Reactive Load Forecasting Model Based on PSR-LSTM in Power System

To optimize reactive power control strategy, in allusion to randomness and nonlinearity of reactive load a reactive load forecasting model based on phase space reconstruction and long short term memory neural network was proposed to improve voltage quality and reduce network loss. Utilizing C-C method the optimal reconstruction dimension and latency time were determined and by means of calculating the maximal Lyapunov exponent the chaos characteristic of reactive load was illustrated. The phase space reconstruction technology was used to map the reactive power sequence to high-dimensional space and in high-dimensional space the long short-term memory (abbr. LSTM) neural network is utilized to perform the prediction. Finally, taking data of active and reactive power load of a certain region in Hainan province for example, by use of Kolmogorov entropy it was verified that the chaos degree of reactive power load was larger than the chaos degree of active power. The feasibility of the proposed method is verified by results of calculation example. Using the proposed method the accuracy of reactive load prediction may be improved and the dispatching and controlling of reactive power in the power system may be more rationally.