粒子群神经网络混合算法在负荷预测中的应用

电力系统负荷预测结果的准确性关系到电力系统的调度运行、生产计划和供电质量,为此在研究短期负荷预测中应用了粒子群PSO和BP神经网络相结合的混合算法。该算法先应用粒子群优化算法算出BP神经网络的连接权向量和阈值,每次迭代求出最优粒子的权向量和阈值及BP网络在这组权向量和阈值的实际输出值,最后得出第i个粒子的适应度函数。与其他方法相比,该算法预测精度较高:平均相对误差≤1.48%,最大相对误差≤4.10%,而且收敛速度快,预测结果满足短期负荷预测误差要求。

Application of Particle Group and Neural Network Hybrid Algorithm in Load Forecast

The load forecast of power system is one of the important tasks of power dispatch and service department,whose accuracy has a close relation with dispatch operation,production plan and quality of power supply.So going further to the research of load forecasting method and model has important realistic meanings.The hybrid algorithm which combines the particle swarm optimization with BP neural network is applied in the short-term daily load forecasting of power system,and it has a good accuracy.The basic principle of particle swarm optimization algorithm is discussed,and the operational method and procedure of the particle swarm-neural network hybrid algorithm are given.Then by the XOR problem,the PSO-BP hybrid algorithm is proved to have a quicker convergence and a higher computational precision.Then,the daily load forecasting model using the PSO-BP hybrid algorithm is established,which has been applied in the daily load forecasting of a city.Compared the predicted results with other used methods,the PSO-BP hybrid algorithm is proved to have a higher prediction accuracy: the average relative error is not more than 1.48% and the maximum relative error is not more than 4.10%.From the comparison we could find that while the PSO-BP hybrid algorithm is applied in the short-term load forecasting,it has a quick convergence,a high computational precision and a low error,the predicted results could meet the basic requirements of error in short-term load forecasting.