The invention discloses a decomposition and optimization based short-term wind speed time series prediction method. The method comprises the steps of S1, performing original wind speed series sampling and processing by hour to construct a time series data sequence; S2, performing three-layer wavelet decomposition on the time series wind speed to obtain high-frequency detailed series d1, d2 and d3 and a low-frequency approximate series a3; S3, performing binary interpolation reconstruction on the decomposed series to obtain detailed series D2, D2 and D3 and an approximate series A3; S4, performing normalization processing on each reconstructed series to construct a time series correlation sequence group, and taking the final 24 groups as a test set and the rest groups as a training set; S5, by using data of the training set, optimizing parameters of a least square support vector machine (LS-SVM) by utilizing a chaotic firefly algorithm to construct an optimal LS-SVM; S6, inputting the test set into the optimal LS-SVM in the step S5 to perform wind speed prediction; and S7, reversely normalizing and superposing prediction values to obtain an actual wind speed prediction value. According to the method, the short-term wind speed prediction precision is improved, more effective decision information can be provided for operating personnel, and the safety and reliability of wind power operation are ensured.