基于改进神经网络算法的光伏发电功率短期预测

光伏发电技术因其清洁无污染、安装便利、维护成本低和使用效率高等优势近年来获得了快速的发展,但是光伏输出功率具有明显的随机性和不确定性,当其大规模接入电网后其波动特性表现的更为突出,给电网带来巨大冲击的同时降低了电网运行的可靠性,增添了电网调度运行管理的成本与难度。针对此问题本文提出一种基于粒子群算法和神经网络算法的组合预测方法对光伏发电功率进行短期预测,对传统神经网络功率预测算法寻优性能欠佳的问题进行改善,利用粒子群算法对输入样本进行合理优化,同时利用变步长的动量梯度法对神经学习因子进行不断修正,形成一种组合的功率预测方法用于光伏功率预测。仿真结果表明本文预测模型在日类型天气为晴朗天气时的预测结果最好,精度提升相比传统方法来说13%左右。

Short Term Prediction of Photovoltaic Power Generation Based on Improved Neural Network Algorithm

Photovoltaics technology has gained rapid development in recent years due to its clean,non-polluting,convenient installation,low maintenance costs and high efficiency of use.However,the output power of photovoltaic has obvious randomness and uncertainty.When it is connected to the power grid on a large scale,its wave characteristics are more prominent.It brings huge impact to the power grid and reduces the reliability of the power grid operation,and increases the cost and difficulty of the power grid dispatching operation management.In this paper,a combination prediction method based on particle swarm algorithm and neural network algorithm is proposed to predict the power of photovoltaic power generation in the short term.The particle swarm algorithm is used to optimize the input sample,and the neural learning factor is modified by the momentum gradient method with variable step length.A combined power prediction method is formed for PV power prediction.The simulation results show that the prediction results of this model are best when the daily weather is sunny,and the accuracy improvement is about 13%compared with the traditional method.