continuously updated and maintained. The extracted entities are matched with the entities in the evacuation knowledge graph, and the evaluation scores of each set of data with consistent matching results are output and fed back to the researchers for review.

[0064] Optimize evacuation strategies and conduct simulation optimization through actual evacuation drills.

Claims

1. A research method for the relationship between obstacles in evacuation roads and smoothness of pedestrian evacuation, characterized in that: The specific steps of this research method are as follows: (1) Collect real-time data of evacuation scenarios and preprocess each set of data; (2) Predict pedestrian and vehicle behavior through motion prediction models; (3) Simulate evacuation scenarios based on real-time data and prediction results; (4) Simulate different evacuation strategies and evaluate evacuation smoothness; (5) Optimize evacuation strategies and conduct simulation optimization through actual evacuation drills.

2. The method for studying the relationship between obstacles in evacuation roads and smoothness of pedestrian evacuation according to claim 1 is characterized in that: The specific steps of real-time data preprocessing in step (1) are as follows: Step 1: Collect real-time data of the evacuation scene through sensors and cameras, then use Gaussian filtering to remove noise in each group of real-time signal data and smooth the data, then calculate the standard deviation of the signal data set, and detect and filter out abnormal data based on the calculated standard deviation, unify the data format, and then detect whether there are duplicate data records. If there are duplicate data, delete them; Step 2: Detect the missing values ​​in each group of data, mark the location of each missing value in the corresponding data, perform statistical and visual analysis on the missing values ​​in each group of data to obtain the distribution and impact range of the missing values, and replace the abnormal values ​​or missing values ​​with the average or median of the corresponding K groups of data points found by the KNN algorithm; Step 3: Extract a certain number of image frames from the real-time video data every second to cover the entire video, obtain the pixel values ​​of each group in each group of image frames, and then for each group of pixels, replace the value of the central pixel with the average value of the pixels around the pixel. Repeat the above process to process the entire image to remove the existing noise, watermark or distortion, and adjust the image size and resolution so that each group of images has the same size.

3. The method for studying the relationship between obstacles in evacuation roads and smoothness of pedestrian evacuation according to claim 2 is characterized in that: The specific steps of constructing the motion prediction model in step (2) are as follows: Step 1: Obtain historical data of evacuation scenarios, and extract feature information of each group of data after preprocessing each group of historical data. Then, create new features by randomly replacing or sampling the acquired feature data. Then, through feature scaling and standardization, make different features on the same scale, and randomly divide each group of collected feature data into training set and test set; Step 2: Create a motion prediction model and define the model loss function. Then select the stochastic gradient descent algorithm to adjust the model parameters to minimize the loss function. Divide the training set into small batches and use each batch to update the model weights. Repeat the forward propagation to calculate the output of the model, and then perform backpropagation to calculate the gradient and update the model parameters until all training sets are used up. Step 3: Use an independent test set to evaluate the performance of the trained model. Based on the model evaluation results, if the evaluation results do not meet the preset threshold, for each set of data, the loss value of the motion prediction model training results is calculated, and then the test set is replaced with another subset. The remaining subset is taken as the training set, and the loss value is calculated again until all data are predicted once. The combined parameters corresponding to the minimum loss value are selected as the optimal parameters in the data interval and the original parameters of the motion prediction model are replaced.

4. The method for studying the relationship between obstacles in evacuation roads and smoothness of pedestrian evacuation according to claim 3 is characterized in that: The specific calculation formula of the gradient descent algorithm described in step 2 is as follows: In the formula, θ represents the model parameters that need to be adjusted; α represents the learning rate; The specific calculation formula of the loss function in step 2 is as follows: In the formula, n represents the number of data; represents the actual observed value; Represents the predicted value of the model.

5. The method for studying the relationship between obstacles in evacuation roads and smoothness of pedestrian evacuation according to claim 3, characterized in that: The specific steps of the motion prediction model behavior prediction in step (2) are as follows: Step I: Linearly transform the preprocessed real-time data from different angles through multiple sets of attention heads, extract features that match the features used in model training, and fuse the outputs of multiple sets of attention heads to obtain the corresponding multimodal data set; Step II: Input each group of data in the multimodal dataset into the trained motion prediction model. The input data starts from the input layer of the motion prediction model and passes through each hidden layer of the model. Each hidden layer performs a linear transformation on the input data and transmits it layer by layer through the weights and activation functions between each layer. Then, the output layer outputs the probability distribution of one or more groups of behaviors. Step III: Select the behavior data with the highest probability as the final prediction result, and then explain and visualize the prediction results of the model for users to understand and analyze.

6. The method for studying the relationship between obstacles in evacuation roads and smoothness of pedestrian evacuation according to claim 5, characterized in that: The specific steps of evacuation scenario simulation in step (3) are as follows: Step 1: Use computer graphics technology to simulate the actual evacuation scenario, integrate the collected real-time data into the virtual environment, and ensure the time synchronization of each set of data. Then identify the pedestrian and obstacle information in the real-time data, and build a personnel behavior model and an obstacle model. Step ②: Integrate the prediction data generated by the motion prediction model into the virtual environment, and then start the simulation in the virtual environment. At the same time, the personnel and obstacle models will move in the virtual environment, and monitor the simulation results in real time. After the simulation is completed, record the data during the simulation.

7. The method for studying the relationship between obstacles in evacuation roads and smoothness of pedestrian evacuation according to claim 6, characterized in that: The specific steps of the evacuation smoothness assessment in step (4) are as follows: Step 1: Collect the paths, personnel behaviors, obstacles and evacuation strategy data in the simulation data, then extract the entity information in each set of recorded data and unify the format information of each set of data; Step 2: Collect various knowledge and information related to evacuation smoothness from the Internet and historical evacuation databases, classify, remove duplicates and filter the collected evacuation knowledge, and use NLP technology to identify and extract entities in the processed evacuation knowledge; Step 3: Extract the corresponding attributes of each entity from the relevant knowledge information, establish the relationship between entities, form the connection of the evacuation knowledge graph, process the entities, attributes and relationships into the corresponding graph structure in the form of triples, select a suitable graph database to store and manage the evacuation knowledge graph, and continuously update and maintain the evacuation knowledge graph; Step 4: Match the extracted entities with the entities in the evacuation knowledge graph, output the evaluation scores of each group of data with consistent matching results, and feed them back to the researchers for review.

Images