The Research Of Mining Land Evaluation Classification Rules Based On Population-based Intelligent Optimization Algorithm

With the development of GIS and database technology, people can manage, inquire and storage land use data expediently, which can provide abundant basic data for land evaluation, land use planning etc.. However, the resource data rooting in different department take on the characteristic such as various type, different scale, complicated relationship etc., which can bring out the phenomenon that people hold on lots of land data but lack of knowledge. How to acquire the interesting knowledge and pattern behind land use data to solve different land problem is a question need to resolve urgently. Spatial data mining, as an important embranchment of data mining, can mine the rules and pattern people interested in according to the guiding by requirement and field knowledge. So far, spatial data mining is applied to remote sensing classification, simulation of land use changes etc..Land evaluation is a main research direction of land management, can prepare for land use planning, land use decision support. It is the process of estimate land quality according as evaluation purpose and type and based on the land evaluation factors impacting to land quality. Spatial classification divides the original data into training set and test set, builds spatial classification model based on classification algorithm to train on training set to obtain mapping relationship, usually using spatial classification rules, between condition attributes and decision attributes, after testing the rules on test set and proving the model is rational, the built spatial classification model can be extended and applied to other data. Based on the above theory, we can conclude the comparability between land evaluation and spatial classification, so this article tries to build land evaluation model by spatial data mining, sets evaluation factors and results to be condition attributes and decision attributes respectively, trains the associated relationship between evaluation factors and evaluation results using classification algorithm to be the land evaluation knowledge. Obviously, this method avoids subjectivity and randomicity without depending on the weights of factors. From another point of view, mining spatial classification rules is just a optimized process to find the best rule set from origin database, so the author introduces ant colony optimization algorithm, which is representational in population-based intelligent optimization algorithm, into mining land evaluation spatial classification rules, and in allusion to the disadvantage of ant colony optimization algorithm, uses immune algorithm to improve on the ant colony optimization algorithm’s results. Consequently, the thesis achieves coupling among land evaluation, spatial data mining and population-based intelligent optimization algorithm.In the first instance, the thesis expatiates the research status of land evaluation theory and methods, concludes evaluation methods in common use recently and their disadvantage. Aiming at the deficiency, the article brings forward the thought of mining land evaluation classification rules based on population-based intelligent optimization algorithm in the direction of land evaluation, spatial data mining and population-based intelligent optimization algorithm.Spatial data mining deals with many aspects, such as data, knowledge, training, test etc, so the author selects integration pattern among GIS, land evaluation and spatial data mining, frames the land evaluation classification rules mining architecture based on GIS. This architecture includes four-layer:data layer, knowledge layer, mining layer and man-machine interaction layer, it is constructed guiding by the coupling between land evaluation and spatial data mining, along with the inner cooperation relationship. They are cooperate with each other, data layer can prepare database for mining layer; The tasks of mining layer are restricted and leaded by algorithm base, knowledge base and model base in knowledge layer; Person is the communication bridge between mining layer and knowledge layer, whose prescription can lead the mining tasks, computer provides the friendship visualization analysis for person by sharing data mechanism.Spatial data contain variety, temporal-spatial diversity and redundancy, before we mine land evaluation classification rules, we need to integrate the multi-source heterogenous spatial data. Combining with the character of land evaluation data, the author research the spatial data conformity surrounding four aspects, which are spatial data conversion, spatial data cleaning, spatial data integration and spatial data reduction. And then, the thesis builds the land evaluation spatial data conformity system under model base and algorithm base. The key research topic of this dissertation is to build land evaluation model based on spatial data mining. Based on the theory of building spatial classification model, this dissertation expatiates the model building from data structure design, spatial data sampling, classification model training, classification rules pruning and classification model test. Firstly, data structure is designed based on the training algorithm, training set and test set is sampled based on sampling principle. Secondly, the author presents the amalgamation among artificial ant colony system, artificial immune system, land evaluation and spatial data mining, which means to introduce population-based intelligent optimization algorithm into land evaluation classification rules mining field. There might be some redundant factors or rules mined by ant colony optimization algorithm and immune algorithm, so we must add the rules pruning step after model training, there are two different pruning strategy, namely rules attributes pruning and redundant rules pruning, and applies immune algorithm to prune the rule set to attain the last rule set. Lastly, test the rule set, if rational, we can apply the land evaluation model based on spatial data mining and population-based intelligent optimization algorithm.Another core topic in this dissertation is the design of population-based intelligent optimization algorithm which using in land evaluation classification rule mining. Three-layer algorithm structure, namely input layer, mining layer and output layer, is designed after being familiar with the algorithm mechanism at first. Whereafter, the math description of algorithm, training flow and arithmetic operators are discussed detailedly in classification model training and classification rules pruning. To prove the rationality and feasibility of architecture, conformity frame, spatial data mining model and optimization algorithm described above, the dissertation takes Puning city, Guangdong province for example, codes programe by Matlab, integrates the model and algorithm to achieve farmland grade rules mining automatically in Puning city. Contrasting basic ant colony optimization algorithm to ant colony optimization algorithm based on immune algorithm, the results of ant colony optimization algorithm based on immue algorithm are better than another one, so the author selects the former results to guide other data classification. Afterwards, the thesis analyzes the relationship between farmland grade and natural condition, economic fators and influencing degree of road to explain the rationality of farmland grade conclude by this article’s method. Finally, compared with decision tree and...