Urban health state estimation method based on deep learning network and multi-source heterogeneous data

Abstract

The invention discloses a city health state estimation method based on a deep learning network and multi-source heterogeneous data. The method comprises the following steps: carrying out preprocessing and feature processing on air quality data; preprocessing the traffic jam data, and processing related characteristics of the traffic jam data; integrating the results of the data feature extraction process; and reasoning and evaluating the health condition of the city. In order to understand the influence of urban health on the life, behavior and selection of urban residents, the invention provides a public health (HOTP) framework. The prediction result of the HOTP method is matched with the actual operation of the city. It shows that the method skillfully integrates metropolitan commuting information and air pollution data, and the overall health level of a city can be accurately evaluated.

Description

Technical Field

[0001] The present invention belongs to the field of urban transportation, and specifically relates to a method for estimating urban health status based on a deep learning network and multi-source heterogeneous data. Background Art

[0002] Smart cities typically use sensor infrastructure combined with data to monitor and detect in real time possible anomalies or events of interest in the city, such as traffic accidents, aviation accidents, and pollution. Sensor devices are becoming part of smart city infrastructure, which can be embedded in urban infrastructure (such as surveillance cameras), mobile devices (such as smartphones or mobile applications), or static devices (such as pollution monitoring stations) to provide services that improve public welfare and the quality of life of residents. These services involve aspects such as traffic management, environmental monitoring, smart transportation, and housekeeping services. Smart city systems usually generate a large amount of data, which allows us to study, understand, and simulate multiple parameters of the city during its operation, such as traffic patterns, crowd behavior, dynamics, and the diffusion of air pollution throughout the city.

[0003] Digital technologies are thus changing the way smart cities are planned and monitored. One of the challenging issues is how to assess the health of residents within an urban area by taking into account spatial and temporal factors, including factors such as traffic conditions and air pollution. By monitoring air pollution emissions and communicating this information to the public, it is possible to understand how the health of a city affects people’s lives, behaviors, and choices in an urban environment. For example, children, the elderly, or people who engage in outdoor sports (such as cycling or jogging) can benefit from this information by avoiding places with high pollution levels or congestion. At the same time, this information can also encourage people to use alternative means of transportation and avoid areas with stress and congestion that may cause traffic congestion. Therefore, identifying and leveraging heterogeneous data types in cities plays an important role in improving public welfare and enhancing public awareness.

[0004] Previous studies have analyzed various perspectives on urban life issues. For example, the criteria for determining the best location of new retail stores using geographic information considers density characteristics of different types and nearby areas, as well as user mobility criteria, including transitions between venues and mobile users from distant areas. There have been some past works focusing on inferring traffic information based on crowdsourcing schemes or ring sensor networks, as well as using urban data to assess the environmental footprint of cities. However, the limitations of these studies are that they focus on specific aspects of urban life issues (such as traffic or air pollution) without considering multiple types of urban data. Our goal is to use heterogeneous multi-source urban data, namely pollution and traffic data, to assess the health of cities. Traffic data are obtained from multiple large-scale sources, including Beijing Transportation Development Research Institute, Transportation Big Data Research Center of Tsinghua University Institute of Data Science, Saiwen Transportation Network, and Baidu Map Intelligent Transportation Joint Laboratory. These data may be noisy, missing values, and measurement errors. The data on urban air quality comes from the China National Environmental Monitoring Center. The Ministry of Ecology and Environment of China is responsible for collecting national air quality data and providing citizens with information about air pollution. Air quality is not fixed, it changes every hour, and even the air quality of each community can vary greatly. Therefore, using air pollution and traffic data to assess the health of a city is a challenging task. Summary of the invention

[0005] In order to solve the problems existing in the above-mentioned prior art, the present invention proposes a method for estimating the health status of a city based on a deep learning network and multi-source heterogeneous data. We propose a HOTP framework to identify the health status of a city by integrating heterogeneous sources of urban data. First, the air quality data is preprocessed and the data features are processed. Then we preprocess and process the traffic congestion data. Finally, the above two data feature extraction results are fused to judge the health status of the city by reasoning. The prediction results are in line with the actual urban operation rules, which shows that the HTOP method we proposed organically integrates the city's traffic data and air pollution data, and can accurately evaluate the health level of the entire city.

[0006] The present invention proposes a method for estimating urban health status based on a deep learning network and multi-source heterogeneous data, which specifically includes the following steps:

[0007] Step 1: Collect urban air quality data;

[0008] Step 2: Obtain the city traffic congestion situation;

[0009] Step 3: Preprocessing urban air quality data;

[0010] Step 4: Pre-processing of urban traffic congestion;

[0011] Step 5: Obtain air quality preprocessing data and urban traffic congestion preprocessing data, and integrate the two preprocessing data;

[0012] Step 6: Reason and assess the health of the city.

[0013] Furthermore, step 1 is specifically as follows:

[0014] Accurately collect air quality data of target cities, covering concentration data of key pollutants such as PM2.5, PM10, SO2, NO2, CO, O3, etc., to ensure the comprehensiveness and accuracy of the data.

[0015] The collected data is expressed in vector form:

[0016]

[0017] where c x represents the concentration of pollutant x.

[0018] Furthermore, step 2 is specifically as follows:

[0019] Systematically obtain traffic congestion data of target cities, including core traffic parameters such as road congestion index, traffic flow, average vehicle speed, etc., to comprehensively reflect the city’s traffic conditions.

[0020] Traffic congestion is represented by the following vector:

[0021] Traffic congestion data = (I 拥堵指数 , F 交通流量 , V 平均车速 )

[0022] Among them I 拥堵指数 、F 交通流量 、V 平均车速 They represent congestion index, traffic flow and average vehicle speed respectively.

[0023] Furthermore, step 3 is specifically as follows:

[0024] The air quality index is calculated by combining the concentration values ​​of various pollutants in the air. The air quality index of a single pollutant, namely the air quality sub-index (IAQI), is calculated by formula (1):

[0025]

[0026] Among them, IAQI is the air quality index of a single pollutant, C P is the mass concentration value of the pollutant item, BP H C P The high value of similar pollutant concentration limit, BPH C P The high value of similar pollutant concentration limit, BP L C P The lower value of the similar pollutant concentration limit, IAQI H For BP H Corresponding air quality index, IAQI L For BP L The corresponding air quality index. Then the air quality index (AQI) is

[0027] AQI=max{IAQI1, IAQI2, IAQI3,..., IAQI n}

[0028] Finally, the air quality index of area C at the current time t is divided by the threshold value marking dangerous pollution conditions to obtain the influencing factor for evaluating air quality.

[0029]

[0030] Furthermore, step 4 is specifically as follows:

[0031] Use traffic data to evaluate traffic conditions. Use the congestion index as the evaluation standard for traffic conditions. The congestion index is the ratio of actual travel time to smooth travel time. The larger the congestion index, the higher the degree of congestion. The congestion index can be calculated by the following formula:

[0032]

[0033] Where V f is the free flow speed of the road section, v is the current vehicle speed of the road section, and F is a variable parameter.

[0034] Furthermore, step 5 is specifically as follows:

[0035] Combining air pollution data with transportation data can make predictions about the health of a city:

[0036]

[0037] Among them, ω T is the traffic index weight parameter, ω P is the air index weight parameter, K T is the normalized value of traffic congestion index, K P is the normalized value of the air index. After obtaining accurate prediction values ​​of air pollution data and traffic congestion data, the prediction values ​​are integrated to predict the health status of the city.

[0038] Furthermore, step 6 is specifically as follows:

[0039] According to the different characteristics of traffic data and air data, the GBDT decision tree algorithm and the long short-term memory network LSTM algorithm are used to predict the future traffic congestion status and air pollution index. Four parameters, MAD (Median Absolute Deviation), MAE (Mean Absolute Error), RMSE (Root Mean Square Error), and STD (Standard Deviation), are selected to evaluate the accuracy of the prediction. Finally, the traffic congestion data and air pollution data are integrated to jointly evaluate the health status of the city.

[0040] Compared with the prior art, this patent has the following beneficial effects:

[0041] The present invention proposes a method for estimating the health status of a city based on a deep learning network and multi-source heterogeneous data. In view of the fact that current research focuses on analyzing data from a single source and the recent use of heterogeneous data sources without determining the two factors of pollution information and traffic data and their interaction to determine the health level of a city, the focus is on determining the health level of a city by combining pollution information with traffic data. In step one, we select the main pollution information according to the pollution level and pollution sources in different areas of the city. This is crucial to understanding the factors affecting urban pollution. In step two, we select data from a collective traffic monitoring network based on the characteristic that traffic conditions in different areas are different. The spatiotemporal distribution or statistical characteristics of these data are helpful in identifying typical traffic conditions and extracting information that is helpful for traffic planning. In steps three and four, we preprocess the acquired air quality data and urban traffic congestion conditions according to the needs of our method, and present these two data in digital form for further experiments. After preprocessing the data, the air pollution data and traffic congestion data are merged to jointly calculate the health status of a city. This innovative strategy breaks through the current practice of most scholars using only a single data source to infer traffic or pollution status. It organically integrates the city's traffic data and air pollution data, which can clearly understand the health level of the entire city and provide a basis for judging the excellence of the prediction algorithm. In addition, in order to more accurately judge the error of the prediction algorithm, four parameters, MAD (Median Absolute Deviation), MAE (Mean Absolute Error), RMSE (Root Mean Square Error), and STD (Standard Deviation), are selected to evaluate the accuracy of the prediction, further improving the reliability of the prediction algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a flow chart of a method according to an embodiment of the present invention.

[0043] Figure 2 These are the actual value and predicted value of the traffic congestion status in the embodiment of the present invention.

[0044] Figure 3 It is the actual value and predicted value of the air pollution index in the embodiment of the present invention.

[0045] Figure 4 This is the city health status prediction of the embodiment of the present invention. DETAILED DESCRIPTION

[0046] This embodiment provides a method for estimating the health status of an urban area based on a deep learning network and multi-source heterogeneous data, which specifically includes the following steps:

[0047] Step 1: Collect urban air quality data;

[0048] Accurately collect air quality data of target cities, covering concentration data of key pollutants such as PM2.5, PM10, SO2, NO2, CO, O3, etc., to ensure the comprehensiveness and accuracy of the data.

[0049] The collected data is expressed in vector form:

[0050]

[0051] where c x represents the concentration of pollutant x.

[0052] Step 2: Obtain the city traffic congestion situation;

[0053] Systematically obtain traffic congestion data of target cities, including core traffic parameters such as road congestion index, traffic flow, average vehicle speed, etc., to comprehensively reflect the city’s traffic conditions.

[0054] Traffic congestion is represented by the following vector:

[0055] Traffic congestion data = (I 拥堵指数 , F 交通流量 , V 平均车速 )

[0056] Among them I 拥堵指数 、F 交通流量 、V 平均车速 They represent congestion index, traffic flow and average vehicle speed respectively.

[0057] Step 3: Preprocessing urban air quality data;

[0058] The air quality index is calculated by combining the concentration values ​​of various pollutants in the air. The air quality index of a single pollutant, namely the air quality sub-index (IAQI), is calculated by formula (1):

[0059]

[0060] Among them, IAQI is the air quality index of a single pollutant, C P is the mass concentration value of the pollutant item, BP H C P The high value of similar pollutant concentration limit, BP H C P The high value of similar pollutant concentration limit, BP L C P The lower value of the similar pollutant concentration limit, IAQI H For BP H Corresponding air quality index, IAQI L For BP L The corresponding air quality index. Then the air quality index (AQI) is

[0061] AQI=max{IAQI1, IAQI2, IAQI3,..., IAQI n}

[0062] Finally, the air quality index of area C at the current time t is divided by the threshold value marking dangerous pollution conditions to obtain the influencing factor for evaluating air quality.

[0063]

[0064] Step 4: Pre-processing of urban traffic congestion;

[0065] Use traffic data to evaluate traffic conditions. Use the congestion index as the evaluation standard for traffic conditions. The congestion index is the ratio of actual travel time to smooth travel time. The larger the congestion index, the higher the degree of congestion. The congestion index can be calculated by the following formula:

[0066]

[0067] Where V f is the free flow speed of the road section, v is the current vehicle speed of the road section, and F is a variable parameter.

[0068] Step 5: Obtain air quality preprocessing data and urban traffic congestion preprocessing data, and integrate the two preprocessing data;

[0069] Combining air pollution data with transportation data can make predictions about the health of a city:

[0070]

[0071] Among them, ω T is the traffic index weight parameter, ω P is the air index weight parameter, K T is the normalized value of traffic congestion index, K P is the normalized value of the air index. After obtaining accurate prediction values ​​of air pollution data and traffic congestion data, the prediction values ​​are integrated to predict the health status of the city.

[0072] Step 6: Reasoning and assessing the health of the city;

[0073] According to the different characteristics of traffic data and air data, the GBDT decision tree algorithm and the long short-term memory network LSTM algorithm are used to predict the future traffic congestion status and air pollution index. Four parameters, MAD (Median Absolute Deviation), MAE (Mean Absolute Error), RMSE (Root Mean Square Error), and STD (Standard Deviation), are selected to evaluate the accuracy of the prediction. Finally, the traffic congestion data and air pollution data are integrated to jointly evaluate the health status of the city.

[0074] Gradient Boosting is a machine learning technique for regression and classification problems. The prediction model it produces is an integration of weak prediction models. For example, if a typical decision tree is used as a weak prediction model, then it is a gradient boosted tree (GBDT). Like other boosting methods, it builds models in a staged manner, but it generalizes the general boosting method by allowing optimization of any differentiable loss function. LSTM is a variant of a recurrent neural network (RNN) that is specifically designed to process sequence data. Traditional RNNs are prone to the problem of vanishing gradients when processing long sequences, making them difficult to train. LSTM introduces a gating mechanism that can effectively alleviate the problem of vanishing gradients, thereby being able to process longer sequence data. Moreover, compared to traditional RNNs, LSTM has better memory performance and can retain distant context information when processing sequence data.

[0075] The urban health status estimation method based on deep learning network and multi-source heterogeneous data includes a data preprocessing module: responsible for the collection, analysis and preprocessing of urban air quality data, as well as the acquisition, analysis and preprocessing of urban traffic congestion conditions; a multi-source data fusion module: adopts a multi-source data fusion method based on HOTP, including air quality sub-index, air quality index, and calculation of influencing factors and congestion index for evaluating air quality; an urban health status prediction module: calculates the predicted value of urban health status through the fused data.

[0076] First, the air quality data and urban traffic congestion of the target city are obtained, and the two types of data are analyzed and preprocessed respectively to calculate the air quality sub-index, air quality index, air quality impact factor, and traffic congestion index required by the prediction method; then the air pollution data and traffic congestion data are integrated to jointly calculate the health status of a city. This method can organically integrate the city's traffic data and air data, providing a basis for the evaluation of the subsequent urban health prediction algorithm. The GBDT decision tree algorithm is used to predict future traffic congestion. GBDT improves the prediction accuracy by integrating weak prediction models and optimizing the loss function. The long short-term memory network LSTM is used to predict future air pollution. As a variant of the recurrent neural network (RNN), LSTM effectively alleviates the gradient vanishing problem of the traditional RNN when processing long sequence data by introducing a gating mechanism, improves memory performance, and can better retain distant context information. Finally, the accuracy of the prediction is evaluated by four parameters: MAD, MAE, RMSE, and STD.

[0077] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.

Claims

1. A method for estimating urban health status based on deep learning network and multi-source heterogeneous data, characterized in that: The specific steps include: Step 1: Collect urban air quality data; Step 2: Obtain the city traffic congestion situation; Step 3: Preprocessing urban air quality data; Step 4: Pre-processing of urban traffic congestion; Step 5: Obtain air quality preprocessing data and urban traffic congestion preprocessing data, and integrate the two preprocessing data; Step 6: Reason and assess the health of the city.

Images