基于模糊聚类的BOA-SVR分时段精细化短期负荷预测

Time-division refined short-term load forecasting based on BOA-SVR and fuzzy clustering

近年来,随着智能电网技术不断发展,电力系统的运行规模和复杂程度日益增大,对负荷的精度要求也越来越高,因此提出一种采用模糊C均值聚类结合动态自适应权重和柯西变异的蝙蝠优化算法(bat optimization algorithm,BOA)优化支持向量回归(support vector regression,SVR)的分类型分时段精细化负荷预测方法来提高负荷预测的精度。该方法首先采用模糊C均值聚类对历史数据样本进行聚类并评价其聚类效果、确定聚类数目,随后将预测日样本分类型建立分时段且考虑不同时段实时变化的气象影响因素的SVR预测模型;然后以预测精度为判断依据,利用动态自适应权重的方法和柯西分布逆累积分布函数改进蝙蝠算法对SVR模型进行寻优,寻找最优参数以建立SVR预测模型,并对预测日负荷进行预测;最后采用不同的预测方法对河南省某地区连续3天的负荷进行预测。预测结果以及误差分析表明:与无参数优化的SVR预测模型以及粒子群优化(particle swarm optimization, PSO)算法优化的SVR预测模型相比较,所提出的方法预测精度更高,对于不同类型日的预测精度均可达到96%以上,具有可行性。

In recent years, in the process of continuous development of smart grid technology, the operation scale and complexity of power system are increasing, and the requirements for accuracy of load are becoming higher and higher. A classified time-division refined load forecasting method of fuzzy C-means clustering combined with dynamic adaptive weighting and Cauchy variant bat optimization algorithm(BOA) for the optimization of support vector regression(SVR) is proposed to improve the accuracy of load forecasting. Firstly,fuzzy C-means clustering is used to cluster historical data samples and evaluate their clustering effect to determine the number of clusters. Then, SVR prediction model is established by classifying the samples of the predicted days into different periods and considering the real-time changes of meteorological factors in different periods of the predicted day. Based on the prediction accuracy, the dynamic adaptive weighting method and the Cauchy distribution inverse cumulative distribution function are used to improve the bat algorithm to optimize the SVR model. The optimal parameters are found to establish the parameter-optimized SVR prediction model,and the daily load is predicted. Finally, the different forecasting methods are used to predict the load in a threeday period in a certain area of Henan Province. The SVR model is compared with the parameter-optimized SVR prediction model and the particle swarm optimization(PSO) algorithm optimized SVR prediction model.According to the prediction results and error analysis results, the method proposed in this paper has higher prediction accuracy, and the prediction accuracy for different types of days can reach more than 96%.Therefore, the proposed method is feasible.