Urbana mobilnost stanovnika uobičajeno se procjenjuje postupcima usredotočenim na specifične domene (promet, logistika, ekologija, društvo i dr.), uz korištenje domenski ograničenih skupova podataka, pokazatelja i indeksa vezanih za ciljane podskupove ukupne urbane populacije (korisnici javnog prijevoza, biciklisti, vozači i dr.). U radu je predložen novi pristup procjene urbane mobilnosti stanovnika temeljen na podacima o telekomunikacijskoj aktivnosti (glasovni poziv, tekstualna poruka, pristup internetu) korisnika u javnim pokretnim telekomunikacijskim mrežama. Koristi se pristup istraživanja urbane mobilnosti stanovnika unutar domene inteligentnih transportnih sustava, kao informacijsko-komunikacijske nadgradnje klasičnog sustava prometa i transporta. Pokazatelji urbane mobilnosti u radu se izvode iz podataka o telekomunikacijskim aktivnostima i ujedinjuju se u indeks urbane mobilnosti stanovnika korištenjem metodologije višeslojnog prilagodljivog sustava neizrazitog zaključivanja zasnovanog na neuronskoj mreži. Uspostavljen je kompleksan postupak procjene urbane mobilnosti korištenjem modela indeksa urbane mobilnosti, koji je definiran postupkom strojnog učenja ANFIS (engl. adaptive neuro-fuzzy inference system). Razmotren je početni sustav neizrazitog zaključivanja, učenje modela, provjera kvalitete modela te su razmotrena ograničenja, pogreške i nedostaci. Model je praktično primijenjen u programskom okružju na skupu eksperimentalno prikupljenih podataka.
|Parallel abstract (English)|| |
The purpose of this PhD thesis was to prove the possibility of application of telecommunication users' activities data (Call Data Records) to suit the needs of providing an urban mobility estimate, then to identify urban mobility indicators, to define the process of their calculation, and define a model which will from the parameters of these indicators give an estimate of urban mobility. The defined goal of my research was to establish a process of the population's urban mobility estimate as a quantitative measure of the process of urban migrations caused by socio-economic activities, one which is in the function of determining a new urban mobility estimation index. A structured record of users' network activities within a public mobile communication network used for the purpose of charging telecommunication services was used as a primary research material. An algorithm that uses this information as a basis to identify the movement of users within an urban agglomeration covered by this research was developed. Based on the information about the identified migrations, the indicators of urban movement were defined, and the process of their calculation was determined and validated. What then follows is a procedure in which the relation between the values of urban mobility indicators and values of mobility estimation is determined. This relation was defined though the method of surveys. The survey was defined in such a way so that through a number of questions expert opinions could be gathered on how and to what degree the combination of values of certain urban mobility indicators affects mobility. According to their own judgement, experts allocated an appropriate value to each question using the suggested procedure for mobility estimation. The method of fuzzy logic, that is, the ANFIS (adaptive neuro-fuzzy inference system) method which functions on the principle of applying conclusion methods which characterise neural networks with the goal of determining parameters of an indirect conclusion system (Fuzzy Inference System – FIS) was used. The result is a system that permits the usage of well-known neural network learning algorithms, ones that cannot be used in fuzzy logic systems, while at the same preserving the possibility of using fuzzy logic. The answers of experts were used as an initial collection of data to establish the model. The model is described with 27 rules, each of which has its own interval of exit mobility estimate values dependent on entry parameter values of entry indicators, defined with a total of 32 fuzzy logic systems which were assigned their appropriate affiliation functions. A form of the exit variable and a learning (optimisation) method as well as the number of epochs were defined. Each of the established fuzzy logic systems were tested for reliability using the RMSE method. The smallest learning mistake was generally observed in all scenarios which used the hybrid learning method and in which the exit function was in linear form. The average model validation error (for a model chosen as optimal) is 0,2103, and the average model validation error is 0,1907. As the most optimal model, the first-class fuzzy logic Sugeno model was chosen, one in which all three entry values are determined with a trapezoid type affiliation function, and each consists of three overlapping linear type functions. A hybrid learning model with three learning epochs was used. The result of the model is an estimate of urban mobility for the matching pair of urban areas (partial mobility index), which is then used to calculate the mobility index of the entire urban agglomeration in the process of urban mobility estimation. Implementation of the urban mobility estimation process was executed using several programming environments. Data collection loading, entry data table conversion and creating the base cell table as well as the decomposition of space was handled with a programming code developed in the R programming environment. The open source geographic analysis programming tool QGis was used for space decomposition visualisation. Validation is executed though several steps. Validation of the ANFIS model is part of an integral process of fuzzy logic system forming. What follows then is the validation of algorithms which were defined as part of the mobility validation procedure. The point of validation in this part of the procedure is to confirm that the defined algorithms and the written programming support are correctly calculating the segment they were assigned to. The validation of all algorithms was successful, with which the fact that all written algorithms are properly executing their assigned operations was confirmed. Then comes the model results validation. According to indicator calculation a data segment of 20% for each time frame is extracted, and based on them, partial indexes and the urban mobility index are calculated. Then, the results found in that data are compared with results from the remaining data in the segment, and the discrepancy between results is determined. The procedure of urban mobility estimation was applied on a real publicly available sub-segment of data taken from a record used to charge telecommunication services in a public mobile communication network. The result of the research proves the hypothesis of this scientific research paper which states that the urban mobility of inhabitants within an urban environment can in a specific time frame be described with an index of urban mobility based on information about recorded telecommunication activities of users in a public mobile communication network. The goal of the paper was fulfilled and the possibility of telecommunication users' activities data application for the purpose of an urban mobility estimate was proven. Indicators of urban mobility and the process of their calculation were defined, followed by a model which will from the values of these indicators give an estimate of urban mobility. A procedure of the population's urban mobility estimate as a quantitative measure of urban migrations caused by socio-economic activities.