嵌入维数自适应最小二乘支持向量机状态时间序列预测方法

Condition Time Series Prediction Using Least Squares Support Vector Machine with Adaptive Embedding Dimension

针对航空发动机状态时间序列预测中嵌入维数难于有效选取的问题,提出一种基于嵌入维数自适应最小二乘支持向量机(LSSVM)的预测方法。该方法将嵌入维数作为影响状态时间序列预测精度的重要参数,以交叉验证误差为评价准则,利用粒子群优化(PSO)进化搜索LSSVM预测模型的最优超参数与嵌入维数,同时通过矩阵变换原理提高交叉验证过程的计算效率,并最终建立优化后的LSSVM预测模型。航空发动机排气温度(EGT)预测实例表明,该方法可自适应选取适用于状态时间序列预测的最优嵌入维数且预测精度高,适用于航空发动机状态时间序列预测。

To deal with the difficulty of selecting an appropriate embedding dimension for aeroengine condition time series prediction,a method based on least squares support vector machine（LSSVM）with adaptive embedding dimension is proposed.In the method,the embedding dimension is identified as a parameter that affects the accuracy of the aeroengine condition time series prediction;particle swarm optimization（PSO）is applied to optimize the hyperparameters and embedding dimension of the LSSVM prediction model;cross-validation is applied to evaluate the performance of the LSSVM prediction model;and matrix transform is applied to the LSSVM prediction model training to accelerate the cross-validation evaluation process.Experiments on an aeroengine exhaust gas temperature（EGT）prediction demonstrates that the method is highly effective in embedding dimension selection.In comparison with conventional aeroengine condition time series prediction methods,the LSSVM prediction model with the optimized hyperparameters and embedding dimension has better prediction performance.