Multi-agent-based space-time activity analysis and prediction method for urban empty-nest elderly

Abstract

The invention discloses a multi-agent-based space-time activity analysis and prediction method for urban empty nesters. Comprising the steps of collecting extrinsic features of community residents and constructing a space-time activity sand table, identifying urban empty-nest elderly people and staying points thereof, constructing a typical space-time activity chain of the urban empty-nest elderly people, constructing a multi-agent model of the urban empty-nest elderly people, simulating and screening demands of the empty-nest elderly people, and generating a public service facility layout optimization scheme. And the requirements of the empty nesters are visualized. According to the method, high-precision resident extrinsic features and high-resolution spatial data are acquired, and a multi-agent model is adopted to simulate and screen the demands of the empty-nest elderly, so that full-coverage, precise and multi-dimensional identification of the activity demands of the empty-nest elderly under an urban scale is realized; and a reliable basis is provided for perfecting urban community public service facilities and improving the life quality of old people.

Description

Technical Field

[0001] The present invention relates to the field of urban planning, and in particular to a multi-agent-based spatiotemporal activity analysis and prediction method for urban empty-nesters. Background Art

[0002] With the acceleration of urbanization and the intensification of population aging, the number of empty-nesters in cities is growing rapidly, especially in first- and second-tier cities. As an elderly group that lives independently but is psychologically lonely, empty-nesters face many needs such as life care, health monitoring, and emotional companionship. However, there are widespread problems such as insufficient aging-friendly urban spaces and facilities, and the needs of empty-nesters are not addressed. The development of digital services has provided the possibility of solving the problem of empty-nesters, giving rise to the need for more refined and intelligent spatiotemporal activity analysis and prediction technologies, with the aim of providing targeted support for the multi-dimensional needs of empty-nesters through efficient technical means.

[0003] At present, there are certain limitations in the spatiotemporal activity analysis and prediction of empty-nest elderly people. In terms of research accuracy, traditional monitoring methods mostly rely on static data, such as family information, community registration forms, etc., which cannot reflect the dynamic behavior trajectory of empty-nest elderly people in real time, lacking timeliness and accuracy; in terms of research methods, traditional location-based service technologies can only obtain fragmented activity data of the elderly, making it difficult to form a complete activity trajectory; in terms of research conclusions, due to the lack of multi-dimensional interactive analysis capabilities, it is difficult to effectively combine the behavior trajectory of empty-nest elderly people with their health status, social interaction, psychological needs and other factors, resulting in the one-sidedness of the research results, and also limiting the reliability and effectiveness of the prediction results. Therefore, with the help of intelligent means, a spatiotemporal activity analysis and prediction method for urban empty-nest elderly people is proposed, which can improve the recognition accuracy while realizing the simulation and prediction of the needs under the interactive influence of multi-dimensional factors such as health and social interaction of the elderly, providing a reliable basis for improving urban community public service facilities and improving the quality of life of the elderly. Summary of the invention

[0004] In order to address the deficiencies mentioned in the above-mentioned background technology, the purpose of the present invention is to provide a multi-agent based method for spatiotemporal activity analysis and prediction of urban empty-nest elderly people. This method obtains high-precision residents' external characteristics and high-resolution spatial data, and adopts a multi-agent model to simulate and screen the needs of empty-nest elderly people, thereby realizing full coverage, precision and multi-dimensional identification of the activity needs of empty-nest elderly people at the urban scale, and providing a reliable basis for improving urban community public service facilities and improving the quality of life of the elderly.

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] A multi-agent-based spatiotemporal activity analysis and prediction method for urban empty-nesters includes the following steps:

[0007] Step 1: Collect the external characteristics of community residents and build a space-time activity sand table

[0008] High-definition monitoring equipment installed in the community records outdoor activity images of active people in real time to form a resident image database. Facial recognition is performed and associated with identity information to form a dataset of explicit characteristics of active people, including four types of labels: age, gender, marital status, and community number. LBS data generated by wearable smart bracelet devices is used to collect behavioral trajectory data of the elderly.

[0009] The urban community unit boundary data, urban road network data, land use data, building form data, business POI data and active crowd explicit feature data set are input into the cloud computing platform to automatically build a multi-source spatiotemporal database of urban communities. The data is then imported into the geographic information platform to build a spatiotemporal activity sandbox for urban community crowds.

[0010] Step 2: Identify urban empty-nesters and their destinations

[0011] The explicit feature dataset of the active population in step 1 is retrieved on the cloud computing platform, and the active population is clustered using the ST-DBSCAN technology to extract population attribute characteristics of 7 indicators in two dimensions, namely, entity attributes and activity attributes. Any 14 consecutive natural days are selected, and the population with less than two people traveling together every day is identified as urban single population, among which urban single population aged 65 and above is identified as urban empty-nest elderly.

[0012] The behavior trajectory data, land use data and business POI data of urban empty-nesters are extracted on the cloud computing platform to identify the residence, stay points and travel activity chains of empty-nesters. The location where an individual stays for more than or equal to 8 hours from 19:00 to 6:00 is defined as the residence; the center of the polygonal location where the behavior trajectory displacement is less than or equal to 100 meters within 15 minutes is marked as the stay point, and the spatial standard of travel activities is defined as a displacement distance greater than 100 meters within 15 minutes, and a trip is identified between two stops in different places.

[0013] Step 3: Construct a typical spatiotemporal activity chain of urban empty-nesters

[0014] Select any 14 consecutive natural days, with 15 minutes as a time step, and form 24*4 time steps per day. According to the land use type and POI type of the individual stay point in the current time step, the decision result chain of the five activities of empty nesters, including home H, medical treatment M, recreation G, shopping S, and cultural and sports E, is obtained.

[0015] The 96 time steps of 14 empty-nest elderly individuals in the past two weeks are connected in series to form a double-column spatiotemporal activity chain of activity code (h / m / g / s / e)-activity location (x, y). The 14 spatiotemporal activity chains of the individual in the past are simplified into a typical spatiotemporal activity chain. The highest frequency activity code of each time step is used as the activity code of the time step. If the code with the same frequency appears in a single time step, the activity consistent with the activity code of the previous time step is selected. The activity locations within 14 corresponding to the activity code are spatially clustered to obtain the activity location (X, Y) in a single time step. The 96 activity codes of the typical spatiotemporal activity chain are connected in series to complete the construction of the double-column typical spatiotemporal activity chain of activity code (H / M / G / S / E)-activity location (X, Y) of the empty-nest elderly individuals. For the elderly who have no activity tracks in the past three days, the generation of typical spatiotemporal activity chains is suspended and marked as elderly people to be visited.

[0016] Step 4: Build a multi-agent model for urban empty-nesters

[0017] Use Python language to build the multi-agent environmental factor layer of urban empty-nesters. The environmental factor layer includes city level, community transportation accessibility, density of health care facilities around the community, density of shopping facilities around the community, density of cultural and sports facilities around the community, and density of urban open space. Build the multi-agent layer of urban empty-nesters and import the population attribute characteristics of the 7 indicators in step 2.

[0018] The q-learning algorithm is used to map the multi-agents of empty-nesters in cities from the environment layer to the learning of activities to obtain reward signals. In each step, the multi-agents of empty-nesters in cities can select an action a from the finite set of activities H, M, G, S, and E in any state s. After the environment accepts the action, the probability of the accompanied state and the unaccompanied state changes. The utility function is divided into a utility curve based on the duration of the activity, a utility curve based on the density of health care facilities, a utility curve based on traffic accessibility, a utility curve based on the density of cultural and sports facilities, and a utility curve based on the density of shopping facilities. The multi-agent model of empty-nesters in cities is loaded onto the sandbox road network. In each time step, the q value of maintaining or changing the activity in the current state of each agent is compared, and the action with a high q value is the action of the next time step. The simulation results of 10 activity chains of urban empty-nest elderly multi-agents are fitted with the typical spatiotemporal activity chains of individuals in step three. If the activity code (H / M / G / S / E) and activity position (X, Y) meet more than 80% fitting rate, it indicates that the multi-agent model of empty-nest elderly is well constructed; otherwise, it returns to the q function iteration process to further improve the q value.

[0019] The multi-agent of urban empty-nest elderly people updates their activity status in the next time step. If they remain active for 24 hours, they will be marked as elderly people waiting to be visited.

[0020] The ordered Logit model is used to calculate the state transition probability of urban empty-nesters. The control variables include the age, gender, companionship or no companionship, and current residential community of urban empty-nesters. At the same time, an age segment is formed every 5 years, and a 10% weakening of the q value is superimposed. If the q value of a single multi-agent is lower than the average value of the next age segment, the agent is eliminated. The number of multi-agents is updated daily and compared with the explicit characteristic data set of urban community residents on that day in step 1. If the error rate is less than 10%, the multi-agent model of urban empty-nesters is considered to be successfully constructed, forming a dynamic simulation and prediction of the urban empty-nester population.

[0021] Step 5: Simulate and screen the needs of empty-nest elderly multi-agents

[0022] Based on the spatiotemporal activity decision of the multi-agent empty-nesters in the city in step 4, calculate the proportion of the accompanied state and the unaccompanied state in the process of improving the q value. If the contribution of the accompanied state to the improvement of the q value is higher than 50%, it indicates that the activity needs of the urban empty-nesters tend to be medical, recreational, shopping, and cultural and sports. Conversely, it indicates that the activity needs of the urban empty-nesters tend to be home. According to the activity preferences of the typical spatiotemporal activity chain in step 3, set the characteristic indicators of home needs, medical needs, travel needs, recreational needs, and shopping needs. Use the autoencoder technology to anchor the demand state in turn according to the demand type, output the multi-agent demand state that meets the requirements, and start the next category of demand state screening until all the set multi-agent activity demand simulations and screening are completed.

[0023] Step 6: Generate an optimized layout plan for public service facilities

[0024] According to the multi-agent activity demand categories in step five, the demand population of each category and its dynamic change trend are counted. The high-density clusters of multi-agent demand are identified for occasional and daily needs, and the locations of daily high-density clusters in the spatiotemporal activity sandbox are marked as the areas to be optimized for related activity facilities. The seven types of labels carried by each multi-agent in step two are mined to determine the consistency between the companionship status of the multi-agent and the use status of the facilities. Through the pathfinding analysis of the shortest distance between the actual road and the activity location, the safe paths and road congestion of walking, cycling and driving are obtained, and the optimization plans for transportation facilities, health care facilities, shopping facilities, cultural and sports facilities, and open space are generated.

[0025] Step 7: Visual display of optimization plan and output of empty-nest elderly people’s needs

[0026] The Dashboard information visualization platform is used to visualize the output results of various activity needs of urban empty-nesters in step 5. The empty-nesters are placed on the time-space activity sandbox according to the activity demand categories, the empty-nesters' population structure with different activity needs is calculated, and the community demand tendency type is output. The population density, age structure, required facilities, and spatial distribution data of empty-nesters who need to meet various activities are transmitted to the large screen displaying the activity needs of urban empty-nesters. The management personnel will implement feedback on the facility update in the order in which the areas to be optimized emerge, so as to realize interactive update and transformation decision-making and optimization of the elderly residential areas.

[0027] Furthermore, in the step one, a sand table of the spatiotemporal activities of urban community populations is constructed. Based on the latitude and longitude data and timestamp data of the elderly's activities collected by wearable smart bracelet devices and the urban community unit boundary data, urban road network data, land use data, building form data and business POI data collected by high-resolution panoramic cameras, the above data are uniformly input into the geographic information platform in shp format to construct a sand table of the spatiotemporal activities of urban community populations.

[0028] Furthermore, the population attribute characteristics of the two dimensions of natural attributes and social attributes in step 2, which include 7 indicators, are shown in Table 3:

[0029] Table 3 Crowd attribute characteristics

[0030]

[0031] Furthermore, the land use type of the individual stay point in the current time step in the step three, that is, if the stay point is located in Class R land, it is identified as a home activity; if the stay point is located in Class A5 land, it is identified as a medical activity; if the stay point is located in Class G land, it is identified as a recreational activity; if the stay point is located in Class B land, it is identified as a shopping activity; if the stay point is located in Class A2 or A4 land, it is identified as a cultural and sports activity.

[0032] Furthermore, in step 4, after the environment accepts the action, the probability of the accompanied state and the unaccompanied state changes, specifically with the following probability changes:

[0033] P[s=s′ / S,a]=δ(s,a,s′)

[0034] In the formula, s is the state of the current time step, a is the current action, s′ is the state of the next time step, and the probabilities of the accompanied state and the unaccompanied state are obtained;

[0035] The iterable q function is used to represent the probability of the agent's travel decision action at this moment:

[0036]

[0037] Where α is the learning rate and γ∈[0, 1] is the discount factor.

[0038] Furthermore, in step 4, the ordered Logit model is used to calculate the state transition probability of urban empty-nesters, and the specific formula is:

[0039]

[0040] Among them, Y i，t is the health status of empty-nest elderly i at the end of period t, Y i，t-1 is the health status of empty-nest elderly i at the end of the previous period.

[0041] Furthermore, the characteristic indicators of home needs, medical needs, travel needs, recreation needs, and shopping needs in step 5 include the demand characteristics of empty-nesters, including activity preference duration, activity preference location, activity companions, and activity preference frequency. Activity preference duration is the total time step of the highest frequency activity code other than residence in the typical spatiotemporal activity chain. Activity preference location is the activity location polygon connection of the highest frequency activity code other than residence in the typical spatiotemporal activity chain. Activity companions are the total number of multi-agents located in the multi-agent action buffer in the same time step. Activity preference frequency is the ratio of the total number of the highest frequency activity code other than residence in the typical spatiotemporal activity chain to the total number of this high frequency activity code.

[0042] Furthermore, in step 5, the autoencoder technology is used in turn according to the demand type to anchor the demand state, and the formula is:

[0043]

[0044] Where L represents the reconstruction error function, such as mean square error or cross entropy, and x i represents the input of the i-th sample, represents the output of the i-th sample, λ represents the regularization coefficient, and R(θ) represents the regularization term, such as L 1 or L 2 norm, etc., θ represents the model parameters.

[0045] Continue training until the normal distribution test P>0.05 is met, and the threshold is set to (μ±2σ). If the threshold interval is met, the corresponding multi-agent demand state is output, and the next category of demand state screening is started until all the set multi-agent activity demand simulations and screenings are completed.

[0046] Beneficial effects of the present invention:

[0047] 1. The present invention collects and analyzes the external characteristics of community residents, especially empty-nest elderly people, to achieve accurate identification and care improvement for the empty-nest elderly group. This helps governments, communities and non-governmental organizations to more accurately locate people in need of help, thereby providing more personalized and caring care services.

[0048] 2. The present invention reveals the spatiotemporal laws of the daily activities of empty-nest elderly people, such as travel habits, frequented places, activity time periods, etc., through the constructed spatiotemporal activity sandbox and typical spatiotemporal activity chains, which deepens the in-depth understanding of the spatiotemporal activity patterns of empty-nest elderly people; this in-depth understanding helps to understand the living conditions and needs of empty-nest elderly people, and provides a basis for formulating more scientific and reasonable public service facility plans.

[0049] 3. The present invention forms an intelligent prediction and early warning mechanism for the health status of empty-nesters. Based on the demand simulation and screening of the multi-agent model, it is possible to predict the future survival status trend and potential needs of empty-nesters, such as health services, emergency rescue, etc. This helps to establish an early warning mechanism and reduce service delays or lack of survival care caused by lonely living conditions.

[0050] 4. The present invention realizes decision-making support and scientific planning of public service facilities. According to the spatiotemporal activity analysis and demand forecast of empty-nesters, the generated public service facility layout optimization plan can more effectively allocate community resources, such as adding convenience stores, medical points, leisure facilities, etc., to ensure that empty-nesters can easily obtain the required services and improve their quality of life. The data analysis and prediction results provided by this method provide strong decision-making support for governments and social organizations in formulating aging policies, urban planning, community services, etc., and help to achieve more scientific, reasonable and humane social governance.

[0051] 5. The present invention realizes the visualization of demand and public participation. The visualization of the demand of empty-nesters not only enables decision makers to intuitively understand the problem, but also facilitates the display of service improvement results to the public, stimulating social attention and support. At the same time, the improvement of public participation can also promote the emergence of more innovative solutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 is a flow chart of the method of the present invention;

[0053] Figure 2 It is the construction of a time and space activity sandbox;

[0054] Figure 3 Screening for empty-nest elderly people’s activity needs;

[0055] Figure 4 Visualize the areas to be optimized for activity facilities. DETAILED DESCRIPTION

[0056] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0057] A multi-agent-based spatiotemporal activity analysis and prediction method for urban empty-nesters. Figure 1-4 As shown, the following steps are included:

[0058] 1. Collect the explicit features of community residents and build a space-time activity sandbox. Use high-definition monitoring equipment installed in the community to record outdoor activity images of active people in real time to form a resident image database. Perform facial recognition and associate it with identity information to form a dataset of explicit features of active people, including four types of labels: age, gender, marital status, and community number. Use the LBS data generated by wearable smart bracelet devices to collect behavioral trajectory data of the elderly.

[0059] The urban community unit boundary data, urban road network data, land use data, building form data, business POI data and active crowd explicit feature data set are input into the cloud computing platform to automatically build a multi-source spatiotemporal database of urban communities. The data is then imported into the geographic information platform to build a spatiotemporal activity sandbox for urban community crowds.

[0060] The data fields of the explicit characteristics of the active population are shown in Table 1:

[0061] Table 1 Data fields of explicit characteristics of active population

[0062]

[0063] The LBS data fields are shown in Table 2:

[0064] Table 1 LBS data fields

[0065]

[0066]

[0067] The construction of the urban community population spatiotemporal activity sandbox is based on the latitude and longitude data and timestamp data of the elderly's activities collected by wearable smart bracelet devices and the urban community unit boundary data, urban road network data, land use data, building form data and business POI data collected by high-resolution panoramic cameras. The above data are uniformly input into the geographic information platform in shp format to construct the urban community population spatiotemporal activity sandbox.

[0068] 2. Identify urban empty-nesters and their stay points. Retrieve the explicit feature dataset of the active population in step 1 on the cloud computing platform, cluster the active population using the ST-DBSCAN technology, extract the population attribute features of 7 indicators in two dimensions, namely, entity attributes and activity attributes, and select any 14 consecutive natural days to identify the population with less than two people traveling together every day as the urban single population, among which the urban single population aged 65 and above is identified as the urban empty-nesters.

[0069] The behavior trajectory data, land use data and business POI data of urban empty-nesters are extracted on the cloud computing platform to identify the residence, stay points and travel activity chains of empty-nesters. The location where an individual stays for more than or equal to 8 hours from 19:00 to 6:00 is defined as the residence; the center of the polygonal location where the behavior trajectory displacement is less than or equal to 100 meters within 15 minutes is marked as the stay point, and the spatial standard of travel activities is defined as a displacement distance greater than 100 meters within 15 minutes, and a trip is identified between two stops in different places.

[0070] The crowd attribute characteristics of the 7 indicators in the two dimensions of ontology attribute and activity attribute are shown in Table 3:

[0071] Table 3 Crowd attribute characteristics

[0072]

[0073]

[0074] 3. Construct a typical spatiotemporal activity chain of urban empty-nesters. Select any 14 consecutive natural days, with 15 minutes as a time step, and form 24*4 time steps per day. According to the land use type and POI type of the individual stay point in the current time step, the decision result chain of the five activities of empty-nesters, namely, home H, medical treatment M, recreation G, shopping S, and cultural and sports E, is obtained.

[0075] The 96 time steps of 14 empty-nest elderly individuals in the past two weeks are connected in series to form a double-column spatiotemporal activity chain of activity code (h / m / g / s / e)-activity location (x, y). The 14 spatiotemporal activity chains of the individual in the past are simplified into a typical spatiotemporal activity chain. The highest frequency activity code of each time step is used as the activity code of the time step. If the code with the same frequency appears in a single time step, the activity consistent with the activity code of the previous time step is selected. The activity locations within 14 corresponding to the activity code are spatially clustered to obtain the activity location (X, Y) in a single time step. The 96 activity codes of the typical spatiotemporal activity chain are connected in series to complete the construction of the double-column typical spatiotemporal activity chain of activity code (H / M / G / S / E)-activity location (X, Y) of the empty-nest elderly individuals. For the elderly who have no activity tracks in the past three days, the generation of typical spatiotemporal activity chains is suspended and marked as elderly people to be visited.

[0076] Among them, the land use type of the individual stay point in the current time step, that is, if the stay point is located in Class R land, it is identified as a home activity; if the stay point is located in Class A5 land, it is identified as a medical activity; if the stay point is located in Class G land, it is identified as a recreational activity; if the stay point is located in Class B land, it is identified as a shopping activity; if the stay point is located in Class A2 or A4 land, it is identified as a cultural and sports activity.

[0077] 4. Build a multi-agent model for urban empty-nesters. Use Python to build a multi-agent environmental factor layer for urban empty-nesters. The environmental factor layer includes city level, community transportation accessibility, density of health care facilities around the community, density of shopping facilities around the community, density of cultural and sports facilities around the community, and density of urban open spaces. Build a multi-agent layer for urban empty-nesters and import the population attribute characteristics of the 7 indicators in step 2.

[0078] The q-learning algorithm is used to map the multi-agents of empty-nesters in cities from the environment layer to the learning of activities to obtain reward signals. In each step, the multi-agents of empty-nesters in cities can select an action a from the finite set of activities H, M, G, S, and E in any state s. After the environment accepts the action, the probability of the accompanied state and the unaccompanied state changes. The utility function is divided into a utility curve based on the duration of the activity, a utility curve based on the density of health care facilities, a utility curve based on traffic accessibility, a utility curve based on the density of cultural and sports facilities, and a utility curve based on the density of shopping facilities. The multi-agent model of empty-nesters in cities is loaded onto the sandbox road network. In each time step, the q value of maintaining or changing the activity in the current state of each agent is compared, and the action with a high q value is the action of the next time step. The simulation results of 10 activity chains of urban empty-nest elderly multi-agents are fitted with the typical spatiotemporal activity chains of individuals in step three. If the activity code (H / M / G / S / E) and activity position (X, Y) meet more than 80% fitting rate, it indicates that the multi-agent model of empty-nest elderly is well constructed; otherwise, it returns to the q function iteration process to further improve the q value.

[0079] The multi-agent of urban empty-nest elderly people updates their activity status in the next time step. If they remain active for 24 hours, they will be marked as elderly people waiting to be visited.

[0080] The ordered Logit model is used to calculate the state transition probability of urban empty-nesters. The control variables include the age, gender, companionship or no companionship, and current residential community of urban empty-nesters. At the same time, an age segment is formed every 5 years, and a 10% weakening of the q value is superimposed. If the q value of a single multi-agent is lower than the average value of the next age segment, the agent is eliminated. The number of multi-agents is updated daily and compared with the explicit characteristic data set of urban community residents on that day in step 1. If the error rate is less than 10%, the multi-agent model of urban empty-nesters is considered to be successfully constructed, forming a dynamic simulation and prediction of the urban empty-nester population.

[0081] Among them, the probability of the environment changing to the accompanied state and the unaccompanied state after accepting the action is as follows:

[0082] P[s=s′ / s,a]=δ(s,a,s′)

[0083] In the formula, s is the state of the current time step, a is the current action, s′ is the state of the next time step, and the probabilities of the accompanied state and the unaccompanied state are obtained;

[0084] The iterable q function is used to represent the probability of the agent's travel decision action at this moment:

[0085]

[0086] Where α is the learning rate and γ∈[0, 1] is the discount factor.

[0087] Among them, in the step 4, the ordered Logit model is used to calculate the state transition probability of urban empty nesters, and the specific formula is:

[0088]

[0089] Among them, r i，t is the health status of empty-nest elderly i at the end of period t, Y i，t-1 is the health status of empty-nest elderly i at the end of the previous period.

[0090] 5. Simulation and screening of the needs of multi-agents of empty-nest elderly people. Based on the spatiotemporal activity decisions of multi-agents of urban empty-nest elderly people, calculate the proportion of the accompanied state and the unaccompanied state in the process of improving the q value. If the contribution of the companionship state to the increase in the q value is higher than 50%, it means that the activity needs of urban empty-nest elderly people tend to be medical, recreation, shopping, and culture and sports. Otherwise, it means that the activity needs of urban empty-nest elderly people tend to be at home. According to the activity preferences of the typical spatiotemporal activity chain in step three, set the characteristic indicators of home needs, medical needs, travel needs, recreation needs, and shopping needs. Use the autoencoder technology to anchor the demand state in turn according to the demand type, output the multi-agent demand state that meets the requirements, and start the demand state screening of the next category until all the set multi-agent activity demand simulations and screening are completed.

[0091] Among them, the characteristic indicators of home needs, medical needs, travel needs, recreation needs, and shopping needs, that is, the demand characteristics of empty-nesters, including activity preference duration, activity preference location, activity companions, and activity preference frequency. Activity preference duration is the total time step of the highest frequency activity code except residence in the typical spatiotemporal activity chain. Activity preference location is the activity location polygon connection of the highest frequency activity code except residence in the typical spatiotemporal activity chain. Activity companions are the total number of multi-agents located in the multi-agent action buffer in the same time step. Activity preference frequency is the ratio of the total number of the highest frequency activity code except residence in the typical spatiotemporal activity chain to the total number of this high frequency activity code.

[0092] Among them, the autoencoder technology is used in turn according to the demand type to perform data dimension reduction, and the formula is:

[0093]

[0094] Where L represents the reconstruction error function, such as mean square error or cross entropy, and x i represents the input of the i-th sample, represents the output of the i-th sample, λ represents the regularization coefficient, and R(θ) represents the regularization term, such as L 1 or L2 norm, etc., θ represents the model parameters.

[0095] Continue training until the normal distribution test P>0.05 is met, and the threshold is set to (μ±2σ). If the threshold interval is met, the corresponding multi-agent demand state is output, and the next category of demand state screening is started until all the set multi-agent activity demand simulations and screenings are completed.

[0096] 6. Generate an optimization plan for the layout of public service facilities. According to the multi-agent activity demand categories in step five, count the demand population in each category and its dynamic change trend. Identify the occasional and daily high-density clusters of multi-agent demand one by one, mark the location of the daily high-density clusters in the space-time activity sandbox as the area to be optimized for the relevant activity facilities, and mine the 7 types of labels carried by each multi-agent in step two to determine the consistency between the companionship status of the multi-agent and the use status of the facilities. Through the path-finding analysis of the shortest distance between the actual road and the activity location, the safe paths and road congestion for walking, cycling and driving are obtained, and the optimization plan for transportation facilities, health care facilities, shopping facilities, cultural and sports facilities, and open space is generated.

[0097] 7. Visual display of optimization plans and output of empty-nest elderly people’s needs. Use the Dashboard information visualization platform to visualize the output results of various activity needs of urban empty-nest elderly people, place empty-nest elderly people on the time-space activity sandbox according to the activity demand categories, calculate the empty-nest elderly population structure with different activity needs, and output the community demand tendency type. The population density, age structure, required facilities, and spatial distribution data of empty-nest elderly people who need to be satisfied by various activities are transmitted to the large screen displaying the activity needs of urban empty-nest elderly people. Management personnel will implement feedback on facility updates in order of the emergence of the areas to be optimized, and realize interactive renewal and transformation decisions and optimization of elderly residential areas.

[0098] Example

[0099] The technical solution of the present invention will be described in detail below by taking the renovation of facilities in Nanjing Yuejianglou Community as an example.

[0100] (1) Taking Nanjing Yuejianglou Community as the target community, we collected the external characteristics of Yuejianglou community residents and constructed a spatiotemporal activity sand table, including:

[0101] (1.1) High-definition monitoring equipment installed in the Yuejianglou community records the outdoor activity images of active people in real time to form a resident image database. Facial recognition is performed and associated with identity information to form a dataset of explicit characteristics of active people, including four types of labels: age, gender, marital status, and community number. LBS data generated by wearable smart bracelet devices is used to collect behavioral trajectory data of the elderly.

[0102] (1.2) Input the unit boundary data of Yuejianglou community, Nanjing road network data, Nanjing land use data, building form data, business POI data and active crowd explicit feature data set into the cloud computing platform to automatically build a multi-source spatiotemporal database of urban communities, and import the data into the geographic information platform to build a spatiotemporal activity sandbox of the Yuejianglou community in Nanjing.

[0103] (2) Identify empty-nest elderly people in Nanjing Yuejianglou Community and their places of stay, including:

[0104] (2.1) The explicit feature dataset of the active population in step 1 is retrieved on the cloud computing platform, and the active population is clustered using the ST-DBSCAN technology to extract the population attribute characteristics of 7 indicators in two dimensions, namely, entity attributes and activity attributes. Any 14 consecutive natural days are selected, and the population with less than two people traveling together every day is identified as the single population in Yuejianglou Community, among which the urban single population aged 65 and above is identified as the empty-nest elderly in Yuejianglou Community.

[0105] (2.2) Extract the behavior trajectory data, land use data and business POI data of the empty-nest elderly in Yuejianglou Community on the cloud computing platform to identify the residence, stay point and travel activity chain of the empty-nest elderly. Define the location where the individual stays for more than or equal to 8 hours from 19:00 to 6:00 as the residence; mark the center of the polygon location where the behavior trajectory displacement is less than or equal to 100 meters within 15 minutes as the stay point; the spatial standard of travel activities is defined as the displacement distance greater than 100 meters within 15 minutes, and the interval between two stops at different places is identified as a trip.

[0106] (3) Construct a typical spatiotemporal activity chain for empty-nest elderly people in Yuejianglou Community, including:

[0107] (3.1) Select any 14 consecutive natural days, with 15 minutes as a time step, and form 24*4 time steps per day. According to the land use type and POI type of the individual stay point in the current time step, the decision result chain of the empty-nesters in Yuejianglou Community for the five activities of home H, medical treatment M, recreation G, shopping S, and cultural and sports E is obtained.

[0108] (3.2) The 96 time steps of 14 empty-nest elderly individuals in Yuejianglou community in the past two weeks are connected to form a double-column spatiotemporal activity chain of activity code (h / m / g / s / e)-activity location (x, y). The 14 spatiotemporal activity chains of the individual in the past are simplified into a typical spatiotemporal activity chain. The highest frequency activity code of each time step is used as the activity code of the time step. If the code with the same frequency appears in a single time step, the activity consistent with the activity code of the previous time step is selected. The activity locations within 14 corresponding to the activity code are spatially clustered to obtain the activity location (X, Y) in a single time step. The 96 activity codes of the typical spatiotemporal activity chain are connected to complete the construction of the double-column typical spatiotemporal activity chain of activity code (H / M / G / S / E)-activity location (X, Y) of the empty-nest elderly individuals. For the elderly who have no activity tracks in the past three days, the generation of the typical spatiotemporal activity chain is suspended and they are marked as elderly people to be visited. The staff of Yuejianglou community will go to their residence to inquire about the situation.

[0109] (4) Construct a multi-agent model for empty-nest elderly people in Yuejianglou Community, including:

[0110] (4.1) Use Python to build a multi-agent environmental factor layer for empty-nest elderly people in Yuejianglou Community. The environmental factor layer includes city level, community transportation accessibility, density of health care facilities around the community, density of shopping facilities around the community, density of cultural and sports facilities around the community, and density of urban open space. Build a multi-agent layer for empty-nest elderly people in Yuejianglou Community and import the population attribute characteristics of the 7 indicators in step 2.

[0111] (4.2) The q-learning algorithm is used to map the multi-agents of empty-nesters in cities from the environment layer to the learning of activities to obtain reward signals. At each step, the multi-agents of empty-nesters in cities can select an action a from the finite set of activities H, M, G, S, and E in any state s. After the environment accepts the action, the probability of the accompanied state and the unaccompanied state changes. The utility function is divided into a utility curve based on the duration of the activity, a utility curve based on the density of health care facilities, a utility curve based on traffic accessibility, a utility curve based on the density of cultural and sports facilities, and a utility curve based on the density of shopping facilities. The multi-agent model of empty-nesters in cities is loaded onto the sandbox road network. At each time step, the q value of maintaining or changing the activity in the current state of each agent is compared, and the action with a high q value is the action of the next time step. The simulation results of 10 activity chains of urban empty-nest elderly multi-agents are fitted with the typical spatiotemporal activity chains of individuals in step three. If the activity code (H / M / G / S / E) and activity position (X, Y) meet more than 80% fitting rate, it indicates that the multi-agent model of empty-nest elderly is well constructed; otherwise, it returns to the q function iteration process to further improve the q value.

[0112] (4.3) The multi-agent of the urban empty-nest elderly updates its activity status at the next time step. If the activity remains active for 24 hours, it will be marked as an elderly person waiting to be visited, and the staff of Yuejianglou Community will go to the residence to inquire about the situation.

[0113] (4.4) The ordered logit model is used to calculate the status transition probability of urban empty-nesters. The control variables include the age, gender, companionship status or no companionship status, and current residential community of urban empty-nesters. At the same time, an age segment is formed every 5 years, and a 10% reduction in the q value is superimposed. If the q value of a single multi-agent is lower than the average value of the next age segment, the agent is eliminated. The number of multi-agents is updated daily and compared with the data set of explicit characteristics of urban community residents on that day in step 1. If the error rate is less than 10%, the multi-agent model of urban empty-nesters is considered to be successfully constructed, forming a dynamic simulation and prediction of the urban empty-nester population.

[0114] (5) Simulation and screening of the needs of multiple intelligent agents for empty-nest elderly people in Yuejianglou Community, including:

[0115] Based on the multi-agent spatiotemporal activity decision of the empty-nest elderly in Yuejianglou Community in step 4, calculate the proportion of the accompanied state and the unaccompanied state in the process of improving the q value. If the contribution of the accompanied state to the improvement of the q value is higher than 50%, it indicates that the activity needs of the urban empty-nest elderly tend to be medical, recreational, shopping, and cultural and sports. Conversely, it indicates that the activity needs of the urban empty-nest elderly tend to be at home. According to the activity preferences of the typical spatiotemporal activity chain in step 3, set the characteristic indicators of home needs, medical needs, travel needs, recreational needs, and shopping needs. Use the autoencoder technology to anchor the demand state in turn according to the demand type, output the multi-agent demand state that meets the requirements, and start the next category of demand state screening until all the set multi-agent activity demand simulations and screening are completed.

[0116] (6) Generate an optimization plan for the layout of public service facilities in Yuejianglou Community, including:

[0117] According to the multi-agent activity demand categories of empty-nest elderly people in Yuejianglou Community in step 5, the demand population of each category and its dynamic change trend are counted. The high-density clusters of multi-agent demand are identified for occasional and daily situations, and the locations of daily high-density clusters in the spatiotemporal activity sandbox are marked as the areas to be optimized for related activity facilities. The seven types of labels carried by each multi-agent in step 2 are mined to determine the consistency between the companionship status of the multi-agent and the use status of the facilities. Through the pathfinding analysis of the shortest distance between the actual road and the activity location, the safe paths and road congestion of walking, cycling and driving are obtained, and the optimization plans for transportation facilities, health care facilities, shopping facilities, cultural and sports facilities, and open space in Yuejianglou Community are generated.

[0118] (7) Visual display of optimization plan and output of needs of empty-nest elderly in Yuejianglou community, including:

[0119] The Dashboard information visualization platform is used to visualize the output results of various activity needs of empty-nesters in Yuejianglou Community in step 5. Empty-nesters in Yuejianglou Community are placed on the time-space activity sandbox according to the activity demand categories, and the empty-nesters population structure with different activity needs is calculated to output the community demand tendency type. The population density, age structure, required facilities, and spatial distribution data of empty-nesters who need to meet various activities are transmitted to the large screen displaying the activity needs of empty-nesters in the city. The management personnel will implement feedback on the facility update in the order in which the areas to be optimized appear, so as to realize interactive update and transformation decision-making and optimization of Yuejianglou Community.

[0120] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. A multi-agent-based spatiotemporal activity analysis and prediction method for urban empty-nesters, characterized by: The steps include: Step 1: Collect the external characteristics of community residents and build a space-time activity sand table High-definition surveillance equipment installed in the community records outdoor activity images of people in real time to form a resident image database; Perform facial recognition and associate it with identity information to form a dataset of explicit characteristics of the active population, including four types of labels: age, gender, marital status, and community number. Use the LBS data generated by wearable smart bracelet devices to collect behavioral trajectory data of the elderly. Input the urban community unit boundary data, urban road network data, land use data, building form data, business POI data and active crowd explicit feature data set into the cloud computing platform to automatically build a multi-source spatiotemporal database of urban communities, and import the data into the geographic information platform to build a spatiotemporal activity sandbox for urban community crowds; Step 2: Identify urban empty-nesters and their destinations The explicit feature dataset of the active population in step 1 is retrieved on the cloud computing platform, and the active population is clustered using the ST-DBSCAN method to extract the population attribute characteristics of 7 indicators in two dimensions, namely, entity attributes and activity attributes. Any 14 consecutive natural days are selected, and the population with less than two people traveling together every day is identified as the urban single population, among which the urban single population aged 65 and above is identified as the urban empty-nest elderly; The behavior trajectory data, land use data and business POI data of urban empty-nesters are extracted on the cloud computing platform to identify the residence, stay points and travel activity chains of empty-nesters; the location where an individual stays for more than or equal to 8 hours from 19:00 to 6:00 is defined as the residence; the center of the polygonal location where the behavior trajectory displacement is less than or equal to 100 meters within 15 minutes is marked as the stay point, and the spatial standard of travel activities is defined as a displacement distance greater than 100 meters within 15 minutes, and the period between two stays in different places is identified as a trip; Step 3: Construct a typical spatiotemporal activity chain of urban empty-nesters Select any 14 consecutive natural days, with 15 minutes as a time step, and form 24*4 time steps per day; according to the land use type and POI type of the individual stay point in the current time step, the decision result chain of the five activities of home H, medical treatment M, recreation G, shopping S, and cultural and sports E for the empty-nest elderly is obtained; The 96 time steps of 14 empty-nest elderly individuals in the past two weeks are connected in series to form a double-column spatiotemporal activity chain of activity code (h / m / g / s / e)-activity position (x, y); the 14 spatiotemporal activity chains of the individual in the past are simplified into a typical spatiotemporal activity chain; the highest frequency activity code of each time step is used as the activity code of the time step. If the code with the same frequency appears in a single time step, the activity consistent with the activity code of the previous time step is selected; the activity positions within 14 corresponding to the activity code are spatially clustered to obtain the activity position (X, Y) in a single time step, and the 96 activity codes of the typical spatiotemporal activity chain are connected in series to complete the construction of the double-column typical spatiotemporal activity chain of activity code (H / M / G / S / E)-activity position (X, Y) of the empty-nest elderly individuals; for the elderly who have no activity tracks in the past three days, the generation of typical spatiotemporal activity chains is suspended and they are marked as elderly people to be visited; Step 4: Build a multi-agent model for urban empty-nesters Use Python language to build a multi-agent environmental factor layer for urban empty-nesters. The environmental factor layer includes city level, community transportation accessibility, density of health care facilities around the community, density of shopping facilities around the community, density of cultural and sports facilities around the community, and density of urban open space; build a multi-agent layer for urban empty-nesters and import the population attribute characteristics of the 7 indicators in step 2; The Q-learning algorithm is used to map the multi-agents of empty-nesters in cities from the environment layer to the learning of activities to obtain reward signals; the multi-agents of empty-nesters in cities can select an action a from the finite activity set H, M, G, S, E at each step in any state s, and the probability of the accompanied state and the unaccompanied state changes after the environment accepts the action; the utility function is divided into utility curves based on activity duration, utility curves based on health care facility density, utility curves based on traffic accessibility, utility curves based on cultural and sports facility density, and utility curves based on shopping facility density; the urban empty-nesters The multi-agent model of the elderly is loaded onto the sandbox road network. In each time step, the q value of each agent in the current state of maintaining or changing activities is compared, and the action with the higher q value is the action of the next time step; the activity chain simulation results of 10 times of urban empty-nest elderly multi-agents are fitted with the typical spatiotemporal activity chain of individuals in step 3. If the activity code (H / M / G / S / E) and activity position (X, Y) meet more than 80% fitting rate, it indicates that the multi-agent model of empty-nest elderly is well constructed; otherwise, it returns to the q function iteration process to further improve the q value; The multi-agent of the empty-nest elderly in the city updates its activity status in the next time step. If it remains active for 24 hours, it will be marked as an elderly person waiting to be visited; The ordered Logit model is used to calculate the status transition probability of urban empty-nesters. The control variables include the age, gender, companionship status or no companionship status, and current residential community of urban empty-nesters. At the same time, an age segment is formed every 5 years, and a 10% weakening of the q value is superimposed. If the q value of a single multi-agent is lower than the average value of the next age segment, the agent is eliminated. The number of multi-agents is updated daily and compared with the explicit characteristic data set of urban community residents on that day in step 1. If the error rate is less than 10%, the multi-agent model of urban empty-nesters is considered to be successfully constructed, forming a dynamic simulation and prediction of the urban empty-nester population. Step 5: Simulate and screen the needs of empty-nest elderly multi-agents Based on the spatiotemporal activity decision-making of the multi-agent empty-nesters in the city in step 4, calculate the proportion of the accompanied state and the unaccompanied state in the process of improving the q value. If the contribution of the accompanied state to the improvement of the q value is higher than 50%, it indicates that the activity needs of the urban empty-nesters tend to be medical, recreational, shopping, and cultural and sports. Otherwise, it indicates that the activity needs of the urban empty-nesters tend to be at home; according to the activity preferences of the typical spatiotemporal activity chain in step 3, set the characteristic indicators of home needs, medical needs, travel needs, recreational needs, and shopping needs, and use the autoencoder technology to anchor the demand state in turn according to the demand type, output the multi-agent demand state that meets the requirements, and start the next category of demand state screening until all the set multi-agent activity demand simulations and screening are completed; Step 6: Generate an optimized layout plan for public service facilities According to the multi-agent activity demand categories in step five, count the demand population of each category and its dynamic change trend; identify the occasional and daily high-density clusters of multi-agent demand one by one, mark the location of the daily high-density clusters in the spatiotemporal activity sandbox as the area to be optimized for the relevant activity facilities, and mine the 7 types of labels carried by each multi-agent in step two to determine the consistency between the multi-agent's companionship status and the facility usage status; obtain the safe paths and road congestion for walking, cycling and driving through the path-finding analysis of the shortest distance between the actual road and the activity location, and generate optimization plans for transportation facilities, health care facilities, shopping facilities, cultural and sports facilities, and open space; Step 7: Visual display of optimization plan and output of empty-nest elderly people’s needs Use the Dashboard information visualization platform to visualize the output results of various activity needs of urban empty-nest elderly in step 5. Place the empty-nest elderly on the time-space activity sandbox according to the activity demand category, calculate the population structure of empty-nest elderly with different activity needs, and output the community demand tendency type; transmit the population density, age structure, demand facilities, and spatial distribution data of empty-nest elderly who need to be satisfied by various activities to the large screen displaying the activity needs of urban empty-nest elderly. Management personnel will implement feedback on facility updates in order of the emergence of the areas to be optimized, so as to realize interactive renewal and transformation decision-making and optimization of elderly residential areas.

2. The multi-agent-based spatiotemporal activity analysis and prediction method for urban empty-nesters according to claim 1 is characterized in that: In the step 1, a sand table of the spatiotemporal activities of people in urban communities is constructed. Based on the latitude and longitude data and timestamp data of the activities of the elderly collected by wearable smart bracelet devices and the urban community unit boundary data, urban road network data, land use data, building form data and business POI data collected by high-resolution panoramic cameras, the above data are uniformly input into the geographic information platform in shp format to construct a sand table of the spatiotemporal activities of people in urban communities.

3. The multi-agent-based spatiotemporal activity analysis and prediction method for urban empty-nesters according to claim 2 is characterized in that: The crowd attribute characteristics of the two dimensions of natural attributes and social attributes in step 2, totaling 7 indicators, are shown in Table 3: Table 3 Crowd attribute characteristics 4. The multi-agent-based spatiotemporal activity analysis and prediction method for urban empty-nesters according to claim 3 is characterized in that: The land use type of the individual stay point in the current time step in the step three, that is, if the stay point is located in Class R land, it is identified as a home activity; if the stay point is located in Class A5 land, it is identified as a medical activity; if the stay point is located in Class G land, it is identified as a recreational activity; if the stay point is located in Class B land, it is identified as a shopping activity; if the stay point is located in Class A2 or A4 land, it is identified as a cultural and sports activity.

5. The multi-agent-based spatiotemporal activity analysis and prediction method for urban empty-nesters according to claim 4 is characterized in that: In step 4, the probability of the environment being in the accompanied state and the unaccompanied state after accepting the action changes, specifically with the following probability changes: P[s=s' / s,a]=δ(s,a,s′) In the formula, s is the state of the current time step, a is the current action, s′ is the state of the next time step, and the probabilities of the accompanied state and the unaccompanied state are obtained; The iterable q function is used to represent the probability of the agent's travel decision action at this moment: Where α is the learning rate and γ∈[0,1] is the discount factor.

6. The multi-agent-based spatiotemporal activity analysis and prediction method for urban empty-nesters according to claim 5 is characterized in that: In step 4, the ordered Logit model is used to calculate the state transition probability of urban empty-nesters. The specific formula is: Among them, Y i,t is the health status of empty-nest elderly i at the end of period t, Y i,t-1 is the health status of empty-nest elderly i at the end of the previous period.

7. The multi-agent-based spatiotemporal activity analysis and prediction method for urban empty-nesters according to claim 6 is characterized in that: The characteristic indicators of home needs, medical needs, travel needs, recreation needs, and shopping needs in step 5, namely, the demand characteristics of empty-nesters including preferred activity duration, preferred activity location, activity companions, and preferred activity frequency; The activity preference duration is the total time step of the highest frequency activity code except residence in the typical space-time activity chain; the activity preference position is the activity position polygon connection of the highest frequency activity code except residence in the typical space-time activity chain; the activity companions are the total number of multi-agents located in the multi-agent action buffer in the same time step; the activity preference frequency is the ratio of the total number of the highest frequency activity code except residence in the typical space-time activity chain to the total number of this high frequency activity code.

8. The multi-agent-based spatiotemporal activity analysis and prediction method for urban empty-nesters according to claim 7 is characterized in that: In step 5, the autoencoder technology is used to anchor the demand state according to the demand type, and the formula is: Where L represents the reconstruction error function, such as mean square error or cross entropy, and x i represents the input of the i-th sample, represents the output of the i-th sample, λ represents the regularization coefficient, R(θ) represents the regularization term, such as L1 or L2 norm, etc., and θ represents the model parameter; Continue training until the normal distribution test P>0.05 is met, and the threshold is set to (μ±2σ). If the threshold interval is met, the corresponding multi-agent demand state is output, and the next category of demand state screening is started until all the set multi-agent activity demand simulations and screenings are completed.

Images