Research Of Efficient Neighbor Query On Agent Model For Evolutionary Computation Based On LSH

In recent years,evolution algorithm gradually become an efficient way to solve the complex problem that traditional optimization methods cannot deal with.But the calculation of fitness requires a lot time.One of the solutions to reduce time consumption is to estimate fitness by agent model.In general,agent model queries neighborhood data by comparing the distance between the position of target data and that of other data one by one which,as a result,causes a high time complexity.To reduce time spent on querying neighborhood data in agent model,the locality sensitive hashing(LSH)algorithm is applied in this article based on LSH using p-stable distribution,presenting LSH using collision p-stable distribution.The algorithm reduces the number of hash function and chooses the data with a high collision rate as candidate set,thus can query neighborhood data more efficiently.To test the modified algorithm,two high-dimensional data is performed to have a test,combining it with surrogate algorithm,adding it to particle swarm optimization.Finally,the test using four test functions verify the efficiency of the algorithm in evolutionary algorithm agent model.Based upon the research,the design realizes the key-value database prototype of evolutionary computation in historical data.Besides basic function of CURD,the system also adds the function of neighbor query to improve the efficiency of neighborhood query in agent model.The neighborhood query is implemented by using collision p-stable LSH.In addition,when there is excessive in-memory data,it will store them to disk regularly and build a multistage structure to divide the hot data and cold data,so as to realize persistent storage and efficient inquiry in external storage.