Research On The Combined Forecast Model Of Urban Short-term Water Supply Based On PSO-ANN-LSSVM

The short-term water supply forecast for the city is the basis for the optimal dispatch of the urban water supply system,which can make decisions for the optimal dispatch of the water company,improve water resource utilization,and save energy consumption.In order to further improve the accuracy of urban short-term water supply forecasting,this paper collects the measured data(time step is 15 minutes)of a city’s water plant as a research object,and develops a combined short-term urban water supply forecasting model based on PSO-ANN-LSSVM The main research results and conclusions are as follows:(1)Based on the data pre-processing technology and chaos theory,the original water supply time series is denoised and the chaos characteristics are judged.First of all,it is judged that the existence of noise in the original time series will seriously interfere with the prediction model and affect the accuracy of the prediction results.Therefore,the standard deviation noise reduction method is used to reduce noise by one,two,and three times,and a total of four sets of data are obtained.Then use mutual information method and Cao method to calculate the optimal delay time and the optimal embedding dimension respectively to reconstruct the phase space of the four sets of water supply time series after noise reduction,and then model each prediction model Of the establishment.Finally,use the maximum Lyapunov exponent for quantitative analysis to perform chaotic characteristics It is determined that the maximum Lyapunov exponents of the water supply time series of each group are greater than zero after calculation,indicating that the four water supply time series have chaotic characteristics and predictability.(2)Based on neural network,least squares support vector machine(LSSVM)and particle swarm intelligence optimization algorithm(PSO),predictive analysis of water supply examples is performed.First of all,for short-term urban water supply with high degree of non-linearity and difficult prediction,this paper uses BP,RBF neural network and LSSVM model with strong non-linear processing capacity to predict water supply based on phase space reconstruction.The prediction results show that:(1)the above three prediction models can predict the overall trend of water supply,but there is much room for improvement in local prediction effects;(2)proper data noise reduction processing can effectively improve the prediction accuracy of the model.Considering that the prediction accuracy of BP and RBF neural networks is limited by parameter settings,this paper uses the particle swarm intelligent optimization algorithm(PSO)to optimize the parameters of BP and RBF neural networks to obtain the PSO-BP and PSO-RBF optimized prediction models.Four groups of water supply examples are used for model verification and analysis.The final prediction results show that:(1)the two optimization models can also track the overall change trend of the four sets of water supply time series,and after PSO optimization parameters and data noise reduction,the prediction accuracy is greatly improved and the prediction error is reduced;(2)the water supply of each group In the time series,the MAPE,MSE,and R~2 three evaluation indicators of the PSO-BP neural network are increased by about 5%,0.03,and 0.06 compared with the BP neural network.The MAPE,MSE,and R~2 three evaluation indicators of the PSO-RBF neural network are higher than the RBF neural network.It is improved by about 3%,0.02,and 0.03,respectively;(3)The optimized model has significantly improved the local prediction effect,and appropriate data noise reduction processing can effectively improve the prediction accuracy of the model.(3)Based on PSO-BP,PSO-RBF optimized forecasting model and LSSVM model,the combined forecasting method was improved,and a new PSO-ANN-LSSVM urban short-term water supply series-parallel combined forecasting model was proposed.Aiming at the problem of low accuracy of the single prediction model,this paper first uses the PSO-BP and PSO-RBF optimization models and the LSSVM model to make separate predictions in series,and obtains the PSO-BP-LSSVM tandem prediction result and PSO-RBF-LSSVM tandem Prediction results,and then this paper attempts to use the optimal combination prediction method to use weighted combination of the two prediction results to obtain the prediction results of the series-parallel weighted combination prediction model.The final results show that:(1)The MAPE,MSE,and R~2 three evaluation indicators of the series-parallel weighted combination prediction model are improved by about 0.9%,0.01,and 0.04 compared with the PSO-BP-LSSVM tandem combination prediction model,respectively,compared with the PSO-RBF-LSSVM tandem combination prediction model Increased by approximately 0.5%,0.02,and 0.01,respectively;(2)Compared with each single model(BP,RBF,LSSVM,PSO-BP,PSO-RBF),the three evaluation indicators have been significantly improved,and the MAPE improvement range is 4%to 10%;MSE increase range is 0.03 to 0.06;R~2 increase range is 0.01 to 0.15;(3)One of the standard deviations has the best noise reduction effect.