The Prediction Of Short-term Water Demand Based On A Spiking Self-organizing Fuzzy Neural Network

With the incessant expansion of the city scale and the continuous improvement of the residents' living standards,the water demand of the city continues to increase,therefore the stable and safe operation of the water supply networks has been a necessary prerequisite for ensuring social production and citizens' lives.At the same time,the operation and management of water supply networks has been put forward higher requirements of reducing energy consumption and controlling leakage under the current situation of serious environmental pollution and shortage of energy and water resources.Water demand forecasting research is an important link and basic premise for ensuring the optimal operation of water supply pipe networks,and the forecasting accuracy will directly affect the effectiveness and reliability of water supply optimization scheduling.On account of the complicated influencing factors,high nonlinearity,strong ambiguity,and large randomness of urban water demand,this paper proposes a prediction model based on spiking self-organizing fuzzy neural network(SSOFNN)to achieve short-term water demand forecasting.The main research work of this paper includes the following points:1.As for the parameters training of neural network optimization design,this paper adopts improved Levenberg-Marquardt(ILM)algorithm as the learning algorithm of fuzzy neural network in order to obtain more accurate and effective water demand forecast results.The LM algorithm,as a second-order algotithm,solves the problems that the first-order algorithm is easily trapped in the local minimum and the convergence rate of the random algorithm is slow,but there is still a problem of heavy computation burden and large storage space in the application of big data samples.The ILM algorithm optimizes the computational flow and replaces the Jacobian matrix multiplication with the calculation of the gradient vector and the quasi-Hessian matrix.The comparison experiments have shown that the ILM algorithm can improve the learning efficiency of the artificial neural network and save the calculation storage space.2.For the fixed-structure fuzzy neural network,the number of hidden-layer neurons is difficult to determine at first,and the network structure cannot be adjusted online according to the data scale.Aiming at this problem,this paper designs a spiking self-organizing fuzzy neural network.The spiking mechanism,which is proposed based on the information transmission model of cerebral cortex and the Integrate-and-Fire model of spiking neurons,is adopted to accomplish the growth and pruning of the fuzzy neural network structure,and then the dynamic adjustment of the network structure in the training process could be achieved,and the performance of the fuzzy neural network in water demand forecast applications is improved simultaneously.3.A short-term water demand forecasting model based on spiking self-organizing fuzzy neural network is established.Principal component analysis(PCA)is used to reduce dimensionality of multiple short-term water demand influencing factors,then the linear-independent principal component variables could be obtained as input data for the prediction model.The influencing factors of urban water demand are complex,varied and intercoupling with each other.Selecting the auxiliary variables appropriately is an important prerequisite for obtaining accurate and reasonable prediction results.In this paper,11 preliminary determined impact factors including meteorological conditions and daily types are processed through PCA to obtain principal component variables,which contain most information of the original data,and further to perform short-term water demand forecasting.On this basis,the fuzzy neural network parameters is trained by the ILM algorithm,and the network structure is adjusted by the spiking mechanism to establish a shortterm demand forecasting model based on SSOFNN.The daily water consumption data of a university in Beijing is used as an example to design a simulation experiment.The results have shown that this prediction model has higher prediction accuracy and faster convergence speed.4.An intelligent supervision platform for water supply networks is designed and developed,including a real-time water consumption monitoring module,a water consumption statistical analysis module,and a water demand forecasting module.The platform includes two subsystems: the front end and the back end.The front end uses the Vue.js and Element UI component library to implement the user interface design and data display,and the back end uses the prevailing frameworks like Spring,Mybatis to implement data transmission and logic processing,and the SSOFNN water demand forecasting model proposed in this paper is embedded to realize the intelligent supervision of the water distribution networks.