The Research Of Data-knowledge-driven Diagnosis Method For Sludge Bulking

The activated sludge process（ASP）uses the adsorption and degradation of the microbial community to remove suspended solid particles and biochemical organics in wastewater and convert the discharged wastewater into usable water resources.ASP which owns the advantages of simple structure and convenient operation,has been the mainstream process in the urban wastewater treatment process（WWTP）.However,sludge bulking has always been a thorny problem that restricts the safe and stable operation of ASP.Sludge bulking leads to the degradation of sludge settling performance,the loss of sludge and the exceeding-standard of effluent water quality,and even destroys the entire operating system.Due to the strong nonlinear and time-varying characteristics,complicated mechanism and changeable causes of sludge bulking,it is still a challenge to accurately diagnose sludge bulking.Therefore,it is of paramount importance to study the accurate diagnosis method of sludge bulking to suppress sludge bulking and ensure the normal and stable operation of WWTP.In order to accurately diagnose sludge bulking,a data-knowledge-driven sludge bulking diagnosis method is proposed.First,the key water quality variables and fault categories were obtained through analyzing the mechanism and the influencing factors of sludge bulking.Second,a detection model based on recursive kernel principal component analysis（RKPCA）was designed to improve the detection accuracy by capturing the nonlinear and time-varying characteristics in the process of sludge bulking.Then,a diagnosis model based on Bayesian network（BN）was proposed to identify the root cause variables and severity level of sludge bulking by analyzing the causal relationship and probability relationship among the process variables.Finally,an intelligent diagnosis system was developed and applied to the urban wastewater treatment plant to realize the effective detection and identification of sludge bulking.The main research works of this paper are as follows:1.The mechanism analysis of sludge bulking:By studying the process mechanism of sludge bulking,the 11 key water quality variables:influent flow rate,influent chemical oxygen demand(COD_in),influent total nitrogen(TN_in),influent total phosphorus(TP_in),dissolved oxygen（DO）,temperature（T）,sludge load（F/M）,suspended solids concentration in aerobic zone（MLSS₁）,suspended solids concentration in secondary settling tank（MLSS₂）,recycled sludge volume（RSV）and effluent p H(p H_out),are obtained.According to the formation reasons and severity of sludge bulking,the categories are classified as:low DO sludge bulking,low T sludge bulking,low TN_in sludge bulking,low TP_in sludge bulking,low/high COD_in sludge bulking and micro sludge bulking and severe sludge bulking.2.The intelligent detection of sludge bulking based on RKPCA:Due to the existence of nonlinear and time-varying characteristics of sludge bulking process,it is difficult to accurately detect sludge bulking phenomenon.In this paper,a detection model based on RKPCA is designed.First,the input variables of the detection model are determined by analyzing the formation mechanism of sludge bulking.Second,the Gaussian kernel function is used to map the input variable data to the feature space and perform eigenvalue decomposition to capture the nonlinear characteristics of the sludge bulking process.Finally,a sliding window mechanism is designed to update the training sample data set,and the parameters of the model are adaptively updated to adapt to the time-varying characteristics of WWTP.Experimental results show that the proposed detection model can accurately detect the occurrence of sludge bulking.3.The intelligent diagnosis of sludge bulking based on BN:Due to the complicated mechanism and changeable causes of sludge bulking,it is difficult to accurately identify the root cause variables and severe level of sludge bulking.In this paper,a diagnosis model based on BN is proposed.First,the granger causality analysis（GC）method is used to preliminarily evaluate the causal relationship among process variables from the sludge bulking sample,and then the mechanism knowledge is used to confirm and supplement the evaluated causal relationship to obtain the structure of BN.Second,the maximum likelihood estimation（MLE）method is designed to evaluate the probability relationship among process variables from the sludge bulking samples to get the parameter of BN.Finally,the mean reconstruction contribution method based on RKPCA（MRC-RKPCA）is designed to isolate the fault candidate variables of sludge bulking samples,and then sludge bulking can be diagnosed through the reverse and forward diagnosis processes of BN.The experimental results show that the proposed diagnostic model can accurately identify the root cause variables and severe level of sludge bulking.4.The development of sludge bulking intelligent diagnosis system:In view of the problem that water plants often suffers from sludge bulking and there is no an universal and effective sludge bulking diagnosis system,a sludge bulking intelligent diagnosis system is developed.First,according to the requirement analysis,the functional modules required by the diagnostic system are determined.Second,the functional modules are packaged to complete the system integration.Finally,the developed intelligent diagnosis system was tested in a wastewater treatment plant.The experimental results show that the developed diagnosis system can realize effective detection and identification of sludge bulking.It can ensure the normal and stable operation of WWTP.