The Key Technologies Of Association Rules Mining For Situation Assessment

The modern war is based on information.The key for the result of war is whether a commander can be aware of the situation by the advantage of information.Therefore,situation assessment(SA)is irreplaceable in the modern war.Most traditional methods building the SA model depends on priori knowledge,selected parameters and supervised data,which can not update themselves dynamically.Therefore,many researches try to apply association rules mining to SA,which gets the model from unsupervised data.However,most studies do not take the complexity of data got from the war into accounts.This thesis focuses on the association rules mining for SA,which considers some characteristics of the data got from battlefield,which are the multi-type of parameter,the large scale,the high dimension and the distortion.The detail works are shown as the follows.1.A central limit theorem(CLT)based algorithm,which is called CLTF,is proposed to mine the fuzzy association rules among battlefield objects.CLTF can transform the object data of the battlefield to the fuzzy transactional dataset.Then,through the data reduction based on CLT,it can fast form the fuzzy association rules constructed by the objects of war.CLTF trys to solve two problems.Firstly,the transactional dataset used by association rules mining can not be directly obtained from the battlefield,and when people try to transform the object data to the transactional dataset,discreting the numerical parameters of objects is very difficult.Secondly,the large scale of data generated by the war is a big challenge.For these two problems,CLTF uses membership functions to divide numerical parameters when forming the fuzzy transactional dataset,transforms the fuzzy transactional dataset to the virtual binary transactional datasets based on the characteristic of product t-norm,and takes CLT to build a sample of original data to speed up the mining process.Furthermore,the errors caused by sampling based on CLT can be controlled in a very small range with high probability.2.A data reduction algorithm based on central limit theorem and Union bound,which is called CLTU,is proposed.CLTU can efficiently reduce the large scale of transactional dataset which is transformed from the object data of battlefield,and can provide the tight guarantee of the results mined from the reduced data.CLTU is an extention of the sampling part of CLTF.Although the sampling in CLTF an ensure that any error of support can be controlled in a small range,these errors can be accumulated,which is a hidden trouble.Therefore,CLTU controls the error more strictly,which can ensure that the maximum error of all the supports is limited in a given range with a given least probability.Furthermore,the CLTU takes pre-sampling,sampling and transaction combination to avoid the over-estimated scale of the reduced data.3.An binary particle swarm optimization(BPSO)based algorithm,which is called BPSOHD,is proposed to mine the long pattern from high dimensional data of battlefied.BPSOHD can efficiently searches the long pattern in a large scale of targets and events of battlefield.There are two difficulties of mining long patterns.Firstly,the support of long pattern is very small,which is easily confused with the noisy pattern,so the minimum support threshold is hard to set.Then,the high deimensionality of data got from the battlefield hides the real long pattern in too many potential results.Therefore,BPSOHD takes the BPSO to search the long pattern without the minimum support threshold.Meanwhile,BPSOHD improves the BPSO by reducing the dimensionality of initial particles and reducing the dimensionality of dataset dynamically,so the efficiency of BPSO is enhanced.4.An algorithm to rank and map association rules,which is called ARSA,is proposed based on data distortion.ARSA can reasonably select association rules and uses them to assess the situation when the data used to form association rules is distortion.Because the reconnaissance system can not totally avoid missing some objects and misrecognition,the association rules got from the data is not the same as the objective rules.Therefore,through some reasonable hypothesis,we theoretically analyze the effect of data distortion.Through the analysis,ARSA sorts all the rules and selects those rules which are both mapped by the current known objects and with high ranks.ARSA can reduce the negative effect caused by the data distortion to some extension.Based on the above studies,we design and realize a prototype system.Through an example,the reliability and efficiency of this system is shown.