Geospatial intelligence
Technological progress in the field of artificial intelligence (AI) has brought new opportunities and new challenges to the intelligent development and integrated innovation of geospatial-related research and applications. Although the early concept of artificial intelligence and the development of theoretical algorithms can be traced back to the 1940s and 1950s, the main driving force for its recent rapid development comes from the rapid development and development of deep learning models and development frameworks (such as Tensorflow, Keras, PyTorch). Industrialization is becoming increasingly mature, the production of big data in various industries has exploded, and the computing performance of computer hardware (such as graphics processing units GPU and high-performance computing platforms HPC) and terminal equipment has been continuously upgraded, which can support training and Deploy artificial intelligence models to support data-driven intelligent decision-making and industrial transformation. Geospatial Artificial Intelligence (GeoAI for short) refers to an interdisciplinary research direction that combines geospatial related science and artificial intelligence. It improves the dynamic perception, intelligent reasoning and knowledge of geographical phenomena through the research and development of machine spatial intelligence. Discover capabilities and seek to solve major scientific and engineering problems in the interaction between humans and the earth's environmental systems (such as population migration prediction, urban expansion prediction simulation, intelligent transportation decision-making under complex conditions, high-precision map production and autonomous driving, global change response Impact on agricultural products, natural disaster emergency rescue projects, etc.). The emergence and development of geospatial intelligence is related to disciplines such as geography, cartography and geographical information systems, remote sensing science and technology, earth system science, resources environment and urban and rural planning, intelligent transportation and computer science (especially machine learning and knowledge graphs) Cross-integration, innovation and development are closely integrated.
Research framework of geospatial intelligence
Intelligence in urban spaces
The seventh census data released recently shows that China's urbanization level has exceeded 63%, and it is already in the second half of the urbanization process. Against this background, the demand for urban space is changing from focusing on the rapid construction of "increments" to focusing on the refined governance of "stock". Urban space monitoring and evaluation, quality improvement and refined governance transformation are becoming hot spots. Echoing this general trend, the national "14th Five-Year Plan" outline clearly proposes the national strategy of "accelerating the pace of digital society construction and comprehensively improving urban quality", which requires us to apply newly emerged intelligent algorithms in the construction of high-density urban spaces. In order to meet the urgent needs of management, explore the path of coordinated development of high density and high quality.
After rapid development in recent years, the current intelligent research and practice at the urban spatial level has basically solved the problem of building a digital background for cities. A considerable number of cities and related agencies have implemented cloud services and built a large number of big data centers. , and achieve large-scale, high-precision acquisition of massive data through various cameras, the Internet of Things, and wearable devices. With the platform integration of new technologies such as mobile Internet, urban big data, Internet of Things, and wearable devices, a large-scale and high-precision built environment data integration is formed, as well as matching high-resolution citizen behavior and perception data. As the spatiotemporal accuracy of urban data continues to improve, the visualization of these massive new data can also reveal intuitive pictures that were previously difficult to effectively obtain, providing a solid foundation for individual-level behavioral activities and spatial morphological status assessment.
However, the demand for intelligence in urban space goes far beyond the visual display of massive data and the simple status monitoring and usage evaluation it supports. The humanization and quality of high-density urban space construction and management require the ability to efficiently handle and respond to a large number of complex and cross-connected urban problems, which cannot be solved through current data visualization. In other words, the application of intelligent algorithms cannot stop at the acquisition and cleaning of basic data on urban space and citizen behavior, but should be more deeply integrated into urban planning, design, governance and other fields, through various deep learning algorithms. The introduction of AI will promote intelligent changes that can be sensed, modeled, analyzed, predicted, explained, and decision-making, and is expected to bring changes to this series of experience-led industry paradigms.
Spatiotemporal big data intelligence
The Internet, sensing technology and large-scale computing infrastructure have generated a variety of dynamic and static big data in urban space, of which more than 80% are related to time and space, such as air quality readings, weather, and taxi movements. Trajectories, real-time traffic conditions, etc. These data all have at least a time or space dimension, and may have other attribute dimensions. Spatiotemporal data can be divided into point data and network data according to the data structure; according to whether the spatiotemporal information changes dynamically, it can be divided into three categories: spatial static, temporal static, and spatiotemporal dynamic. It can be divided into six categories through combination. Among them, point data that is static in space and dynamic in time: for example, sensors are mostly installed in fixed positions, but the readings they produce constantly change with time; point data that both time and space change with time: two There is no correlation between spatiotemporal dynamic point data, such as Didi taxi usage records in different time periods; spatially static and temporally dynamic point data: for example, the road network can be expressed by a network. Although the road network is a spatially static data structure, Superimpose dynamic traffic flow information on it, and the pipe network superimposed with dynamic flow information becomes a data structure that is static in space and dynamic in time; network structure data with events that are dynamic in both time and space: the nodes and edges of the entire network are constantly changing. . Taking trajectory data as an example, people, vehicles, and objects in space pass through different locations at different time periods, reflecting different states. Space-time points are connected in time order to form a chain structure.
Spatiotemporal big data means rich knowledge about a city. If used correctly, it can help solve various urbanization development and scene empowerment challenges. By integrating basic geographic information data resource pool (2D, 3D), sensory Internet of Things data resource pool (environment sensing data), and dimensional spatiotemporal data resource pool (people, vehicles, objects, fields), a unified geographic information data, The data governance framework for spatiotemporal data and business data, as well as the rich and flexible spatiotemporal service system framework, can support the efficient docking of massive data and complex emergency applications.
The development of digital twin cities
The conceptual model of digital twin first appeared in 2003, proposed by Grieves M. W．The professor proposed it in the product life cycle management course of the University of Michigan in the United States. It was called the "mirror space model" at the time and was later defined as the "digital twin". In 2010, the National Aeronautics and Space Administration (NASA) introduced the concept of digital twins for the first time in its space technology roadmap to achieve comprehensive diagnostic and predictive functions of flight systems to ensure continued safe operation throughout the system's service life.
Since October 2019, the National Development and Reform Commission, the Ministry of Science and Technology, the Ministry of Industry and Information Technology, the Ministry of Natural Resources, the Ministry of Housing and Urban-Rural Development and other ministries and commissions have intensively issued policy documents to effectively promote the rapid development of urban information modeling (CIM) and building information modeling (BIM) related technologies, industries and applications. , assisting the construction of digital twin cities.
At present, the construction of digital twin cities is in the 1.0 stage and mainly faces the following problems and challenges. First, it is mainly based on urban static data, but urban scenes are complex and the status of the city changes rapidly. How to capture dynamic data that reflects the city in a timely and accurate manner is the basis for smart city applications, and it is also a big problem. The second is to rely on information directly expressed by data to solve problems. The deeper knowledge behind the data needs to be further explored, and the depth and breadth of urban data utilization is still insufficient. The third is the lack of visual analysis capabilities for human-computer interaction and the integration of human-computer intelligence. The ability to counter-control the physical world through display is even more insufficient. The display and perception of the city do not form a closed loop. Fourth, it focuses on restoring real cities and cannot proactively create future scenarios through the application of intelligent models to help users see tomorrow's urban problems and guide today's development path accordingly. Fifth, construction is carried out mainly by the government without forming an open ecosystem, resulting in poor sustainability.
Development stages of artificial intelligence
Policy support: Artificial intelligence moves toward “ubiquitous intelligence” and promotes the development of intelligent industries
Artificial intelligence is an important driving force for a new round of technological revolution and industrial transformation, and its strategic importance has become a general consensus around the world. Starting from the 2015 State Council's official inclusion of artificial intelligence into one of the key tasks of "Internet +", my country's emphasis on artificial intelligence has continued to increase, and its development so far can be roughly divided into four stages.
Among them, the second phase is landmark. The 2017 Government Work Report included "artificial intelligence" for the first time, and artificial intelligence was officially elevated to a national strategy. In the same year, the State Council issued the "New Generation Artificial Intelligence Development Plan", which clearly stated the "three-step" strategic goal: by 2025, the new generation of artificial intelligence will be widely used in fields such as intelligent manufacturing, intelligent medical care, smart cities, intelligent agriculture, and national defense construction. application. The third stage is when artificial intelligence begins to be deeply integrated with the real economy. The fourth phase starts in 2020. Artificial intelligence is included in the "new infrastructure" policy and becomes one of the main supporting technologies for new technology infrastructure. It will assume a more important role in promoting the digital transformation and intelligent upgrading of trillions of real economy industries. and integrated innovation. National policy supports artificial intelligence to move towards "ubiquitous intelligence", that is, artificial intelligence technology will be widely penetrated into new infrastructure construction. Its essence is not only directed at the industrial development of artificial intelligence itself, but also to find application scenarios in the real economy to empower productivity. upgrade.
Technology development: from "perceptual intelligence" to "cognitive intelligence", allowing machines to "understand" and "explain"
Artificial intelligence was first proposed at the Dartmouth Conference in 1956. Since its development, it is generally believed that artificial intelligence is divided into three levels: computational intelligence, perceptual intelligence and cognitive intelligence. Computational intelligence refers to fast computing and massive storage capabilities; perceptual intelligence refers to the ability of machines to "listen, speak, and see" and to identify and judge concrete things. Its development benefits from the use of convolutional neural networks (Convolutional Neural Networks). Deep learning models represented by Network (CNN) and Recurrent Nerul Network (RNN) have developed rapidly. However, deep learning is difficult to effectively utilize prior knowledge, and its opacity and uninterpretability have become constraints on the development of artificial intelligence. Obstacles; cognitive intelligence is the ability to understand and explain, aiming to enable machines to understand semantics, logical reasoning and learning judgment. Therefore, the realization of cognitive intelligence needs to be driven by knowledge, which involves key technologies such as knowledge representation, semantic understanding, associative reasoning, intelligent question answering, emotional computing, decision planning and so on.
The current development of artificial intelligence is still in a state of weak artificial intelligence, and the research focus is undergoing a transition from perceptual intelligence to cognitive intelligence. Perceptual intelligence targeting vision and hearing and other recognition technologies has broken through the red line of industrialization, but it can only assist or replace humans in a certain aspect of human work. When people can use machines to identify more things, it will naturally trigger the need for deep automated knowledge services such as understanding and analysis of things. However, the field of cognitive intelligence that requires external knowledge, logical reasoning or domain migration is still in its infancy.
Knowledge graph empowerment: as the underlying support for cognitive intelligence
The concept of Knowledge Graph was first formally proposed by Google in 2012 and is mainly used to support the next generation of search and online advertising businesses. Knowledge graph is essentially a knowledge base based on semantic network, which aims to describe concepts, entities, events and the relationships between them in the objective world. Knowledge graph is the evolution and development of the symbolism research paradigm in the era of big data and artificial intelligence. It takes knowledge as the processing object and enhances the machine's recognition by simulating the human brain's knowledge cognition, problem solving, knowledge question and answer, knowledge reasoning and other functions. Intellectual ability, learning ability, reasoning ability. Its development is inseparable from the Semantic Web. Compared with the traditional Semantic Web, the advantages of the Knowledge Graph are: (1) It has richer semantic expression capabilities and can support applications in more scenarios; (2) It can be well integrated with artificial intelligence technology to realize cognitive intelligence and explainable artificial intelligence; (3) Data based on graph structure facilitates the storage and integration of knowledge.
Cognitive intelligence is an artificial intelligence technology that can be implemented in real terms, has extensive and diverse application requirements, and can generate huge social and economic value. As the underlying support for cognitive intelligence, knowledge graph will accelerate the advancement of artificial intelligence from "perceptual intelligence" to "cognitive intelligence", and has surpassed the level of artificial intelligence in many fields such as e-commerce, social networking, logistics, finance, medical care, justice, and manufacturing. More and more cases are emerging in which knowledge graphs empower artificial intelligence. Knowledge graphs provide new perspectives and opportunities, and bring about a new era of new technology, industry and society - the dawn of the cognitive era.
The proposal of spatiotemporal artificial intelligence
The development of artificial intelligence has penetrated into application fields related to geospatial intelligence, urban spatial intelligence, spatiotemporal big data intelligence, and digital twin urban development. We see that in these fields, no matter in terms of theory, technological innovation and application practice, a core underlying innovation is artificial intelligence based on space and time: spatio-temporal artificial intelligence (Spatio-Temporal AI). With the development of urban digital twins, spatiotemporal artificial intelligence (ST-AI) has begun to receive widespread attention from industry, academia and research. For this reason, we organized industry experts to publish this white paper "Spatial-temporal Artificial Intelligence Empowers Digital Twin Cities (2021)" for the first time. The release of this book aims to promote the combination of AI technology and urban digital twin scenarios, promote technology upgrades through scenario applications, and empower urban digital transformation through technological innovation.
Current problems faced by digital twin city construction
(1) Insufficient exploration of urban application scenarios and lack of deep understanding of business logic.
A large number of digital twin products are divorced from application scenarios and actual business needs, blindly follow trends, lack a deep understanding of business logic, and ignore effective support for urban operation and governance. There are many twin applications in urban planning, construction management and other industries, but urban governance, services and other applications based on digital twins are generally insufficient in depth. Since city-level heterogeneous big data collection and cross-industry and cross-domain applications are still in their infancy, algorithm models in various majors and industries are not yet mature and have yet to be developed. Moreover, the market lacks standardized construction guidelines, and there are relatively few construction plans with long-term considerations, systematic layout, and in-depth application requirements.
(2) There is a lack of unified spatio-temporal intelligent platform specifications, and the digital base urgently needs to be integrated. The digital twin
city is derived from the practice of GIS, BIM, and CIM of "one map", but is higher than the "one map" construction. The urban management department has a clear understanding of the city's digital base. There is a strong demand for maps, such as the urban information model platform promoted by the housing and urban-rural construction system, the spatio-temporal big data platform led by natural resources and land planning, the city base map based on the public security, political and legal lines for urban security and comprehensive management, etc. Each of them has a strong demand. The base map is a self-contained system, which generally only supports applications within the system and cannot be used by other departments on demand. For a long time, smart city lines have strong coordination and weak coordination, and there is no clear leading department. Even though some places have established big data bureaus, coordination Coordination is still insufficient. How to integrate multiple basemaps to form a city-level unified digital basemap and data assets is the first issue to be faced in the construction of digital twin cities.
(3) Spatiotemporal data islands lack correlation and integration, and the value of data is far from being released. The
core of digital twin cities is to build a unified spatiotemporal intelligent base. However, a unified standard spatiotemporal data underlying framework that is compatible with heterogeneous information has not yet been formed. Institutions and manufacturers It is difficult to ensure unified coding and accurate fusion expression of multi-modal data if they advance independently. Moreover, there are certain difficulties in the fusion processing of multi-source data such as urban vector spatio-temporal, government business data, Internet of Things, raster data, model data, and point cloud data, and there are barriers between formats. Although data silos have always received people's attention, they have not been solved. Inconsistent data formats in business systems of various departments, unclear data permissions, and incomplete data docking mechanisms will all restrict the role of digital twin cities.
(4) The integration of algorithm models and dynamic data is not deep, and the ability to quickly analyze and assist decision-making is insufficient. The
new types of surveying and mapping, sign perception, collaborative computing, all-element expression, visualization, digital space simulation, situation prediction, etc. involved in the current digital twin city are of far greater value. It has not yet been released, and the development and integrated application of key technologies still need to be strengthened. Mass data loading technology, cloud-edge computing collaboration technology, simulation technology, etc. are not mature enough; the ability to use artificial intelligence, especially spatiotemporal artificial intelligence, and edge computing to quickly analyze and process dynamic data is insufficient.
(5) The investment and operation mechanism is unclear, and an open industrial ecosystem has not yet been formed.
From building components to land parcels to various professional units, and finally to the city, a layer-by-layer collection of urban operation and live data is formed, which is used to evaluate the city's " Is there a correlation between physical” construction genes and “virtual” operational performance. For example, how a well-performing science and technology park depends on the degree of land mix, the level of greening, the accessibility of the road network, the rent of housing, or the density of cafes. Establish an effective correlation mechanism between operational performance and construction genes, empower ordinary people and stakeholders, and help them participate in urban construction or urban operation management more conveniently and rationally. How to establish a relationship between citizens, experts and the government Real-time and effective communication and interaction channels need to be practiced everywhere.
Future development needs of digital twin cities: digital twin 2.0 integrated with smart technology
(1) Fine-grained urban smart scene services
should be designed according to local conditions. The top-level design of digital twin cities should be closely integrated with urban governance modernization scenarios and business needs. At the same time, the future development rules of the city and the evolution direction of information technology should be taken into consideration to promote Digital twin city implementation application. In the context of "ubiquitous intelligence", with the help of artificial intelligence technology, the construction of digital twin cities will be closely integrated with application scenarios such as economy, life, and governance to promote the digital transformation of real economic industries and truly serve urban smart scenarios.
(2) Construction of a unified spatio-temporal intelligent base oriented by business needs.
With the development of new technologies, new applications, and new scenarios in the digital economy, the demand for data sharing across regions, industries, and businesses is increasing day by day. The construction of digital twin cities urgently requires the formulation of a unified Rules framework, laws and regulations and basic platform. Follow the principle of combining goal orientation and demand orientation for overall planning, fully consider the future development laws of the city and the evolution direction of information technology, unify common needs, and collaboratively promote the construction of spatiotemporal intelligent bases. Establish a digital twin application system based on a unified base, evaluate the city base map and data resources of the existing line system, follow unified standards to build a spatio-temporal intelligent platform with advanced technology, complete data, and strong scalability, and gradually promote the application scenarios Rich and systematic iterative development.
(3) The fusion of multi-source heterogeneous spatio-temporal data and data asset management
should solve the data fusion, information sharing and business collaboration mechanism between all levels and systems of the digital twin city, focusing on solving the connection, cooperation and correlation between the systems. Constraint relationship issues, use systematic scientific methods to guide the planning and design of complex giant systems of smart cities, and improve the scientificity, standardization and operability of top-level design. Accelerate the improvement of standards and specifications for the fusion and processing of multiple heterogeneous data, and unify the format and encoding of various types of data such as vectors, raster grids, models, point clouds, government affairs, and perception to form full-cycle data standards and specifications, leveraging knowledge graphs in data Advantages in fusion to build fusion processing capabilities for multi-source heterogeneous data.
(4) Deep mining and analysis of spatiotemporal data based on efficient algorithm models
On the basis of providing conventional general spatial analysis capabilities for urban multi-source data, it combines artificial intelligence technology to build in-depth spatial analysis capabilities with algorithm models such as machine learning, deep learning and knowledge graphs, using classification, clustering, regression, deep learning, etc. Supervised and unsupervised learning methods realize information classification, mining and prediction, etc. Different types of analysis capabilities are modeled according to specific application scenarios to form spatial analysis capabilities for specific topics, such as park industry positioning analysis and dynamic assessment related to park planning, feasibility analysis and sales forecast for commercial site selection, Intelligent dispatching and travel planning for the transportation field, etc. It provides real-time online analysis and modeling capabilities, can build spatial data analysis pipelines, organize data and algorithm models in the form of directed acyclic graphs, and schedule and execute the analysis process according to strategies.
(5) Comprehensive sharing and cooperation of the industrial ecology jointly built by multiple stakeholders.
The digital twin and intelligentization of the city cannot be achieved overnight. It is necessary to build a new model of urban digital transformation that is co-constructed, co-governed and shared to provide sustained momentum. The implementation of Digital Twin City 2.0 is inseparable from the support of technology platform companies with full-stack spatiotemporal artificial intelligence, but one company alone is not enough. It needs to rely on relevant technology alliances to gather relevant enterprises and universities and research institutes with advantageous disciplines. Let’s complete it together. It is necessary to establish an innovation ecology, allowing the government to unite with enterprises, institutions, universities, and scientific research institutions in digital city-related fields to form a cross-industry, open alliance. Responsible for the research on digital twin city theory, standards, solutions and other public issues, establishing a new mechanism for effective industry-university-research cooperation in upstream and downstream industries, and forming the standards and quality of digital twins. Responsible for the preparation of plans for each digital twin application scenario, organizing application scenario demonstrations, and supporting the construction and development of future digital twin cities.