Multi-agent simulation and planning optimization method based on man-machine interaction

Abstract

The invention discloses a multi-agent simulation and planning optimization method based on man-machine interaction. The method comprises the steps of target site basic data acquisition, man-machine interaction intelligent platform establishment, community facility planning multi-scheme generation, community multi-agent simulation and supply and demand index measurement and calculation, and community facility planning scheme comparison, selection, optimization and comprehensive output. According to the method, a community facility planning scheme is generated in a man-machine interaction mode, a community facility supply and demand scene is simulated and evaluated through a multi-agent model, a man-machine interaction planning design process can be combined with the multi-agent model, and a foundation is laid for intelligent community facility layout and multi-agent auxiliary planning design. According to the method, the limitation that traditional community facility planning heavies the current situation of community facility construction and lightens the actual demand characteristics of community residents is overcome, and the problem that a planning scheme is intelligently generated and deduced and evaluated in a man-machine interaction mode is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent planning and design, and specifically relates to a community "multi-agent" simulation and facility planning optimization based on human-computer interaction. Background Art

[0002] There are two limitations in the traditional community facility planning scheme: on the one hand, it only focuses on the construction status of community facilities, while ignoring the actual use of community residents as users. On the other hand, it only considers the requirements of the thousand-person index for community facilities, but not the actual needs of community residents. Therefore, traditional community facility planning requires a comprehensive approach that takes into account both the actual use of community residents and the actual needs of community residents to achieve supply and demand matching in community facility planning. This background requires us to simulate the spatial and temporal distribution of residents and identify the supply and demand of multi-scheme community facilities through human-computer interaction methods in technological innovation to achieve a balance between supply and demand in community facility planning. Summary of the invention

[0003] Purpose of the invention: In view of the shortcomings of the prior art, the present invention provides a "multi-agent" simulation and planning optimization method based on human-computer interaction, which solves the two limitations of traditional community facility planning: only focusing on the construction status of community facilities, while ignoring the actual usage of community residents as users; only considering the requirements of the thousand-person index for community facilities, but ignoring the actual needs of community residents.

[0004] The technical solution adopted by the present invention is a "multi-agent" simulation and planning optimization method based on human-computer interaction, comprising the following steps: S1, basic data collection of the target site: Collect the current geospatial information data and the current population spatiotemporal distribution data of the target community, perform spatial calibration and spatial superposition, standardize the data, and import them into the database for storage; the community current geospatial information data includes oblique photography data, community boundary data, community building data, land use function data, road traffic data, and community facility area of ​​interest (AOI) data; S2, construction of human-computer interaction intelligent platform for community facility planning simulation: This step is the core step of this patent. The construction of the human-computer interaction intelligent platform for community facility planning simulation is divided into three steps: first, run the data integration module to obtain the model training data set, then run the multi-agent simulation module to complete the multi-agent model training, and finally output to the three types of clients of the human-computer interaction device; Importing the current geographic spatial information data and the current population spatiotemporal distribution data collected in step S1 into a computer workstation and calculating them to obtain a model training data set, including the built environment characteristic indicators of each target community, the community attributes of the population spatiotemporal distribution data, the current population residence indicators, and the characteristic indicators of the current population using facilities; Set the Q-Learningr algorithm of the community crowd multi-agent model, and construct rules based on the current crowd's use of facilities and time costs; train the established multi-agent model in batches based on the model training data set until the average reward value is higher than the historical average in long-term decision-making, and complete the multi-agent model training; The model training data set and multi-agent model are pushed from the human-computer interaction intelligent platform to the human-computer interaction device; the human-computer interaction device is composed of three types of client devices, namely, the planner interaction device, the street office interaction device, and the resident interaction screen; the planner interaction device includes eye movement data analysis and statistics, and electronic map annotation functions; the street office device includes community facility status statistics and community facility plan display; the resident interaction screen includes plan interaction display and resident facial emotion collection functions; the human-computer interaction intelligent platform includes community facility planning simulation and facility planning optimization process; the human-computer interaction device calls the community facility planning database of the human-computer interaction intelligent platform during operation, and the data generated by the human-computer interaction device are uploaded and stored in the community facility planning database of the human-computer interaction intelligent platform; S3, generation of multiple community facility layout plans based on planner human-computer interaction: The oblique photography data described in step S1 is rotated at a fixed angle and a panoramic photo is captured by a computer terminal, and the oblique photography data is projected on a 4K resolution large screen as a basic photo set for eye movement experiment and played sequentially; then, a head-mounted eye tracker is used to extract the eye movement data generated by the community planner when viewing each photo; wherein the sampling rate of the head-mounted eye tracker is not less than 300 Hz, and the planner will be prompted to focus on the two aspects of building quality and facility performance when viewing each photo; The eye movement data is automatically processed using Tobii Pro Lab software to extract eye movement indicators from each area in the photo, including the number of fixations and annotation duration. The average annotation duration of each area is superimposed and sorted, and the top 20 marked as key areas are selected and transmitted to the planner's human-computer interaction device terminal. Then, the planner conducts multi-scheme facility layout planning for the selected key areas in front of the human-computer interaction device terminal, and scans and uploads them to the human-computer interaction intelligent platform. S4, community "multi-agent" simulation and supply and demand estimation: Number each plan proposed by the planner in step S3, and spatially associate the number data with environmental characteristics and population distribution characteristics to form a plan indicator set; input the plan indicator set into the multi-agent model trained in step S2 to simulate the population distribution, and screen out facility planning plans that meet the supply and demand index distribution range by calculating the public service facility demand index, the public service facility supply index, and the supply and demand index, and record them in the community facility planning database in step S2; S5, Scheme comparison and optimization based on human-computer interaction among residents: The community facility planning database stored in the back-end of the human-computer interaction intelligent platform is wirelessly transmitted, and these selected plans are connected to the resident interactive screen in real time. Residents can zoom in and out on the plans displayed on the screen to traverse the community facility layout of each selected plan; the facial expressions of residents during the operation are automatically captured by the camera of the resident interactive screen, and based on the facial expression recognition interface of the back-end of the human-computer interaction intelligent platform, OpenCV is used to parse and summarize the emotional proportion of residents to each plan, and generate a resident perception database of each planning plan; the resident perception database includes the community facility layout plan number, the proportion of negative emotions of residents, and the proportion of positive emotions of residents; the resolution of the resident interactive screen is not less than 3840*2160, and the shooting pixels of the camera device are not less than 20 million pixels; if the positive emotion of the residents is greater than or equal to 80%, the corresponding plan is marked as the public preferred plan; if the positive emotions of the residents of all plans are less than 80%, the plan is fed back to step S3 to regenerate the planning plan; the plan with the highest positive emotion index is selected as the final plan and pushed to the interactive device of the street office of the corresponding community; based on the final plan, the street office proposes a specific implementation schedule; S6, Comprehensive output of community facilities plan: The final plan of the community facility layout in step S5 is superimposed with the current geographic spatial information data and the current population spatiotemporal distribution data in step S1, and output to a three-dimensional digital sand table for comparative display.

[0005] Furthermore, the implementation process of step S1 is as follows: The community building data is collected using a mobile laser scanning system, which requires that the ranging accuracy of the mobile laser scanning system is less than 10mm, the scanning rate is greater than 1.5 million points / second, and the camera system pixel is greater than 20 million pixels; the community facility area of ​​interest (AOI) data is collected through the Baidu Map open platform; community boundary data, land use function data, and road traffic data are obtained through relevant departments; the current data on the spatiotemporal distribution of the community population includes LBS data and other high-spatial-resolution mobile positioning data, which are obtained through relevant departments.

[0006] Furthermore, the implementation process of step (2) is as follows: The community site data collected in step S1 is imported into a computer workstation with a computing power of 10 petaFLOPS and a memory of 320 GB for calculation. The community analysis unit and the community space vectorized data are associated and the built environment characteristic index of each community analysis unit is calculated. The community population spatiotemporal distribution data and the community analysis unit are associated to obtain the community attributes of the population spatiotemporal distribution data; the current population spatiotemporal distribution data and the land use function data are associated and the current population residence index is calculated; the population spatiotemporal distribution data and the community facility interest area (AOI) data are associated and the characteristic index of the current population using the facilities is calculated to obtain the model training data set; The rules of the current crowd multi-agent model are set, including environment perception module, cost measurement module, experience library, learning mechanism and decision control module; the environment perception module is used to perceive the crowding penalty factor in the environment; memory storage module: used to save and extract historical facility use information; cost measurement module is used to calculate the time cost and distance cost of using facilities in combination with the environment perception module; experience library is used for the agent to continuously update its own experience as the system evolves, and its content reflects the agent's evaluation of the facilities used in memory; the learning mechanism adopts the Q-Learningr algorithm based on multi-agent; the decision control module is used to combine the experience library and the historical information of the use of facilities to make decisions on the facilities to be used in the next simulation cycle; the established multi-agent model is trained in batches until the average reward value is higher than the historical average value in long-term decision-making, and the multi-agent model training is completed; The three types of client devices are all equipped with a 4K-definition two-way camera, a two-dimensional and three-dimensional space conversion calculator with an accuracy of meters, 2TB of local storage space, a 10,000 mAh portable battery, and a gigabit POE switch port, which can meet the actual needs of outdoor mobile power plugging and indoor fixed power connection, and indoor and outdoor mobile Internet access, and implement the entire process of community facility investigation, construction, updating, and management.

[0007] Furthermore, the implementation process of step S3 is as follows: The human-computer interaction step of the planner is to design the community facility layout by category for the facility-deficient plots marked with red frames in the planner's interactive device, and compare them in real time in the form of virtual reality. The device will identify the facility modification plots as highlighted. After the planner confirms that all modifications are completed, the highlights will be completely cancelled and the plan can be stored in the planner's interactive device. The facilities in each category are specifically divided into seven sections: elderly care and childcare, community meal assistance, housekeeping convenience, health services, sports and fitness, cultural leisure, and children's recreation. The current status map of the community facility layout and the design sketches of multiple plans are completed.

[0008] Furthermore, the implementation process of step S4 is as follows: The implementation process of step S4 is as follows: By calculating the public service facility demand index, the public service facility supply index and the supply and demand index, the supply and demand index interval is determined to be between 0.9 and 1.1, and the facility planning schemes that meet the supply and demand index distribution interval are marked as candidate schemes and recorded in the community facility planning database in step S2; the public service facility demand index numbered n of the community m The calculation method is as follows: , where is the total length of time that user i stays in the public service facility numbered n in community m, is the total length of time that user i stays in community m, is the total number of days that user i stays in the public service facility numbered n in community m, is the total number of days that user i stays in community m; The energy level index of the public service facility numbered n in the community m The calculation method is as follows: , where The number of people living in community m among the users of type n service facilities in community m, The total number of people using the public service facility numbered n in community m; The public service facility supply index of community m numbered n The calculation method is as follows: , where is the energy level index of type n public service facilities in community m, The accessibility of n types of public service facilities in community m to the user's residence; The supply and demand index of the community facility planning scheme is the ratio of the demand index of the n-type public service facilities in community m to the supply index of the community public service facilities; the calculation method is as follows: , where is the supply and demand index of community facility planning scheme i, is the demand index of n types of public service facilities in community m in community facility planning scheme i, It is the public service facility supply index of n types of public service facilities in community m in community facility planning scheme i.

[0009] The beneficial effects of the present invention are: 1. Generate community facility planning schemes through human-computer interaction, simulate community facility supply and demand scenarios through multi-agents, conduct dynamic quantitative calculations, and select the public preferred schemes for community facilities based on interactive means of facial expression recognition and present them in a three-dimensional digital sandbox. This method combines the planning and design process of human-computer interaction with multi-agents, laying a solid foundation for the layout of intelligent community facilities and multi-agent-assisted planning and design.

[0010] 2. Multi-agent simulation and calculation is conducive to improving the organic combination of rational planning and intelligent technology, realizing the intelligent generation and optimization of social intrinsic attributes, and exploring future development trends and possibilities while ensuring the matching of supply and demand of community facilities. This efficient intelligent generation and optimization system not only intuitively presents data performance and virtual effects, but also strengthens the communication of opinions between the users and builders of community facilities, and pays more attention to "people-oriented" urban planning and design.

[0011] 3. Through the video, image, and text feedback created during the human-computer interaction process, a more efficient and scientific iteration of community facility planning schemes is achieved. This method can both retain the leading role of planners in scheme design and enhance the main role of residents in urban construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 A flow chart of the method implemented by the present invention; Figure 2 A diagram showing the construction of a human-computer interaction intelligent platform according to an embodiment of the present invention; Figure 3 A supply and demand calculation diagram of each community facility planning scheme in an embodiment of the present invention; Figure 4 This is a scheme comparison and optimization diagram based on resident human-computer interaction according to an embodiment of the present invention. DETAILED DESCRIPTION

[0013] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0014] like Figure 1-4 As shown, a "multi-agent" simulation and planning optimization method based on human-computer interaction includes the following steps: Step 1: Collect the current geographic spatial information data and the current population spatiotemporal distribution data of the communities within the old city of Nanjing, Jiangsu Province, perform spatial calibration and spatial superposition, standardize the data, and import them into the database for storage; wherein the community current geographic spatial information data includes oblique photography data, community boundary data, building data, land use function data, road traffic data, and community facility area of ​​interest (AOI) data; The detailed process of step 1 is as follows: The current geographic spatial information data and current population spatiotemporal distribution data of communities within the old city of Nanjing, Jiangsu Province are collected, and the community-related data are spatially calibrated and spatially overlapped, the data are standardized, and imported into a database for storage; the current community geographic spatial information data include community boundary data, building data, land use function data, road traffic data, and community facility area of ​​interest (AOI) data, among which the community building data are collected using a mobile laser scanning system, which requires the ranging accuracy of the mobile laser scanning system to be less than 10mm, the scanning rate to be greater than 1.5 million points / second, and the camera system pixel to be greater than 20 million pixels; the community facility area of ​​interest (AOI) data are collected through the Baidu map open platform; the community boundary data, land use function data, and road traffic data are obtained through the local Natural Resources Bureau; the current data on the spatiotemporal distribution of the community population include LBS data and other high-spatial-resolution mobile positioning data, which are obtained through the local Natural Resources Bureau. Step 1.2: Build a database. The DGX workstation equipped with ArcGIS 10.8 will calibrate the data obtained in step 1.1 uniformly according to WGS1984 and perform spatial calibration. The urban spatial vectorized data, urban functional business data, and urban population spatiotemporal distribution data obtained in step 1.1 will be spatially superimposed to form the urban three-dimensional spatial sand table and prediction model dataset of Quanzhou City. Step 2: Import the community site data within the old city of Nanjing, Jiangsu Province collected in step S1 into a computer workstation with a computing power of 10 petaFLOPS and a memory of 320 GB and calculate, associate the community analysis unit with the community space vector data, and calculate the built environment characteristic index of each community analysis unit, associate the community population spatiotemporal distribution data with the community analysis unit to obtain the community attributes of the population spatiotemporal distribution data; associate the current population spatiotemporal distribution data with the land use function data and calculate the current population residence index; associate the population spatiotemporal distribution data with the community facility interest area (AOI) data and calculate the characteristic index of the current population using the facility, and obtain the model training data set; The detailed process of step 2 is as follows: 2.1 Import the community site data within the old city of Nanjing, Jiangsu Province collected in step S1 into a computer workstation with a computing power of 10 petaFLOPS and a memory of 320GB and calculate, associate the community analysis unit with the community space vector data and calculate the built environment characteristic index of each community analysis unit, associate the community population spatiotemporal distribution data with the community analysis unit to obtain the community attributes of the population spatiotemporal distribution data; associate the current population spatiotemporal distribution data with the land use function data and calculate the current population residence index; associate the population spatiotemporal distribution data with the community facility interest area (AOI) data and calculate the characteristic index of the current population using the facility, and obtain the model training data set; 2.2 Set the rules of the current crowd multi-agent model, including environment perception module, cost measurement module, experience library, learning mechanism and decision control module. The environment perception module is used to perceive the crowding penalty factor in the environment; the memory storage module is used to save and extract the historical information of the use of facilities; the cost measurement module is used to calculate the time cost and distance cost of using facilities in combination with the environment perception module; the experience library is used for the agent to continuously update its own experience as the system evolves, and its content reflects the agent's evaluation of the use of facilities in memory; the learning mechanism adopts the Q-Learningr algorithm based on multi-agents; the decision control module is used to combine the experience library and the historical information of the use of facilities to make decisions on the facilities to be used in the next simulation cycle; the established multi-agent model is trained in batches until the average reward value is higher than the historical average in the long-term decision, and the multi-agent model training is completed.

[0015] 2.3 Push the model training data set and multi-agent model from the human-machine intelligent interaction platform to the human-machine interaction device; the human-machine interaction device consists of three types of client devices, namely planner interaction devices, street office interaction devices, and resident interaction screens. The planner interaction device includes eye movement data analysis and statistics, and electronic map annotation functions; the street office equipment includes community facility status statistics and community facility plan display; the resident interaction screen includes plan interaction display and resident facial emotion collection functions. The three types of client devices are all equipped with a 4K resolution two-way camera, a meter-precision two-dimensional and three-dimensional space conversion calculator, 2TB local storage space, 10,000 mAh portable battery, and a gigabit POE switch socket to meet the actual needs of outdoor mobile plug-in and indoor fixed power connection, indoor and outdoor mobile Internet access, and implement the entire process of community facility investigation, construction, update, and management.

[0016] Step 3: Rotate the oblique photography data in step 1 at a fixed angle and capture panoramic photos through a computer terminal, and project them as the basic photo set for the eye movement experiment on a 4K resolution large screen for sequential playback; then, use a head-mounted eye tracker to extract the eye movement data generated by the community planner when viewing each photo. The sampling rate of the head-mounted eye tracker is not less than 300Hz, and the planner will be prompted to focus on the building quality and facility performance when viewing each photo.

[0017] The eye movement data is automatically processed using Tobii Pro Lab software to extract eye movement indicators from each area in the photo, including the number of fixations and annotation duration. The average annotation duration of each area is superimposed and sorted, and the top 20 marked as key areas are selected and transmitted to the planner's human-computer interaction device terminal. Then, the planner conducts multi-scheme facility layout planning for the selected key areas in front of the human-computer interaction device terminal, and scans and uploads them to the human-computer interaction intelligent platform.

[0018] The detailed process of step 3 is as follows: the oblique photography data described in step 1 is rotated at a fixed angle and a panoramic photo is captured by a computer terminal, and it is projected on a 4K resolution large screen as a basic photo set for the eye movement experiment and played sequentially; then, a head-mounted eye tracker is used to extract the eye movement data generated by the community planner when viewing each photo. Among them, the sampling rate of the head-mounted eye tracker is not less than 300Hz, and the planner will be prompted to focus on the two aspects of building quality and facility performance when viewing each photo. The eye movement data is automatically processed by Tobii Pro Lab software to extract the eye movement indicators of each area in the photo, and the eye movement indicators include the number of fixations and the length of annotations; and the average annotation length of each area is superimposed and sorted, and the top 20 marked as key areas are selected and transmitted to the planner's human-computer interaction device terminal. Then, the planner conducts multi-scheme facility layout planning for the selected key areas in front of the human-computer interaction device terminal, and scans and uploads them to the human-computer interaction intelligent platform. The human-computer interaction step of the planner is to design the community facility layout by category for the facility-deficient plots marked with red frames in the planner's interactive device, and compare them in real time in the form of virtual reality. The device will identify the facility modification plots as highlighted. After the planner confirms that all modifications are completed, the highlights will be completely cancelled and the plan can be stored in the planner's interactive device. The facilities in each category are specifically divided into seven sections: elderly care and childcare, community meal assistance, housekeeping convenience, health services, sports and fitness, cultural leisure, and children's recreation. The current status map of the community facility layout and the design sketches of multiple plans are completed.

[0019] Step 4: Number each plan proposed by the planner in S3, and spatially associate the number data with environmental characteristics and population distribution characteristics to form a plan indicator set. Input the plan indicator set into the multi-agent model trained in step S2 to simulate the population distribution. Further, by calculating the public service facility demand index, the public service facility supply index, and the supply and demand index, select the facility planning plans that meet the supply and demand index distribution range, and record them in the Nanjing Old City community facility planning database in step 2; The detailed process of step 4 is as follows: Each plan proposed by the planner in step S3 is numbered, and the numbered data is spatially associated with the environmental characteristics and the population distribution characteristics to form a plan indicator set. The plan indicator set is input into the multi-agent model trained in step S2 to simulate the population distribution of each community in Nanjing Old City. Further, the implementation process of step S4 is as follows: By calculating the public service facility demand index, the public service facility supply index and the supply and demand index, the supply and demand index interval is determined to be between 0.9 and 1.1, and the facility planning schemes that meet the supply and demand index distribution interval are marked as candidate schemes and recorded in the community facility planning database in step S2; the public service facility demand index numbered n of the community m The calculation method is as follows: , where is the total length of time that user i stays in the public service facility numbered n in community m, is the total length of time that user i stays in community m, is the total number of days that user i stays in the public service facility numbered n in community m, is the total number of days that user i stays in community m; The energy level index of the public service facility numbered n in the community m The calculation method is as follows: , where The number of people living in community m among the users of type n service facilities in community m, The total number of people using the public service facility numbered n in community m; The public service facility supply index of community m numbered n The calculation method is as follows: , where is the energy level index of type n public service facilities in community m, The accessibility of n types of public service facilities in community m to the user's residence; The supply and demand index of the community facility planning scheme is the ratio of the demand index of the n-type public service facilities in community m to the supply index of the community public service facilities; the calculation method is as follows: , where is the supply and demand index of community facility planning scheme i, is the demand index of n types of public service facilities in community m in community facility planning scheme i, It is the public service facility supply index of n types of public service facilities in community m in community facility planning scheme i.

[0020] Step 5: The community facility planning database stored in the back-end of the human-computer interaction intelligent platform is wirelessly transmitted to connect these selected plans to the resident interactive screen in real time. Residents can zoom in and out on the plans displayed on the screen to traverse the community facility layout of each selected plan; the facial expressions of residents during the operation are automatically captured by the camera of the resident interactive screen, and based on the facial expression recognition interface of the back-end of the human-computer interaction intelligent platform, the emotional proportion of residents to each plan is analyzed and summarized using OpenCV to generate a resident perception database of each planning plan. The resident perception database includes the community facility layout plan number, the proportion of negative emotions of residents, and the proportion of positive emotions of residents. The resolution of the resident interactive screen is not less than 3840*2160, and the shooting pixels of the facial emotion recognition device are not less than 20 million pixels. If the positive emotions of residents are greater than or equal to 80%, the corresponding plan is marked as the public preferred plan; if the positive emotions of residents in all plans are less than 80%, the plan is fed back to step S3 to regenerate the planning plan. The plan with the highest positive emotion index is selected as the final plan and pushed to the interactive device of the street office of the corresponding community. Based on the final plan, the street office proposes a specific implementation schedule.

[0021] Step 6: The final plan for the layout of facilities in each community in Nanjing Old Town in step S5 is superimposed with the current geographic spatial information data and the current population spatiotemporal distribution data in step S1, and output to a three-dimensional digital sand table for comparative display.

Claims

1. A "multi-agent" simulation and planning optimization method based on human-computer interaction, characterized in that: The following steps are involved: S1, basic data collection of the target site: Collect the current geographic spatial information data and current population spatiotemporal distribution data of the target community, perform spatial calibration and spatial superposition, standardize the data, and import them into the database for storage; the community current geographic spatial information data includes oblique photography data, community boundary data, community building data, land use function data, road traffic data, and community facility interest surface data; S2, construction of human-computer interaction intelligent platform for community facility planning simulation: The construction of the human-computer interaction intelligent platform for community facility planning simulation is divided into three steps: first, run the data integration module to obtain the model training data set, then run the multi-agent simulation module to complete the multi-agent model training, and finally output to the three types of clients of the human-computer interaction device; Importing the current geographic spatial information data and the current population spatiotemporal distribution data collected in step S1 into a computer workstation and calculating them to obtain a model training data set, including the built environment characteristic indicators of each target community, the community attributes of the population spatiotemporal distribution data, the current population residence indicators, and the characteristic indicators of the current population using facilities; Set the Q-Learningr algorithm of the community crowd multi-agent model, and construct rules based on the current crowd's use of facilities and time costs; train the established multi-agent model in batches based on the model training data set until the average reward value is higher than the historical average in long-term decision-making, and complete the multi-agent model training; Push the model training data set and multi-agent model from the human-computer interaction intelligent platform to the human-computer interaction device; The human-computer interaction device consists of three types of client devices, namely planner interaction devices, street office interaction devices, and resident interaction large screens; The planner interaction device includes eye movement data analysis and statistics, and electronic map annotation functions; The street office equipment includes community facility status statistics and community facility plan display; the resident interactive screen includes plan interactive display and resident facial emotion collection functions; the human-computer interaction intelligent platform includes community facility planning simulation and facility planning optimization process; the human-computer interaction equipment calls the community facility planning database of the human-computer interaction intelligent platform during operation, and the data generated by the human-computer interaction equipment are uploaded and stored in the community facility planning database of the human-computer interaction intelligent platform; S3, generation of multiple community facility layout plans based on planner human-computer interaction: The oblique photography data described in step S1 is rotated at a fixed angle and a panoramic photo is captured by a computer terminal, and the oblique photography data is projected on a 4K resolution large screen as a basic photo set for eye movement experiment and played sequentially; then, a head-mounted eye tracker is used to extract the eye movement data generated by the community planner when viewing each photo; wherein the sampling rate of the head-mounted eye tracker is not less than 300 Hz, and the planner will be prompted to focus on the two aspects of building quality and facility performance when viewing each photo; The eye movement data is automatically processed using Tobii Pro Lab software to extract eye movement indicators from each area in the photo, including the number of fixations and annotation duration. The average annotation duration of each area is superimposed and sorted, and the top 20 marked as key areas are selected and transmitted to the planner's human-computer interaction device terminal. Then, the planner conducts multi-scheme facility layout planning for the selected key areas in front of the human-computer interaction device terminal, and scans and uploads them to the human-computer interaction intelligent platform. S4, community "multi-agent" simulation and supply and demand estimation: Number each plan proposed by the planner in step S3, and spatially associate the number data with environmental characteristics and population distribution characteristics to form a plan indicator set; input the plan indicator set into the multi-agent model trained in step S2 to simulate the population distribution, and screen out facility planning plans that meet the supply and demand index distribution range by calculating the public service facility demand index, the public service facility supply index, and the supply and demand index, and record them in the community facility planning database in step S2; S5, Scheme comparison and optimization based on human-computer interaction among residents: The community facility planning database stored in the back-end of the human-computer interaction intelligent platform is wirelessly transmitted, and these selected plans are connected to the resident interactive screen in real time. Residents can zoom in and out on the plans displayed on the screen to traverse the community facility layout of each selected plan; the facial expressions of residents during the operation are automatically captured by the camera of the resident interactive screen, and based on the facial expression recognition interface of the back-end of the human-computer interaction intelligent platform, OpenCV is used to parse and summarize the emotional proportion of residents to each plan, and generate a resident perception database of each planning plan; the resident perception database includes the community facility layout plan number, the proportion of negative emotions of residents, and the proportion of positive emotions of residents; the resolution of the resident interactive screen is not less than 3840*2160, and the shooting pixels of the camera device are not less than 20 million pixels; if the positive emotion of the residents is greater than or equal to 80%, the corresponding plan is marked as the public preferred plan; if the positive emotions of the residents of all plans are less than 80%, the plan is fed back to step S3 to regenerate the planning plan; the plan with the highest positive emotion index is selected as the final plan and pushed to the interactive device of the street office of the corresponding community; based on the final plan, the street office proposes a specific implementation schedule; S6, Comprehensive output of community facilities plan: The final plan of the community facility layout in step S5 is superimposed with the current geographic spatial information data and the current population spatiotemporal distribution data in step S1, and output to a three-dimensional digital sand table for comparative display.

2. The "multi-agent" simulation and planning optimization method based on human-computer interaction according to claim 1 is characterized in that: The implementation process of step S1 is as follows: The community building data is collected by a mobile laser scanning system, which requires a ranging accuracy of less than 10 mm, a scanning rate of more than 1.5 million points per second, and a camera system pixel of more than 20 million pixels; The community facility interest data is collected through the Baidu map open platform; the current data on the spatiotemporal distribution of the community population includes LBS data and other high spatial resolution mobile positioning data.

3. The "multi-agent" simulation and planning optimization method based on human-computer interaction according to claim 2 is characterized in that: The implementation process of step (2) is as follows: The community site data collected in step S1 is imported into a computer workstation with a computing power of 10 petaFLOPS and a memory of 320 GB for calculation. The community analysis unit and the community space vectorized data are associated and the built environment characteristic index of each community analysis unit is calculated. The community population spatiotemporal distribution data and the community analysis unit are associated to obtain the community attributes of the population spatiotemporal distribution data; the current population spatiotemporal distribution data and the land use function data are associated and the current population residence index is calculated; the population spatiotemporal distribution data and the community facility interest surface data are associated and the characteristic index of the current population using the facility is calculated to obtain the model training data set; The rules of the current crowd multi-agent model are set, including environment perception module, cost measurement module, experience library, learning mechanism and decision control module; the environment perception module is used to perceive the crowding penalty factor in the environment; memory storage module: used to save and extract historical facility use information; cost measurement module is used to calculate the time cost and distance cost of using facilities in combination with the environment perception module; experience library is used for the agent to continuously update its own experience as the system evolves, and its content reflects the agent's evaluation of the facilities used in memory; the learning mechanism adopts the Q-Learningr algorithm based on multi-agent; the decision control module is used to combine the experience library and the historical information of the use of facilities to make decisions on the facilities to be used in the next simulation cycle; the established multi-agent model is trained in batches until the average reward value is higher than the historical average value in long-term decision-making, and the multi-agent model training is completed; The three types of client devices are all equipped with a two-way camera with a resolution of 4K, a two-dimensional and three-dimensional space conversion calculator with an accuracy of meters, 2TB of local storage space, a 10,000 mAh portable battery, and a gigabit POE switch port, which can meet the actual needs of outdoor mobile power plugging and indoor fixed power connection, and indoor and outdoor mobile Internet access, and implement the entire process of community facility investigation, construction, updating, and management.

4. The "multi-agent" simulation and planning optimization method based on human-computer interaction according to claim 3 is characterized in that: The implementation process of step S3 is as follows: The human-computer interaction step of the planner is to design the community facility layout by category for the facility-deficient plots marked with red frames in the planner's interactive device, and compare them in real time in the form of virtual reality. The device will identify the facility modification plots as highlighted. After the planner confirms that all modifications are completed, the highlights will be completely cancelled and the plan can be stored in the planner's interactive device. The facilities in each category are specifically divided into seven sections: elderly care and childcare, community meal assistance, housekeeping convenience, health services, sports and fitness, cultural leisure, and children's recreation. The current status map of the community facility layout and the design sketches of multiple plans are completed.

5. The "multi-agent" simulation and planning optimization method based on human-computer interaction according to claim 4 is characterized in that: The implementation process of step S4 is as follows: By calculating the public service facility demand index, the public service facility supply index and the supply and demand index, the supply and demand index interval is determined to be between 0.9 and 1.1, and the facility planning schemes that meet the supply and demand index distribution interval are marked as candidate schemes and recorded in the community facility planning database in step S2; the public service facility demand index numbered n of the community m The calculation method is as follows: , where is the total length of time that user i stays in the public service facility numbered n in community m, is the total length of time that user i stays in community m, is the total number of days that user i stays in the public service facility numbered n in community m, is the total number of days that user i stays in community m; The energy level index of the public service facility numbered n in the community m The calculation method is as follows: , where The number of people living in community m among the users of type n service facilities in community m, The total number of people using the public service facility numbered n in community m; The public service facility supply index of community m numbered n The calculation method is as follows: , where is the energy level index of type n public service facilities in community m, The accessibility of n types of public service facilities in community m to the user's residence; The supply and demand index of the community facility planning scheme is the ratio of the demand index of the n-type public service facilities in community m to the supply index of the community public service facilities; the calculation method is as follows: , where is the supply and demand index of community facility planning scheme i, is the demand index of n types of public service facilities in community m in community facility planning scheme i, It is the public service facility supply index of n types of public service facilities in community m in community facility planning scheme i.

Images