变结构遗传最小二乘支持向量机法预测日用水量

Daily water demand forecasting method based on variable structure genetic least squares support vector machine

为解决日用水量预测模型的动态参数估计问题,提出了基于变结构遗传最小二乘支持向量机的预测模型.以日用水量的主要影响因素和相关日用水量为输入,利用遗传算法对基于LSSVM的历史日用水量模型参数进行寻优,获得模型结构参数序列;采用扩展卡尔曼滤波器估计基于最小二乘支持向量机的预测日用水量模型参数,进而预测下一日用水量.实例分析表明:提出的模型具有较高的预测精度,预测的最大绝对相对误差仅为9.3%,平均绝对相对误差为2.09%.

To dynamically estimate the parameters of daily water consumption model, a new variable-structure genetic least squares vector machine（LSSVM）-based model is proposed. The principal factors of daily water consumption and the correlative daily water consumption are used as the model inputs. With genetic algorithm, the parametersseries of LSSVM-based historical daily water consumption models aredetermined. With the series, Extended Kalman Filter（EKF） is applied to estimate the parameters of LSSVM-based next-day prediction modeland the next-day daily water consumption is forecasted. Case study shows that the proposed model has higher forecasting performance in term of a maximum absolute relative error of 9. 3% and a mean absolute relative error of 2.09% .