The Application Of Improved PSO-LSSVM In Hourly Urban Water Consumption Forecasting

Urban water consumption forecasting is of great significance to ease thecontradiction between supply and demand of water resources, guarantee urbanwater supply and optimize the scheduling, etc. In the project of Shanghaidowntown water consumption forecasting, this thesis builds an hourly waterconsumption forecasting model based on Least Squares Support VectorMachine(LSSVM) and optimizes the model parameters with improved ParticleSwarm Optimization(PSO).Firstly, as to solve the premature convergence of PSO, this thesis proposesan improved algorithm based on swarm activity, and further an extremumtrembling operator is added to the improved algorithm. Experiments of functionoptimization prove the validity and advantage of the two improved algorithms.Secondly, a method based on historical data and real-time data is proposedto fill the missing flow data and is of high accuracy. By calculating thepreprocessed flow data of water works, we get the hourly water consumption data and verify the accuracy with daily water consumption data. The resultsshow the accuracy is better than99%. This thesis selects the model inputs withthe heuristic backwards screening feature selection and correlation analysis,based on the relation between the water series and its influencing factors.Finally, the model based on improved PSO-LSSVM is built and applied toforecast the hourly water consumption of Shanghai downtown and its sub-areas.Simulation results show the accuracy of heuristic backwards screening featureselection is higher than correlation analysis. The average predictive error ofdowntown water consumption is less than1%and its sub-areas are less than2%,so the established model has good generalization ability and meets therequirements of engineering application.