Abstract:In order to avoid the issue of inversing matrix existed in LS-SVM (Least Squares Support Vector Machines) algorithm,the paper presented an improved LS-SVM algorithm: using improved PSO (Particle Swarm Optimization) algorithm to iteratively and optimally solve the linear equation set in LS-SVM algorithm,which can not only speed the training of the algorithm and save the memory,but also get least square solution,raising the computational precision.Using the improved algorithm to recognize the gas-bearing bed in the target stratum of well-C in Changqing gasfield and for comparison with BP neural network algorithm,classic SVM algorithm and traditional LS-SVM algorithm,the results showed the algorithm is characteristics of higher recognition precision,fast convergence speed and coincident to the result of well test,having excellent effect.