ANFIS在短期负荷预测中的应用

Application of Adaptive Neuro-fuzzy Inference System to Short-term Load Forecasting

为使负荷预测更精确,鉴于预测对象的不确定性和非线性,采用ANFIS预测电力系统短期负荷。ANFIS将模糊理论与神经网络融合,利用神经网络实现系统的模糊逻辑推理,采用混合学习算法调整前提参数和结论参数,自动产生模糊规则。该系统具有非线性映射和自学习能力,不基于数学模型,用独特的空间分层方法建立若干模糊推理系统,依靠专家经验获取控制信息,能用于负荷预测的非线性建模,获取负荷数据的最佳估计,克服数据处理过程中存在的不确定性和不完备性。所用ANFIS模型为2输出1输入5层1阶Sugeuo模糊系统。利用某局网负荷数据训练和检测ANFIS网络模型后预测负荷,结果表明该算法鲁棒性好,抗干扰能力强,能有效补偿对象的大纯滞后。

Great attention is always paid to power system short-term load forecasting （STLF）, which is an important task of power utilities. STLF is concerned to the dispatching work and production scheme of power system, it provides original information for power flow calculation and stability analysis. Accurate load forecasting is helpful to the security and stability of power system as well as saving its generation costs. In view of uncertainness and non-linearity of forecast object, the application of Adaptive Neuro-Fuzzy Interference System （ANFIS） model to forecast short-term load is presented in this paper. This paper describes the structure and control mode of ANFIS, main functions of the whole system firstly. Adaptive Neural-fuzzy Inference System adjusts the parameters of the fuzzy inference system with hybrid algorithm and can produce fuzzy rules automatically. ANFIS not only has the ability of nonlinear mapping but also has the ability of self-learning, and instead of being concluded by a mathematics model, the information of control system is obtained by FIS, which is built up by ＂space delamination＂. So ANTIS can be used to achieve the nonlinear model of noise. The model not only achieves the optimal reconstruction but also possesses a desired robust against the effect of uncertainties and incomplete information in data processing. The ANFIS architecture predicts STLF is introduced secondly, five membership functions is selected during the simulation, i. e. it has 25 buses in second layer. This paper gives a certain network data to train and check the ANTIS neural network, then give the simulation example of modeling to forecast short-term load, and the results indicates that this strategy has good robust and anti-disturb ability, and it can retrieve the long time delay object.