Implementation of land redistribution and readjustment processes in zoning applications using artificial intelligence optimization algorithms

Land redistribution and readjustment (LR) problems in urban areas are complex and difficult real-world problems involving many different criteria. In order to solve these problems, first of all, pre-parceling operations should be carried out successfully. While any random candidate solution in the initial population which is created for the solution of the problem is generating or it is updating, the area of the parcels is updated at each iteration and accordingly, the location of each parcel, in other words, its border points must be updated again. In this dissertation study, the pre-parceling process is automatically carried out in three different ways on a region of Konya considering the zoning legislation and a sample development plan: The first is the parceling process according to edge lengths, the second is parceling according to area values and third is parceling using both edge and area values together. In all these parceling operations, the areas of the parcels are calculated using the Gauss method. Moreover, the binary search technique is used in all methods to determine the boundaries of the parcel accurately and to calculate its area in the parceling process. Experimental results show that pre-parceling processes are carried out very quickly and successfully. LR problems are discrete optimization problems that are similar to scheduling and traveling salesman problems in terms of their nature. In addition, these problems have too many criteria to be dealt with simultaneously. Therefore, solving these problems requires artificial intelligence optimization algorithms instead of classical methods. For solving these problems, first of all, an objective function is needed that will enable the algorithms to make the right decision and evaluate the solutions objectively among the candidate solutions. Thanks to this objective function, which is the basic criterion in solving LR problems, the quality of distribution and subdivision plans can be evaluated and compared without the need for any specialist. In this study, an objective function that takes all criteria into account is proposed. In addition, in this dissertation, two different map-based crossover operators called classical and intelligent parcel-based crossover operators are proposed different from crossover techniques in the literature. The discrete artificial bee colony (ABC), differential evolution (DE), genetic (GA), particle swarm optimization (PSO) and tree seed (TSA) algorithms which are proposed in this dissertation study are implemented on a real project area with the help of the proposed objective function and these operators. The results in the experimental studies are compared with the official results obtained manually. In addition to these, thanks to the expert system developed, the gaps and overflowing areas in the urban blocks are completely eliminated and the solutions obtained by algorithms are made available in the real world. When the experimental results are examined, it is clearly seen that all the results obtained by the artificial intelligence algorithms are much more effective than the real results obtained manually in terms of both solution quality and speed. In addition, it is seen that the PSO algorithm is much more effective and robust than other algorithms. In addition, it is seen that the proposed intelligent parcel based crossover operator achieves much more effective results than the classical parcel based crossover operator. Due to the very complex structure of real world problems in zoning applications, the optimal value is not known exactly. Therefore, a synthetic data set is proposed to be used as a benchmarking data set in LR applications whose optimal solution of the problems is certainly known. This data set consists of 5 different problems, 20, 40, 60, 80 and 100, depending on the number of parcels. Each problem set contains 4 different problems as 1, 2, 3 and 4 in terms of the number of land owners per parcel. Therefore, the data set consists of 20 different problems in total. When the experimental results using the synthetic data set are examined, it is seen that the GA method is the most effective algorithm in terms of both speed and performance. Although ABC has better results than the GA in a few problems, ABC is the second most successful algorithm after the GA method in terms of performance. However in terms of time, ABC is an algorithm almost as successful as the genetic algorithm. On the other hand, the results of DE, PSO and TSA algorithms are similar in terms of solution quality. As a result, the experimental results clearly show that the GA method is the most effective technique in solving LR problems in terms of speed, performance and robustness.