Research On Water Quality Prediction Model Of Mopanshan Reservoir Based On Optimized GRU

Ensuring water quality safety plays a crucial role in maintaining people’s health and promoting the green and high-quality development of the economy and society.China has always attached great importance to and paid close attention to this matter.Water quality prediction plays an important role in water quality security.By predicting the trend of water quality change,we can find the change of water quality in time,evaluate the impact of water quality on human body and environment,and formulate corresponding protection and treatment measures to improve the quality of drinking water and water supply safety.Reservoir water system is a complex,dynamic,fragile and important ecological environment system.Reservoir water quality prediction and early warning can help to detect abnormal water quality early,optimize water resources management and decision-making,and promote ecological environmental protection.By using data analysis methods to process water quality data,the main factors and characteristics that affect water quality can be identified,and appropriate intelligent calculation methods can be selected for modeling.A water quality prediction model can be established to predict and warn of future water quality changes,identify water quality problems in a timely manner,and take measures to solve them.However,the current water quality prediction models ignore the non-stationary characteristics of water quality data caused by the complexity of water.Secondly,water quality prediction models are prone to over fitting training data when predicting small sample water quality data.In order to solve the above problems,this paper discusses the algorithm based on Gated Recurrent Unit（GRU）and Particle Swarm optimization（PSO）,Variational Mode Decomposition（VMD）and Firefly Algorithm（FA）for water quality prediction model.The main research work and research results of this paper are as follows:（1）The variational modal decomposition（VMD）method is introduced to decompose the data,reduce the non-stationary characteristics of the data due to the complexity of the water environment,and mine the characteristic information hidden in the water quality data.At the same time,we use particle swarm optimization（PSO）to improve the VMD algorithm to avoid over-decomposition and under-determined decomposition caused by human experience.The experimental results show that the variational modal decomposition algorithm can effectively reduce the non-stationary nature of water quality data,and the performance of the variational modal decomposition algorithm improved by particle swarm optimization algorithm to reduce the non-stationary nature of water quality data is better.（2）Pearson correlation analysis is used to screen out the main influencing factors affecting Chl-a,DO,COD_Mn and NH₃-N,and analyze the correlation between each influencing factor and the main prediction indicators,so as to avoid the decline of the interpretability of the model and the information redundancy caused by too many neural network inputs,and provide a clear goal for water pollution control.（3）The BFGS quasi Newton method is used to improve the efficiency and convergence accuracy of the Firefly algorithm in the later optimization stage,which improves the convergence speed and accuracy of the FA algorithm and makes it suitable for a wider range of optimization problems;Using adaptive inertia weight technology to adjust the step factor allows the algorithm to better explore the search space,improve search quality,and help jump out of the local optimal solution.Experimental results show that the flexibility and adaptability of the improved Firefly Algorithm（IFA）are significantly improved.（4）A PV-IFA-GRU hybrid model for water quality prediction is proposed.GRU neural network is used to solve the problem of nonlinearity and small sample size of water quality data.The improved Firefly algorithm IFA is used to optimize the number of nodes in the hidden layer of GRU network and the learning rate of weights,which reduces the difficulty of adjusting hyperparameter,improves the performance of neural network,and helps to improve the prediction accuracy of water quality prediction model.The PV-IFA-GRU mixed water quality prediction model proposed in this article takes the data detected by the monitoring station of Mopanshan Reservoir as the research object,and verifies the performance of the PV-IFA-GRU model.The experimental results show that when the hybrid model predicts Chl-a,DO,COD_Mn,and NH₃-N,the prediction accuracy R² can reach 98.8%,98.7%,98.4%,and 98.3%,which can adapt to different prediction scenarios and provide technical support for predicting and preventing future reservoir water quality deterioration.