Dream of a realistic SimCity

Machine Learning Artificial Intelligence Digital Transformation Probabilistic Generative Models Support Vector Machine Sparse Modeling Anomaly and Change Detection Relational Data Learning Time Series Data Analysis Simulation and Machine Learning Navigation of this blog
Summary

From Iwanami Data Science Series: “Time Series Analysis – State Space Model, Causal Analysis, and Business Applications. In the previous article, we discussed simulation of protein 3D geometry (misfolding) and analysis by machine learning. In this issue, we will discuss the dream of a realistic SimCity.

SimCity as a problem-solving tool

SimCity is an urban development simulation game that has been a long-running hit since its release in 1989 to this day.

In this game, you play the role of a mayor and build a virtual city, constructing power plants, town halls, and other infrastructure necessary for the year. As the city grows to a certain degree, the game becomes more challenging as the complexity of the issues to be addressed, such as the satisfaction of the residents and the response to disasters, increases.

SimCity is a game for entertainment purposes, but if a simulator based on real-world data and realistic models can be used as an interface like SimCity, even non-experts can be widely and deeply involved in the planning and formulation of marketing and policy measures. SimCity is a realistic simulator based on realistic data and realistic models.

There are two major societal needs for realistic SimCity. One is decision-making support, which is difficult to foresee in this era of social change due to technological development, aging society with a declining birthrate, and changes in laws and regulations. For example, if automated driving technology spreads to Level 4, rents in front of train stations will be high and those in the suburbs will be low, and this will cause changes.

To quantitatively indicate the possibilities and risks that may occur in the future requires specialized human resources, know-how, and other resources, and thus incurs significant costs. The existence of a platform that allows for simple, arbitrary scenarios to be set up and predicted will help companies and governments respond flexibly to the changes of the times.

Another is the pursuit of reality. In the past, the household-based attribute estimation data described below was used in the “Amazing Disaster Prevention Training,” an overnight disaster prevention training program sponsored by the city of Urayasu in Chiba Prefecture. By showing the participating junior high school students the realistic population distribution and damage situation in the event of a depth 7 confidence event, they were able to experience a sense of tension as they considered and acted on their own roles and priorities.

Today, we are inundated with data, computer resources essential for large-scale calculations, and software useful for creating calculation programs are widely available. One form of utilizing these resources is the realistic SimCity.

Challenges in Realizing a Realistic SimCity

SimCity is a multi-agent simulator. The satisfaction level of residents, an important concept in SimCity, is calculated through this process.

In this case, the agents are the residents and the cars. For example, in the case of residents, each individual has attributes such as address, income, and satisfaction level. The agent’s behavior is determined according to the level of satisfaction, and if the level of satisfaction is low, “moving out” occurs. Since the series of processes are sequentially visualized as the actions and attributes of the agents, the causal relationships of the game can be intuitively grasped.

The advantage of SimCity as a simulator is, of course, that all the attributes (elements) that affect the game progression and their relationships can be grasped (= defined) as simple rules. Therefore, SimCity makes it possible to capture the sum of each agent’s actions in a hierarchical manner, and succeeds in representing various social phenomena such as population shifts, pollution problems, and traffic congestion in the game.

However, in the real world, there are many phenomena that cannot be explained in a single way. For example, for a single individual’s purchasing behavior, there are countless possible attributes, such as distance to a store, price, brand, items handled, and so on. In order to create a universal tool that completely reproduces the real world, this entire world would have to be digitally copied, which is not realistic.

Therefore, it is conceivable to narrow down the phenomenon to be represented by the simulator, such as the “moving out” described above, and virtually derive the factors that cause the phenomenon, which are actually difficult to observe, as variables, such as the “satisfaction level” of the residents. For this purpose, principal components obtained statistically from a comprehensive set of observable data may be used, or if no source exists, a hypothetical-deductive method based on axioms or households may be used.

Technology for a realistic SimCity

One of the elements necessary to realize a realistic SimCity is the creation of agent data. In this section, we discuss attribute estimation for the case where residents are used as agents.

Attributes applicable to residents would be information such as family structure, age, and income, which are considered to have an impact on consumption behavior. For example, it is not difficult to imagine that there are differences in lifestyle and household expenditure structure between a single family household and a household consisting of parents and children. Although this information is basic information, it is generally considered personal information and is difficult to handle from the standpoint of privacy protection.

Therefore, inspired by downscaling, which is used in meteorology to predict regional climates, we estimated resident attributes based on statistical survey data. The transition establishment is generated from population statistics compiled by town, street, and character, such as the census.

For example, the number of married-couple households is tabulated in the population statistics by residential area, and the residential area of each household unit can be obtained and estimated from map data such as basic map information (Geospatial Information Authority of Japan) using GIS. In addition, the number of married-couple households aggregated by the age of the head of household can be combined into an estimation matrix, from which attributes such as the household type and age of the head of household can be estimated. The figure below shows the distribution of the elderly population estimated in this way.

Conclusion

Future work toward the realization of a realistic SimCity is still limited to agent parameter estimation, but it is conceivable that behavioral data such as purchase and visit information, which correspond to the results of agents’ actions, will also be estimated based on the relationship with demographic attributes.

To obtain information on agents, some of the content can be pseudo-estimated, while others can be estimated using information on approved members and monitors held by the company. As one example, we describe the Frameworks Logistics Open Data Contest. In this contest, movement trajectory and action history data of delivery trucks and warehouse staff obtained in actual operations are made public in a form that anyone can access in a state close to raw data. The data are also analyzed by combining them with household attribute estimation data.

An early version of SimCity, called Micropolis, is available as open source from the Don Hopkins website.

In the next article, we will discuss the use of emulators and the inverse problem of molecular simulation.

コメント

タイトルとURLをコピーしました