| Water supply system as the city’s infrastructure,water quality directly affects people’s water safety.In the source Waterworks treatment industry,coagulation precipitation is an important process of water treatment system,which determines the quality of water from the Waterworks and the cost of water production,of which the control of coagulant addition is the key.Due to the fluctuation of raw water quality into the source Waterworks,there is a large time lag in the process of coagulation and dosing in the water purification plant,it is difficult to calculate the amount of the drug in a timely and accurate manner.Therefore,modeling the Waterworks coagulation delivery system to predict the amount of drug delivery is very intentional to ensure the quality and safety of water supply and reduce energy consumption.In this study,the Waterworks coagulant addition was predicted by the radial base(RBF)neural network prediction model.In addition,in order to ensure the stability,accuracy and improve the performance of the model,this study optimizes the RBF model by PSO and subtract clustering,and compares the prediction results with the actual operation data of the Waterworks,and finds that the optimization model MRE(average relative error)is 5.63% and MAPE(average absolute error)is 23.80%.The establishment and research results of the Waterworks dosing model system are as follows:(1)Through the global sensitivity analysis method based on variance decomposition,the sensitivity analysis of the main factors of 7 coagulation drugs,such as raw water turbidity,chroma,total number of bacteragions(CFU),was carried out,and the influence factors affecting the coagulant addition of the source Waterworks were calculated as: raw water turbidity(NTU)> The input matrix of raw water flow >(Q)> oxygen consumption(CODMn)> raw water p H,the prediction model is determined to be 4×191,X(Qx1n,NTUx2 n,PHx3n,CODMnx4n),1× 191 drop output matrix,the network structure is clarified.(2)A single RBF network model,RW(random walk)-RBF optimized network model and PSO(particle group)-RBF optimized network model were constructed,comparing the advantages and disadvantages of RW algorithm and PSO algorithm to RBF network,and it was found that PSO-RBF was 5.82% lower than RW-RBF algorithm MAPE and MRE was 0.98% lower.On this basis,a composite control scheme for optimizing RBF neural networks based on PSO and other methods is determined.By subtracting the clustering method to determine the initial center point position of the base function and the number of hidden layer nodes,using PSO algorithm to dynamically update the center point position of the RBF network base function,using pseudo-inverse method to determine the weight of the function,and finally optimized model prediction system can accurately give the coagulant add-on.(3)The study found that the PSO-RBF model was 3.05 percent lower than the single RBF model MRE and the REmax(maximum relative error)was 0.1986 lower.The cooperation and competition between particles make PSO optimize RBF model increase the high-dimensional search capability of multi-dimensional complex space,and can quickly find the optimal solution of neural network weights.The robustness of the model is reduced,the convergence speed is faster and the precision is higher,and the accurate control of coagulation drug is realized.(4)The PSO-RBF model can directly get the mapping relationship between the drug and the water quality of raw water infested,and the time sorting problem of the data no longer constrains the results of the network program.In addition,in order to verify the universality of the model,the study selected Waterworkss in other regions to test the model,the results are better,the modeling process is better than the binary regression model and the rational model,which can provide an effective reference for simulating the amount of Waterworks dosing. |
PDF Full Text Request