Research On Intelligent Prediction Method Of Water Consumption Sequence Based On Deep Learning

The rapid development of the society and urbanization process demand more and more water consumption.Under the existing water resources condition,it is important to improve the reasonableness and accuracy of water consumption prediction for the scientific and reasonable allocation of water resources.The thesis addresses the strong non-smoothness and multi-temporal scale variation of water consumption sequences,as well as the low accuracy and poor credibility of water consumption prediction methods,and constructed intelligent prediction methods of water consumption sequences based on deep learning,which provides important theoretical support and technical support for the rational allocation of regional water resources.Based on the analysis of existing water prediction methods,the thesis combined data decomposition algorithms,intelligent optimization algorithms and deep learning time series algorithms to build a water prediction method based on "decomposition-prediction-synthesis" and carried out validation and analysis:(1)Construction of EEMD-WOA-SRU water sequence intelligent prediction model.To address the limited nonlinear expressiveness of a single model,the improved EEMD-WOA-SRU combined water consumption prediction model was constructed by the improved Ensemble Empirical Mode Decomposition(EEMD),improved Whale Optimization Algorithm(WOA)and Simple Recurrent Unit(SRU)in this thesis.Firstly,to address the endpoint effect problem in EEMD,the thesis used the mirror polar extension method and the Long Short-Term Memory(LSTM)-based data prediction method to suppress the endpoint effect of EEMD,and secondly,the improved whale optimization algorithm was used to optimize the number of neurons in the simple recurrent unit to increase the robustness of the model.The experimental results show that the similarity coefficient of the components obtained using the LSTM data prediction method is improved by 3.5% and the decomposition error is reduced by 0.94% compared with the mirror polar extension method.The prediction results of the constructed improved EEMD-WOA-SRU showed an average reduction of 45.42% in MAE,50.43% in RMSE,and 52.38% in NSE compared to SRU.Compared with EEMD-ELM and EEMD-RNN,the improved EEMD-WOA-SRU model has an average reduction of 17.64% and 9.19% in MAE,an average reduction of 13.55% and9.15% in RMSE,and an average improvement of 4.55% and 0.15% in NSE,possessing higher prediction accuracy and confidence.(2)Construction of SSA-VMD-TCN water sequence intelligent prediction model.The thesis combined the improved Sparrow Search Algorithm(SSA),the improved Variational Mode Decomposition(VMD)and the Temporal Convolutional Network(TCN)to construct the improved SSA-VMD-TCN combined water consumption prediction model.Firstly,this thesis used the extreme value extension method to suppress the endpoint effect of VMD,and secondly used the improved sparrow search algorithm to optimize the hyperparameters of VMD and TCN for parameter search.The experimental results show that in suppressing the endpoint effect of VMD,the components obtained using LSTM data prediction method improve the similarity coefficient by 29.23% and reduce the decomposition error by 70.76%compared with the mirror polar extension method.The improved SSA-VMD-TCN model was constructed with an average reduction of 83.01% in MAE,85.36% in RMSE,and75.04% in NSE compared with TCN.Compared with VMD combined with LSTM network,MAE is reduced by 63.13% on average,RMSE is reduced by 39.72% on average,and NSE is improved by 5.79% on average,and compared with the improved EEMD-WOA-SRU,MAE is reduced by 67.94% on average,RMSE is reduced by 69.61%,and NSE improved by9.72% on average.The improved combined SSA-VMD-TCN model possesses higher prediction accuracy and confidence.This thesis analyzed and modeled the water consumption prediction methods from multiple spatial and temporal scales.The prediction accuracy and credibility of the improved combined EEMD-WOA-SRU and SSA-VMD-TCN models were verified through comparative analysis,and the theory and construction method of the models can provide a reference for regional water resources allocation.