Study On Forecasting Model Of Hourly Water Consumption For Optimal Operation

Along with the increasing area and demand amount of water distribution systems, the complexity of water supply system rise year after year. Traditional operation for water supply system of big city face challenge, optimal operation for water supply system is imperative under the situation. In spite of which method for optimal operation is adopted, they must include three steps: the prediction of hourly water consumption, the simulation of water supply network work state and the decision making of optimal operation. Among upper three steps, the first one is the base and precondition of the latter two. Prediction of hourly water consumption is of crucial importance to optimal operation of urban water supply systems. Although various methods have been produced over a long period of theoretical study, very few practical methods for prediction have yet been found.Usually total water use is separated into three components: periodicity, trend and randomness, but the calculated largest positive Lyapunov exponents show that a distinct chaotic component also exists. Here, based on conventional Wolf's algorithm for largest Lyapunov exponent, an improved calculation method was put forward in which the search for new vector length and its evolution angle weights.There are many different points of view on principal factors for daily urban water consumption, but almost all of them are based on experience. To solve this problem from theory rough set theory is firstly introduced. However in this study it was found that there exists a main practical problem associated with the conventional method: there are usually different results of attribute reduction for time series with different length or same length with different samples in the same information system based on traditional variable precision rough set algorithm. Therefore an algorithm based on weighting coefficient cumulative estimate of rough set theory was put forward. An analysis of principal factors for daily water consumption in Hangzhou was discussed using the improved algorithm show that 'maximum air temperature' , 'relative humidity', 'index of weather' and 'index of weekday' are the principal factors of daily urban water consumption.In allusion to the above two research results that a distinct chaotic component exists in the observed time series of daily water consumption and the nonlinear relations between daily water consumption with its influence factors such as air temperature, weather, air humidity and the day of the week etc, a support vector machines regression (SVR) forecasting model based on chaotic phase space reconstruction was developed. The model's train data was determined by chaotic phase reconstruction of time series of daily water consumption and the time series of its principal factors. The nonlinear relations between them were derived by the SVR method. Examples show that the introduction of the new forecasting method is helpful for improving the precision of forecasting urban daily water consumption.Hourly water consumption data is regard as time series data with a period of 24 hours reflecting our daily life. Therefore, the data can be divided by every 24 hours and regarded each of them as a pattern. If the hourly water consumption divided by the average hourly water same day, the pattern calls diurnal demand profile. In order to discuss the consecutive profiles with mathematic tools, they should be 'cluster'. Considering traditional fuzzy C-means (FCM) algorithm's shortcoming to cluster diurnal demand profile, an improved fuzzy C-means algorithm...