基于深度脊波神经网络的电力系统短期负荷预测模型

Short-Term Load Forecasting Model of Power System Based on Deep Ridgelet Neural Network

针对电力系统负荷的非线性预测问题,本文构造了一种基于深度脊波神经网络的电力系统短期负荷预测模型。该模型的隐含层采用脊波神经元,神经元的激励函数采用脊波变换函数。对该预测模型采用受限的玻尔兹曼机学习原理进行预训练,最后利用粒子群优化算法对其进行深度优化精调。通过对某地区实际电网负荷系统进行仿真预测,结果表明,与传统的BP神经网络、脊波神经网络和常规深度神经网络模型相对比,深度脊波神经网络预测模型的日平均绝对误差百分比分别降低了1.96%、1.12%和0.3%,日最大绝对误差分别降低了3.91%、2.19%和1.78%,验证了深度脊波神经网络预测模型具有较好的预测准确度和稳定性。

To solve the nonlinear forecasting problem of power system load,a short-term load forecasting(STLF)model based on deep ridgelet neural network(DRNN)is constructed,in which the hidden layer adopts ridgelet neurons and the ridgelet transform function is taken as their excitation function. This prediction model is pre-trained according to the restricted Boltzmann machine(RBM)learning principle,and it is further deeply optimized using the particle swarm optimization(PSO)algorithm. An actual grid load system in one certain region is predicted through simulations,and results show that compared with models based on the traditional BP neural network,ridgelet neural network,and the conventional deep neural network(DNN),the daily mean absolute percentage error of the prediction model based on DRNN is reduced by 1.96%,1.12%,and 0.3%,respectively,and its daily maximum absolute error is reduced by3.91%,2.19%,and 1.78%,respectively,showing that the novel model has higher prediction accuracy and better predictive stability.