Short-term Power Load Forecasting Based On Improved Particle Swarm Optimization Algorithm

With economic development,power load forecasting is playing an increasingly important role.Accurate load forecasting can be used as a basis for unit maintenance and personnel transfer.Although the development of smart grid has reduced the pressure on the main grid,the factors of load forecasting are more complicated,which puts forward higher requirements on the accuracy of load forecasting.The work done in this article is as follows(1)Compared with traditional prediction algorithms,BP neural network model has strong learning ability and self-adaptive ability,and strong ability to fit nonlinear data.But the iteration speed is slow and the overall situation is poor.Although the PSO-BP prediction model has improved the prediction accuracy,it is still easy to converge prematurely and fall into local extremes.This paper proposes an algorithm based on the variance of Euclidean distance to characterize population diversity(PA);and based on the probability jump characteristics of simulated annealing algorithm,proposes a simulated annealing(PASA)algorithm based on population diversity.We use the PASA-PSO algorithm to optimize the neural network parameters.The prediction results show that the PASA algorithm not only improve s the diversity of the particle population,but also avoids early convergence due to the sudden jump of probability.The algorithm reduces the average relative error from 4.335% of the PSO model to 3.775%.(2)This paper also proposes a reverse learning initialization strategy based on p-probability.Starting from the initialization of particle swarms,a P-persistent reverse learning initialization particle algorithm(p-OBL)is proposed.Compared with randomly initialized particles,p-OBL tends to retain particles with higher fitness values and randomly retains particles with poor fitness values.This ensures that all particles are closer to the optimal solution and avoid uneven particle distribution.The prediction results show that the p-OB algorithm has a positive effect on the accuracy of the algorithm.(3)The data determines the upper limit of the load forecast accuracy.First of all,this paper selects 55 attributes based on experience to establish the initial decision table.After preprocessing the data set,this paper uses rough set-based attribute reduction algorithm to reduce these 55 attributes to only 11 attributes,and input these11 attributes into the neural network model for load forecasting.