Power marketing risk prediction optimization method

The invention discloses an electric power marketing risk prediction optimization method, and belongs to the field of data prediction, and the method comprises the steps: S1, carrying out the preprocessing of data affecting the electric power marketing risk prediction; s2, using a'relative convergence 'strategy model as a position updating mathematical model of a narrow relativistic search algorithm in a local convergence stage, and improving a position updating strategy of the narrow relativistic algorithm; s3, improving a particle optimization weight factor of a narrow relativistic search algorithm by using the fitness value variation of the particle position, the current iteration particle speed and the optimal fitness value, and optimizing the position to update the mathematical model; s4, optimizing a hidden layer node number s and an initial learning rate alpha parameter of the LSTM model by using an improved narrow relativistic algorithm to obtain a beta SRS-LSTM model; and S5, taking the preprocessed d