Study Of A Daily Water Consumption Forecasting Method Based On Improved Least Squares Support Vector Machine

With the sustainable development of cities in China, the sizes of urban water distribution network are fast increasing, and urban water distribution networks are becoming more and more complex. The traditional experimental dispatch can not meet the needs of the scientific dispatch and operation of water distribution network in China. Optimal dispatches of urban water distribution network are imperative under this situation. Urban daily water consumption forecasting is the first step and key step to optimal dispatch of water distribution network, and its accuracy determines whether the optimal operation scheme can be applied to the real water distribution systems or not.Research progress of urban daily water consumption forecasting is systematically summarized at home and abroad in this paper, and the main work of this paper is below:(1)The wavelet analysis theory is introduced to effectively eliminate the noise of the series of daily water consumption. The correlation of the denoised daily water consumption series was analyzed by the mutual information method.(2)The major influence factors of predicted daily water consumption and denoised revelvant daily water consumption are used as the inputs, and denoised predicted daily water consumption are used as the outputs. A mutative scale chaos algorithm with strong global search capability and faster search speed was introduced to optimize the parameters of least squares support vector machine(LSSVM).A forecasting model based on wavelet analysis, mutative scale chaos genetic algorithm(MSCGA) and LSSVM is proposed. Among MSCGA LSSVM-based model, genetic LSSVM-based model and wavelet genetic LSSVM-based model, the proposed method has best forecasting performance in case study.