Author: ЧАЙКО ВЛАДИМИР ИВАНОВИЧ | CHAIKO VLADIMIR
Introduction
It is impossible to imagine the modern world without space technologies, and the future without flights to other celestial bodies, mining on them and just living.
The recently launched new space race "Exploration of the moon v2.0." leads us to the real creation of colonies on the Moon.
During its centuries-old existence, mankind has accumulated a large amount of knowledge, ideas and technologies for the exploration of celestial bodies. This is how space colonies, intergalactic flights and much more were described in science fiction long before the actual human flight into space.
Unfortunately, most of the accumulated knowledge is just a theory that needs expensive verification in practice.
Progress in the field of computing technology has allowed mankind to conduct experiments not with real physical objects, but with virtual ones – to create computer models. The passage of any computer modeling technologies does not give a 100% guarantee of their operability, however, those technologies that do not pass it are guaranteed to be inoperable. Preliminary verification of theories and ideas using computer modeling will make it possible to cut off a large number of non-working technologies even before testing them experimentally, which will allow humanity to save a large amount of money.
To date, there are a large number of computer modeling methods, each of which has its pros and cons. Most of them are not suitable for testing knowledge in the field of colonization. In this case, the method called by the author "simulation object-oriented modeling" (IOM) is most suitable for verification by the modeling method.
Simulation object-oriented modeling is able not only to test technologies, but also, when combined with self-learning algorithms and neural networks, to create new ones.
Relevance of the work: this work is relevant, as it proposes a method of reducing costs for solving an important and necessary task – verifying and obtaining humanity's knowledge about the colonization of celestial bodies.
Scientific novelty: this work uses a new method of computer modeling (Simulation object-oriented method) and the latest developments in the field of artificial intelligence (self-learning algorithm and neural network).
The purpose of the study is to prove the suitability of using simulated object-oriented modeling, self-learning algorithms and neural networks to test existing and develop new technologies for colonization of celestial bodies.
Research objectives:
To study existing modeling methods for their suitability for testing existing technologies for colonization of other celestial bodies;
To reveal the disadvantages of existing modeling methods;
To prove in practice the effectiveness of using simulation object-oriented modeling to test colonization technologies;
To reveal the superiority of object-oriented simulation over other methods in the framework of verification of colonization technologies;
To study modern technologies in the field of artificial intelligence;
To prove in practice the suitability of the use of artificial intelligence for the development of new technologies for the colonization of celestial bodies.
The object of research: methods for verifying existing and developing new technologies for colonizing celestial bodies through computer modeling and the use of artificial intelligence technologies.
The subject of the study: the computer IOM model "Echo of Pluto" created within the framework of this work.
The model itself is available for download from Yandex.Disk at the link: https://disk .yandex.ru/d/miUHNh2E738AYg.
The technical documentation is available for download from Yandex.Disk at the link: https://disk .yandex.ru/i/asO11hbeqhgP5g
The user manual is available for download from Yandex.Disk at the link: https://disk .yandex.ru/i/NlUTMt0-kxhQRQ
Thanks: to my father, Ivan Ivanovich Chaiko, for criticizing this study.
The main part
A new race for the Moon
Recently, a new space race has begun – "Exploration of the Moon v2.0.". The scale of this race, compared with the previous one, is much larger – not two countries (the USSR and the USA) participate in it, but as many as 6: Russia, the USA, China, India, Japan and Israel. [1] This is due to the fact that scientists found water on the Moon in 2023. [2]
Over the past 4 years, many major powers have launched or carried out their missions to the moon. So, in 2019, China managed to successfully launch the Chang'e-4 probe. [3] In 2023, Russia attempted to launch the Luna-25 lunar rover, after which India successfully landed the Chandrayan-3 probe. [4][5]
At the same time, in the USA, there was a boom in various projects related to the Moon. On August 15, 2023, DARPA launched the LunA-10 project. His main task is to bring together all the developments and existing projects in the field of exploration of celestial bodies and design the lunar infrastructure within 10 years. [6] This led to the creation of the Lunar Operating Guidelines for Infrastructure Consortium (LOGIC). [7]
The global trend is as follows: humanity has seriously engaged in the exploration of the Moon and the creation of real habitable bases on it. First populated by robots, with artificial intelligence, and then by living people. [8] It is worth noting that humanity already has the experience of creating the necessary robots for this, and robots in general. [9][10]
The need to test human knowledge about colonization
Humanity has a huge amount of different knowledge and technologies for the development of other celestial bodies. There is even some experience. They have been accumulating for a very long time: from the dreams of ancient philosophers, from books on space fiction and the works of futurists, to lunar rovers, rovers and, perhaps, the real landing of a living person on the Moon. [11][12][13]
Unfortunately, most of this knowledge is theory. Humanity needs to test all this knowledge for efficiency. Today, this is one of the main problems, without solving which we will not be able to conquer the Moon. As the great Russian scientist academician Chebyshev said: "The convergence of theory with practice gives the most beneficial results, and not only practice benefits from this: the sciences themselves develop under its influence: it opens up new subjects for research or new sides in subjects long known." [14]
There is only one way to prove the efficiency of the technology – to test it in practice. Unfortunately, practical tests are very expensive, and similar tests of space technologies are expensive in Cuba (to the third degree). Even tests in artificially created conditions close to real ones are very expensive. The Biosphere 2 project is a good example of this. [15]
Humanity needs a way to reduce the cost of testing accumulated knowledge and technologies. The solution to this problem is to check with computer modeling before practical tests. This is cheaper and will allow you to "weed out" a large number of broken ideas and theories at an early stage. This method cannot guarantee that the technologies that have been simulated will work, but it can guarantee the unsuitability of those technologies that will not pass it. [16]
Computer modeling and its methods
Computer modeling consists in the following: a real object or process is described mathematically, all its parameters are expressed by a variable (number) and linked together using mathematical formulas (functions). By changing the value of one or more parameters, the "impact" on the object or process is simulated. As a result of changing some parameters (variables), all other parameters (variables) change, which reflects the behavior of the object or process. All calculations are performed by a computer.
When conducting computer modeling, it is worth understanding the following:
During the creation of the model, the programmer must enter all the parameters of the object or process into it. Due to the large number of parameters in complex processes, you can forget to add something;
All parameters must be correctly described mathematically and linked together by mathematical functions. The programmer must correctly output all the functions, which is a difficult task.;
All parameters of an object or process must be measured with sufficient accuracy. At the same time, it is very difficult to determine a sufficient level of accuracy immediately.
All this leads to the fact that no one can guarantee that the created model really behaves like a real simulated object, so the reliability of the simulation result is not guaranteed. [16]
Choosing a modeling method
Computer modeling methods are divided into 3 types:
Analytical – description of a process or object by a single mathematical formula with its subsequent calculation on a computer;
Quantitative is the design of the parameters of a process or object in the form of variables, the dependence between which is established through the coefficients y = f(k,x). The values of the coefficients themselves are determined by statistical methods;
Simulation is a method in which they try to reflect the real behavior of an object or process. The parameters of a process or object are formed in the form of variables, the relationship between which is established through mathematical formulas that accurately describe the effect of one parameter on another.
The first two methods have very serious disadvantages, which the third one lacks:
When creating a model of complex objects or systems, you have to think of their individual parts as a simple variable for ease of implementation. Such a variable does not reflect the internal state of the part of the real system that it models;
The relationship between the variables is described by simple mathematical formulas that describe the real relationship between the parameters approximately;
There is no mathematical formula that predicts events in the future.
Therefore, using these methods to test technologies for suitability is not a good approach to solving the problem. The most appropriate method, in this case, is the simulation method. Considering this, the author suggests using a method that he calls "Simulated Object-oriented modeling" (IOM). [17]
Simulation object-oriented modeling
In the method proposed by the author, the simulated system consists entirely of two types of "parts": polygons and objects located on a plane. Objects interact with polygons and other objects during the "playback" of the model, as a result of which their parameters change. This method differs from all other simulation methods in that it has an element of randomness – the same thing is done in different (approximately the same) time and forces. To implement this, the model uses pseudo-random number generators. The model built on this principle most accurately describes the real life processes of a colony on any celestial body.
Simulation object-oriented model of the "Echo of Pluto"
General description of the model
The Echo of Pluto model proves the suitability and effectiveness of using IOM (with artificial intelligence technologies) to test and obtain knowledge of the conquest of celestial bodies.
The essence of everything that happens in the model is quite simple: a colony is deployed on Pluto. The task of the colonist living in it is to take samples of the surface of Pluto in strictly defined places. During the performance of his task, he (the colonist) spends physical and mental energy. It is necessary to arrange all the buildings of the colony (4 pieces) in such a way that the colonist collects samples as quickly as possible and with as little effort as possible. This is a test of the effectiveness of the methods of placing colony buildings relative to each other in order to identify the most effective arrangement.
In all processes of the model, there is randomness, implemented using 3 different pseudo-random number generators (2 standard and 1 improved by the author). So the fatigue of the colonist accumulates and decreases in a function close to the parabola (y=x2), but each time the parabola is unique due to the presence of some randomness.
You can learn more about the device of the model from its technical documentation. [18]
Choosing a celestial body for modeling
Pluto was chosen as a celestial body for modeling because:
Most of its orbit is located inside the Kuiper Belt. This is a belt of asteroids consisting of gas. The presence of a base on Pluto gives humanity access to a huge amount of energy resources; [19]
Being on the outskirts of the Solar System, Pluto is very well suited as a starting point for traveling beyond its borders; [20]
The total amount of water on Pluto exceeds the total amount of water on planet Earth. [21]
Despite the fact that the Pluto Echo model simulates Pluto, it (the model) is suitable for modeling colonies on other celestial bodies, including Mars, the Moon and Venus.
Forecasting and its methods
During the simulation, the following goal is achieved: to arrange the colony's buildings so that it "works" most effectively. This goal cannot be achieved without forecasting.
To date, there are only three methods of forecasting: static, model and expert. [22]
Statistical forecasting is based on mathematical statistics and statistical methods. It consists in collecting statistics, on the basis of which a forecast is made. An example of such a prediction is the exit poll, a poll of citizens in order to determine the winner of the elections before they are over. [23]
Model forecasting is based on the application of models, both physical and computer. An example of such a simulation is the simulation of the dynamics of the humanity of the Club of Rome. [24]
Expert forecasting is based on a survey of experts. It consists in interviewing a certain number of experts and highlighting common assumptions from them. This general is taken as a forecast. An example of such forecasting is the book by Jorgen Randers "A Global Forecast for the next 40 years. A report of the Club of Rome commemorating the 40th anniversary of The Limits to Growth». [25]
Implementation of 3 types of forecasting in the developed model
All three approaches were implemented in the DIGITAL model "Echo of Pluto" in the form of three modeling modes:
Human mode - within this mode, the user is given the opportunity to place all the buildings of the colony himself using a computer mouse. After placing all the buildings, the model is played and outputs performance data. This is the embodiment of "expert forecasting", because a person (an expert) arranges all the buildings based on his experience and assumptions;
Self-learning algorithm mode - within this mode, a self-learning algorithm deals with the arrangement of all elements. His task is to sort through all possible options for the arrangement of buildings and give the best possible arrangement as an answer. This is the epitome of statistical forecasting;
"Neural network" mode - within this mode, the neural network deals with the arrangement of all elements. With each new playback of the model, the neural network learns and provides a more effective solution to this problem. This is the epitome of model forecasting.
The appearance of all 3 operating modes is shown in Figures 1, 2 and 3 of Appendix A.
Statistical determination of the correct placement of buildings using a self-learning algorithm
A self–learning algorithm is a brute force algorithm that implements all solutions to a problem from a set of possible solutions. As a result of testing all possible solutions through modeling (collecting performance statistics), the most effective option is determined.
As a result of this collection of statistics, and its subsequent analysis, new patterns of the influence of the location of colony buildings on the effectiveness of its work can be discovered.
Search for general rules and methods of building placement using a neural network
A neural network is a program that simulates the physical work of the brain in order to obtain its capabilities (intelligence).
The human brain, according to the theory of multiple intelligence, has 12 types of intelligence. [26] Fortunately for mankind, neural networks can qualitatively simulate only 2 types: bodily-kinetic and logical-mathematical. However, this is enough to use neural networks to gain new knowledge, including in the field of colonization of celestial bodies.
During the simulation, the neural network accumulates experience, on the basis of which it makes assumptions about how best to arrange colony buildings. She checks all her assumptions using the created Pluto Echo model. The result of the simulation becomes her new experience. This process is cyclical and repeats until the neural network develops rules for the correct placement of modules relative to the landscape. This is an implementation of the generative adversarial network (GAN) technology described in 2014. [27] In this network, the model is a generator, and the neural network is a discriminator. [28]
The practical part
The general principle of the Pluto Echo model
The Pluto Echo model was created using the Blitz BASIC programming language. [29]
The model plays out the task of arranging colony buildings in order to determine such an arrangement in which the colonist, in the course of his activity (sampling), spent as little energy and time as possible on completing his task. The number of sampling sites varies with each playback of the model in the range from 10 to 30 pieces.
During his activity, the colonist gets tired both physically and mentally. To simulate the processes of fatigue accumulation and rest, a two-parameter "critical power" model is used, which is used in all international sports competitions. [30]
Moving along the surface, the colonist encounters the difficulties of the local terrain – craters. During their passage, the speed of movement decreases by 30%, and the expenditure of physical and mental strength increases by 50%. There may also be various buildings of the colony itself on the way of the colonist, which he will have to bypass, wasting time and effort.
The result of the simulation cannot be mathematically calculated, because it consists of the totality of the flow of all processes that affect each other and have a share of randomness (as in real life). The final result of the simulation can be found out only by playing the model completely.
Colony structure
The colony consists of 4 modules:
The residential module is a colony module in which the collected samples are stored and the colonist rests;
Communication module – a module that performs the function of communication with the Earth;
A nuclear reactor is a module designed to provide a colony with electricity;
A solar battery is a module designed to provide a colony with electricity from sunlight.
This colony structure was developed by the author together with the domestic neural network "YaGPT" from the Russian company "Yandex". [31]
The appearance of all modules is shown in Figures 1-4 of Appendix B.
The graphical part of the model
All modules are implemented in the form of graphic images in BMP format and are made in the form of pseudo-3D graphics. [32][33] These images, with the exception of the image of the residential module, were created by the Realistic Vision 4.0 neural network.[34]
The colonist's living module was created in a graphic editor by the author himself.
The surface of Pluto and its craters were created based on real NASA images taken by the automatic interplanetary station "New Horizons", which reached Pluto in 2015. [35]
Implementation of a self-learning algorithm
The Pluto Echo model has a built-in self-learning algorithm that works in streaming mode.
This algorithm works as follows:
Implements one of the possible options for the arrangement of colony buildings;
Determines the effectiveness of the chosen solution through modeling;
Compares the obtained efficiency with the previous most effective option;
If the modeled version of the arrangement is more efficient, then the model stores information about the new version in binary format;
If the modeled version of the arrangement is less effective, then the model does not remember it and proceeds to step 1 of this algorithm.
Efficiency refers to the amount of time a colonist spends taking samples.
Given that this algorithm works by iterating through all possible arrangements of buildings, it is linear (linear or sequential search algorithm). Its complexity is defined as O(n)(1): [36]
On=the number of possible object placement variantsreferences
(1)
On=(800∙600)(4+30)≈7∙10198
The scheme of the algorithm is shown in Figure 1 of Appendix B.
Neural network implementation
The Pluto Echo model has a built-in neural network consisting of 119 neurons arranged in 6 layers. Its implementation was carried out according to the main principle of design - the principle of "do not complicate" (KISS principle). [37] It consists of 19 one-dimensional arrays of Float variables of different dimensions. [38] All neurons have no activation function and process absolutely all signals of any magnitude. The sigmoid is not used.
At its core, the neural network solves the regression problem – to take the coordinates of all the places where you want to take a sample, and give out the coordinates where it is best to put the colonist's residential module. [39] The gradient descent method is used to solve this problem. [40]
The processing of neuronal signals is carried out according to the following algorithm:
Summation of all signals entering the neuron;
Multiplying the result by a coefficient (link weight);
Transmitting the received value to the output;
The mathematical description of this process is presented in the form of formula (2):
ni, j=j=1jn(i-1,j)∙k(i,j)
(2)
where ni, j is the neuron j of layer i, k(i,j) is the output coupling coefficient of the neuron j of layer i.
The neural network is trained using a method called by the author the "minimum addition method". This is one of the variations of the error back propagation method, which refers to the methods of supervised learning. [41] The author has not encountered this method in the literature and is probably being proposed for the first time.
The learning algorithm consists of the following sequence of actions:
Calculate the neural network error (O);
If the error is greater than or less than zero, determine the error sign;
If the error sign is positive, then subtract the minimum addition constant (α) from all weights;
If the error sign is negative, then add the minimum addition constant (α) to all weights.
The mathematical description of the method is presented in formula (3).
The minimum addition coefficient is set at the discretion of the neural network developer. In the Pluto Echo model, this coefficient is 0.01.
k(i,j)=k(i,j)+α if O<0k(i,j)=k(i,j)-α if O>0
(3)
where: k(i,j) is the output coefficient of the neuron j of layer i, α is the coefficient of minimum addition, O is the error.
The initial values of the coefficients of neural connections are set in the range 0≤k≤δ randomly, where δ is determined by the formula (3):
δ=1x∙k
(3)
where: δ is the maximum value of the range, x is the number of neurons on the previous layer, k is the tuning coefficient determined empirically.
This formula is a modified formula by the author, generally accepted among neural network programmers (the k coefficient has been added). [42]
In the Pluto Echo model, the value of k was experimentally determined to be 5.
Format for saving simulation results
The simulation results are saved as separate CSV files:
The "Human" mode saves data on physical and mental fatigue, time spent and the number of samples taken in the "oth.csv" file.
The "self-learning algorithm" mode saves the coordinates of all the buildings of the best option in the "algorithm.csv" file.
The "neural network" mode stores the values of all weights of neural connections of the neural network in the file "neiro.csv".
It is worth noting that the file "neiro.csv", at its core, is the experience gained as a result of training a neural network. It can be used to transfer this experience to other programs and neural networks (to implement learning transfer). [43]
The CSV format is a common and publicly available format for storing tabular data. Due to this, the obtained results can be opened and analyzed in any analytical program or DBMS. [44]
Conclusion
Humanity has a lot of knowledge about the colonization of celestial bodies, but most of it is just a theory that needs to be tested in practice.
Computer modeling can significantly reduce the cost of solving this problem by screening out non-working methods at the early stages of verification.
Object-oriented simulation is most suitable for testing existing theories.
As a result of the conducted research, the existing methods of computer modeling were studied, their disadvantages were revealed. An object-oriented simulation model "Echo of Pluto" was created.
The created model embodies all three types of forecasting.
The created model uses all the latest developments of mankind in the field of artificial intelligence: a generative adversarial network (GAN), a self-learning algorithm and a neural network. The possibility of transferring training has been implemented.
The neural network was created on the principle of "do not complicate" (KISS principle) based on the method of "minimal addition" without activation function (sigmoids).
During the practical part of the study, the following was carried out:
The effectiveness and usefulness of using IOM and artificial intelligence technologies to test existing technologies for colonization of celestial bodies has been proven in practice.
The effectiveness and usefulness of the use of IOM and technologies in the field of artificial intelligence to create new knowledge in the field of colonization of celestial bodies has been proven in practice.
The purpose of the study has been achieved.