A data-driven intelligent evaluation method for biodegradability of petrochemical wastewater

Abstract

This invention relates to a data-driven intelligent evaluation method for the biodegradability of petrochemical wastewater, belonging to the field of refining wastewater treatment and resource utilization. Addressing the difficulty in obtaining real-time values ​​of key variables such as BOD in current petrochemical wastewater treatment processes, which hinders accurate online evaluation of biodegradability, this invention designs a data-driven intelligent evaluation model for petrochemical wastewater biodegradability. This model achieves real-time and accurate evaluation of the biodegradability of petrochemical wastewater. Experimental results show that this intelligent evaluation model can accurately evaluate the biodegradability of petrochemical wastewater with high accuracy, ensuring the safe, efficient, and stable operation of the petrochemical wastewater treatment process.

Description

Technical Field

[0001] The present invention designs a data-driven intelligent evaluation method for the biodegradability of petrochemical wastewater, which realizes the real-time and accurate evaluation of the biodegradability of petrochemical wastewater. As one of the main indicators for evaluating wastewater composition, the biodegradability of petrochemical wastewater is an important basis for selecting different treatment processes, preventive measures and operating strategies in the process of petrochemical wastewater treatment. Accurate evaluation of the biodegradability of petrochemical wastewater is crucial to provide corresponding prevention and response measures support to on-site operators of water plants, and has an important impact on cost reduction, efficiency improvement and efficient and stable operation in the process of petrochemical wastewater treatment. It is an important branch of advanced manufacturing technology and belongs to the field of refining wastewater treatment and resource utilization. Background Art

[0002] The biodegradability of petrochemical wastewater is also called the biodegradability of petrochemical wastewater, that is, the difficulty of organic pollutants in petrochemical wastewater to be degraded by microorganisms. It is one of the indicators for evaluating the composition of petrochemical wastewater and an important basis for petrochemical enterprises to choose different processes for treating wastewater and selecting operating strategies. Therefore, achieving real-time and accurate evaluation of the biodegradability of petrochemical wastewater is a key issue currently faced by petrochemical wastewater treatment plants. The most common method for evaluating the biodegradability of petrochemical wastewater is the water quality index evaluation method. However, due to the different detection frequencies and categories of key water quality indicators, the water quality index evaluation method is difficult to obtain real-time and accurate evaluation results and cannot meet the actual needs of petrochemical wastewater treatment plants. Therefore, the use of intelligent means for accurate evaluation is the current research focus.

[0003] The present invention designs a data-driven intelligent evaluation method for the biodegradability of sewage, which mainly analyzes the characteristics of the sewage treatment process, uses online data and offline data to determine the characteristic variables of the biodegradability of sewage, and performs corresponding pretreatment, establishes an intelligent evaluation model for the biodegradability of sewage based on fuzzy neural networks, and uses online data to adjust the model parameters, thereby realizing the intelligent evaluation of the biodegradability of sewage. Summary of the invention

[0004] The present invention obtains a data-driven intelligent evaluation method for biodegradability of petrochemical wastewater. The evaluation model uses four categories of wastewater biodegradability as predictive variables, including biochemical oxygen demand (BOD), primary sedimentation tank chemical oxygen demand (COD), influent chemical oxygen demand (COD), and water biodegradability (BOD). in , total phosphorus TP, ammonia nitrogen NH4-N, pH, dissolved oxygen DO, temperature T, suspended solids concentration SS, redox potential ORP, inlet flow Q in, turbidity, nitrate nitrogen NO3-N, and sludge concentration MLSS are taken as related variables. The data of these variables are used to extract characteristic variables, and a fuzzy neural network is constructed to achieve nonlinear mapping between the biodegradability category of petrochemical wastewater and the characteristic variables. This solves the problem that the biodegradability of petrochemical wastewater is difficult to evaluate accurately in real time, effectively improves the accuracy of wastewater biodegradability evaluation, and shortens the evaluation cycle of wastewater biodegradability.

[0005] The present invention adopts the following technical solutions and implementation steps:

[0006] 1. A data-driven intelligent evaluation method for biodegradability of petrochemical wastewater, comprising the following steps:

[0007] (1) Determine the prediction variables and related variables: The present invention mainly conducts intelligent evaluation on the biodegradability of wastewater in the process of petrochemical wastewater treatment, and determines the biochemical oxygen demand (BOD), the chemical oxygen demand (COD) of the primary sedimentation tank, and the chemical oxygen demand (COD) of the influent. in , total phosphorus TP, ammonia nitrogen NH4-N, pH, dissolved oxygen DO, temperature T, suspended solids concentration SS, redox potential ORP, inlet flow Q in , turbidity, nitrate nitrogen NO3-N, sludge concentration MLSS, a total of 14 variables are wastewater biodegradability related variables, and the four categories of wastewater biodegradability, namely, difficult to degrade, partially degradable, relatively degradable, and easily degradable, are used as predictive variables. The four categories are defined as Class 1 to Class 4 respectively;

[0008] (2) Design a multi-scale feature extraction algorithm for selecting biodegradability characteristic variables of petrochemical wastewater:

[0009] ① According to the sampling period of related variables, the sampling period of dissolved oxygen DO, nitrate nitrogen NO3-N and redox potential ORP is 5 minutes, which are expressed as τ1, τ2, τ3 respectively. The obtained sampling sample vectors are x1(t1), x2(t2), x3(t3), where t1, t2, t3 are the sampling time, sludge concentration MLSS, turbidity, temperature T, pH and influent flow Q respectively. in The sampling period is 15 minutes, which are expressed as τ4, τ5, τ6, τ7, τ8 respectively. The obtained sampling sample vectors are x4(t4), x5(t5), x6(t6), x7(t7), x8(t8), where t4, t5, t6, t7, t8 are the sampling time, ammonia nitrogen NH4-N, primary sedimentation tank chemical oxygen demand COD, influent chemical oxygen demand COD in The sampling period is 2 hours, expressed as τ9, τ 10 ,τ 11 The obtained sampling vectors are x9(t9),x 10 (t 10 ),x 11(t 11 ), where t9, t 10 , t 11 are the sampling times, biochemical oxygen demand BOD, suspended solid concentration SS, and total phosphorus TP respectively. The sampling period is 1 day, denoted as τ 12 , τ 13 , τ 14 respectively. The obtained sampling sample vectors are x 12 (t 12 ), x 13 (t 13 ), x 14 (t 14 ), where t 12 , t 13 , t 14 are the sampling times respectively. The sampling scale vector of the defined variables is τ = [τ1, …, τ i , …, τ 14 , where τ i is a positive integer, i = 1, 2, …, 14, and x i (t i ) = [x i (1), x i (2), …, x i (t i ), …, x i (n i )] are the elements in the vector; The sampling times corresponding to each variable are defined as t i = 1 + kτ i , k = 0, 1, …, n i - 1, n i is the number of samples of the i-th variable and is a positive integer;

[0010] ② Define the step size of time scale fusion as m, m0 ≤ m ≤ m1, where m, m0, and m1 are positive integers and m0 < m1 ≤ 14. Define the number of samples of the fused variable as N, i.e., n i = N; When τ i = m, the sampling vector of each variable after scale fusion is x i ′(t i ′) = x i (t i ), t i ′ is the sampling time of the fused variable; When τ i < m, the sampling times of each variable become t i ′ = 1 + km, and the sampling vector of each variable after scale fusion is x i ′(t i ′) = x i (1 + km); When τ i>m, the sampling vector after the fusion of each variable scale is

[0011] x i ′(t i ′)=[x i (1),x i (1)+x i (n i )-x i (1) / N,x i (1)+2[x i (n i )-x i (1)] / N,…,x i (1)+(N-1)[x i (n i )-x i (1)] / N];

[0012] ③Set the number of feature extraction times L1=1;

[0013] ④j=L1, according to the obtained sampling sample vector, press n i Merge from small to large to form the sample matrix X(t) h×o =[x1(t),x2(t),…,x i (t),…,x o (t)], where x i (t) = x i ′(t i ′); h and o represent the sample matrix X(t) h×o The total number of samples and the number of variables, and h≤N, o≤14; the sample matrix X(t) h×o Perform standardization transformation to obtain the standardized matrix X std (t) h×o =[z1(t),z2(t),…,z2(t),…,z o (t)], where:

[0014]

[0015] Where b = 1, 2, ..., o, x b (t) = [x b (1),x b (2),…,x b (t)] is the matrix X(t) h×o The sample vector of the bth variable in z b (t) = [z b (1),z b (2),…,z b (t)] is the matrix X std (t)h×o The standardized sample vector of the bth variable in , μ b and θ b The matrix x b The mean and standard deviation of (t) are calculated as follows:

[0016]

[0017]

[0018] ⑤According to the standardized matrix X std (t) h×o , calculate the corresponding covariance matrix R(t) o×o =[r1(t),r2(t),…,r o (t)], where the calculation formula is as follows:

[0019]

[0020] Among them, r b (t) = [r b (1),r b (2),…,r b (t)] is the matrix R(t) o×o The sample vector of the bth variable in , represents the correlation coefficient vector between the bth variable and the cth variable, z c (t) = [z c (1),z c (2),…,z c (t)] is the matrix X std (t) h×o The sample vector of the cth variable in , c = 1, 2, ..., o, T is the transpose of the formula;

[0021] ⑥ Solve the covariance matrix R(t) o×o The eigenvalues ​​and corresponding eigenvectors are λ1,λ2,…,λ o and p1,p2,…,p o ;

[0022] ⑦ Arrange the eigenvalues ​​in descending order and calculate the cumulative contribution rate η(e) of the first e related variables. The calculation formula is as follows:

[0023]

[0024] Among them, e∈[1,o], λ c Represents the matrix R(t) o×o The cth eigenvalue is selected, and the first e related variables with cumulative contribution rate η(e) exceeding 85% are selected as the characteristic variables of sewage biodegradability. The extracted characteristic variable set X e(t) is expressed as follows:

[0025] X e (t)=[x1(t),x2(t),…,x j (t),...,x e (t)]; (6)

[0026] Among them, x j (t) = [x j (1),x j (2),…,x j (t)], j = 1, 2, ..., e, is the set X e The jth sample vector in (t);

[0027] ⑧The number of feature extractions L1 increases by 1. If the number of extractions L1 ≤ N, go to step ④ to continue training, otherwise stop feature extraction;

[0028] After the features are extracted by the multi-scale feature extraction algorithm, the first e variables that meet the contribution rate η(e)>85% are selected in descending order from the eigenvalue as the feature variables;

[0029] (3) Design of intelligent evaluation model for biodegradability of sewage: A fuzzy neural network is used to establish a sewage biodegradability evaluation model. The topological structure of the model is divided into four layers: input layer, RBF layer, normalization layer, and output layer. The connection mode of the neural network epp-4 is determined, that is, the number of neurons in the input layer is e, the number of neurons in the RBF layer is p, the number of neurons in the normalization layer is p, and the number of neurons in the output layer is 4. The weights of the neural network are randomly assigned. The input of the fuzzy neural network is expressed as X(t) = [x1(t), x2(t), …, x e (t)], the expected output of the neural network is denoted as y d (t) = [y d 1 (t),y d 2 (t),y d 3 (t),y d 4 (t)], the actual output is expressed as y(t) = [y1(t), y2(t), y3(t), y4(t)], y d q (t) and y q (t) represent the expected output and actual output of the network corresponding to the qth category, q = 1, 2, 3, 4, the expected outputs of the 1st to 4th categories are defined as [1, 0, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0] and [0, 0, 0, 1] respectively. The input of the fuzzy neural network at the tth time is represented by x1(t), x2(t), …, xa (t), the computational function of the fuzzy neural network is:

[0030] ① Input layer: This layer consists of e neurons, and the output of each neuron is:

[0031] u I (t) = x I (t),I=1,2,…,e; (7)

[0032] Among them, u I (t) represents the output of the Ith neuron in the input layer at time t, x I (t) represents the input of the Ith neuron in the input layer at time t;

[0033] ②RBF layer: The RBF layer consists of p neurons, and the output of each neuron is:

[0034]

[0035] in, is the output of the Jth RBF neuron, J = 1, 2, ..., p, c IJ (t) and σ IJ (t) are the center value and Gaussian width of the I-th membership function in the J-th neuron of the RBF layer, c IJ (t)∈[-2,2],σ IJ (t)∈[0.01,1];

[0036] ③ Normalization layer: The normalization layer consists of p neurons, and its output calculation formula is:

[0037]

[0038] Among them, v z (t) is the output of the zth neuron in the normalization layer;

[0039] ④Output layer: The output of the output layer is:

[0040]

[0041] Among them, y q (t) is the output of the qth neuron in the output layer, indicating the output probability of the network on the qth category, q = 1, 2, 3, 4, w q (t) represents the connection weight between the normalization layer neuron and the qth neuron in the output layer, w z q (t) represents the connection weight between the zth neuron in the normalization layer and the qth neuron in the output layer, v(t) = [v1(t), v2(t), …, v p(t) is the output vector of the neurons in the normalization layer;

[0042] ⑤ Define the error function as:

[0043]

[0044] where y d q (t) is the expected output of the q-th neuron in the output layer;

[0045] (5) Train the neural network, specifically:

[0046] ① Given that the number of neurons in the RBF layer and the normalization layer of the fuzzy neural network is p, where p is a natural number and p ∈ [10, 20], the model inputs are X(1), X(2), …, X(t), …, X(P), and P is the number of training samples. Conduct training and design the calculation step L2 = 1;

[0047] ② Let t = L2, calculate the output y(t) of the model, and use the gradient descent method to adjust the parameters of the dynamic evaluation model;

[0048]

[0049]

[0050]

[0051] where c IJ (t), σ IJ (t), and w z q (t) are the center value, width, and weight before model adjustment respectively, and c IJ (t + 1), σ IJ (t + 1), and w z q (t + 1) are the center value, width, and weight after model adjustment respectively, and η ∈ (0, 0.1] represents the learning rate;

[0052] ③ Increase the learning step L2 by 1. If the step L2 < P, go back to step ② for continued training. If L2 = P, stop the calculation;

[0053] (6) Evaluate the biodegradability of sewage using the y(t) solved by the trained neural network, that is, select the category corresponding to the node with the maximum output probability y δ (t) in the output layer of the fuzzy neural network as the final classification result; where y δ(t)=max{y1(t),y2(t),y3(t),y4(t)},δ=1,2,3,4,max represents the symbol of taking the maximum value in the set; if y1(t) is the maximum output probability, that is, δ=1, the evaluation result is difficult to degrade; if y2(t) is the maximum output probability, that is, δ=2, the evaluation result is partially degradable; if y3(t) is the maximum output probability, that is, δ=3, the evaluation result is relatively easy to degrade; if y4(t) is the maximum output probability, that is, δ=4, the evaluation result is easy to degrade.

[0054] The creativity of the present invention is mainly reflected in:

[0055] (1) The present invention aims to solve the problem that petrochemical wastewater treatment plants cannot accurately evaluate the biodegradability of wastewater in real time. The present invention proposes a data-driven intelligent evaluation method for the biodegradability of petrochemical wastewater. According to the wastewater treatment process data, seven characteristic variables related to the biodegradability of petrochemical wastewater are extracted: redox potential (ORP), influent pH, influent ammonia nitrogen (NH4-N), influent chemical oxygen demand (COD), influent total phosphorus (TP), influent suspended solids (SS), and influent biochemical oxygen demand (BOD). An evaluation model for the biodegradability of wastewater is established, which solves the problem that it is difficult to accurately evaluate the biodegradability of petrochemical wastewater in real time.

[0056] (2) The present invention is based on the fact that the current sewage treatment process is a complex, dynamic and time-varying process. The relationship between the biodegradability category of sewage and the characteristic variables not only has the characteristics of nonlinearity and strong coupling, but is also difficult to describe using an accurate mathematical model. Therefore, based on the sewage treatment process data, a fuzzy neural network is used to realize the prediction of the biodegradability category of sewage, which has the characteristics of high prediction accuracy and good adaptability to environmental differences.

[0057] It should be noted that the present invention adopts 7 characteristic variables related to the biodegradability of petrochemical wastewater and designs an intelligent evaluation method for the biodegradability of petrochemical wastewater based on fuzzy neural network. As long as the characteristic variables and methods of the present invention are used to predict the biodegradability category of wastewater, it should fall within the scope of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is the fuzzy neural network structure diagram of the present invention;

[0059] Figure 2 This is the confusion matrix result diagram of the intelligent evaluation of the biodegradability of sewage in the present invention DETAILED DESCRIPTION

[0060] The present invention obtains a data-driven intelligent evaluation method for the biodegradability of wastewater, and selects variables related to the biodegradability of petrochemical wastewater as biochemical oxygen demand (BOD), chemical oxygen demand (COD) in primary sedimentation tank, and chemical oxygen demand (COD) in influent. in, total phosphorus TP, ammonia nitrogen NH4-N, pH, dissolved oxygen DO, temperature T, suspended solids concentration SS, redox potential ORP, inlet flow Q in , turbidity, nitrate nitrogen NO3-N, sludge concentration MLSS, a total of 14 variables;

[0061] The experimental data comes from the daily water quality analysis report of a petrochemical wastewater treatment plant; the biochemical oxygen demand BOD, the chemical oxygen demand COD of the primary sedimentation tank, and the chemical oxygen demand COD of the influent are taken respectively. in , total phosphorus TP, ammonia nitrogen NH4-N, pH, dissolved oxygen DO, temperature T, suspended solids concentration SS, redox potential ORP, inlet flow Q in The test data of turbidity, nitrate nitrogen NO3-N, and sludge concentration MLSS are experimental sample data. After removing abnormal samples, 1000 sets of available data remain. All 1000 sets of data samples are divided into two parts: 800 sets of data are used as training samples, and the remaining 200 sets of data are used as test samples;

[0062] The present invention adopts the following technical solutions and implementation steps:

[0063] 1. A data-driven intelligent evaluation method for biodegradability of petrochemical wastewater, comprising the following steps:

[0064] (1) Determine the prediction variables and related variables: The present invention mainly conducts intelligent evaluation on the biodegradability of wastewater in the process of petrochemical wastewater treatment, and determines the biochemical oxygen demand (BOD), the chemical oxygen demand (COD) of the primary sedimentation tank, and the chemical oxygen demand (COD) of the influent. in , total phosphorus TP, ammonia nitrogen NH4-N, pH, dissolved oxygen DO, temperature T, suspended solids concentration SS, redox potential ORP, inlet flow Q in , turbidity, nitrate nitrogen NO3-N, sludge concentration MLSS, a total of 14 variables are wastewater biodegradability related variables, and the four categories of wastewater biodegradability, namely, difficult to degrade, partially degradable, relatively degradable, and easily degradable, are used as predictive variables. The four categories are defined as Class 1 to Class 4 respectively;

[0065] (2) Design a multi-scale feature extraction algorithm for selecting biodegradability characteristic variables of petrochemical wastewater:

[0066] ① According to the sampling period of related variables, the sampling period of dissolved oxygen DO, nitrate nitrogen NO3-N and redox potential ORP is 5 minutes, which are expressed as τ1, τ2, τ3 respectively. The obtained sampling sample vectors are x1(t1), x2(t2), x3(t3), where t1, t2, t3 are the sampling time, sludge concentration MLSS, turbidity, temperature T, pH and influent flow Q respectively. inThe sampling periods are 15 minutes, denoted as τ4, τ5, τ6, τ7, τ8 respectively, and the obtained sampling sample vectors are x4(t4), x5(t5), x6(t6), x7(t7), x8(t8) respectively, where t4, t5, t6, t7, t8 are the sampling times, ammonia nitrogen NH4-N, primary sedimentation tank chemical oxygen demand COD, influent chemical oxygen demand COD in The sampling period is 2 hours, denoted as τ9, τ 10 , τ 11 , and the obtained sampling sample vectors are x9(t9), x 10 (t 10 ), x 11 (t 11 ), where t9, t 10 , t 11 are the sampling times respectively, biochemical oxygen demand BOD, suspended solid concentration SS, total phosphorus TP. The sampling period is 1 day, denoted as τ 12 , τ 13 , τ 14 , and the obtained sampling sample vectors are x 12 (t 12 ), x 13 (t 13 ), x 14 (t 14 ), where t 12 , t 13 , t 14 are the sampling times respectively. Define the sampling scale vector of the variable as τ = [τ1, …, τ i , …, τ 14 , where τ i is a positive integer, i = 1, 2, …, 14, and x i (t i ) = [x i (1), x i (2), …, x i (t i ), …, x i (n i )] are the elements in the vector; define the sampling time corresponding to each variable as t i = 1 + kτ i , k = 0, 1, …, n i - 1, n i is the number of samples of the i-th variable and is a positive integer;

[0067] ② Define the step size of time scale fusion as m, m0 ≤ m ≤ m1, m, m0 and m1 are positive integers and m0 < m1 ≤ 14. Define the number of samples of the fused variable as N, that is, n i = N; when τ iWhen τ = m, the sampling vector after fusing the scales of each variable is x i ′(t i ′) = x i (t i ), where t i ′ is the sampling time of the fused variable; when τ i < m, the sampling time of each variable becomes t i ′ = 1 + km, and the sampling vector after fusing the scales of each variable is x i ′(t i ′) = x i (1 + km); when τ i > m, the sampling vector after fusing the scales of each variable is

[0068] x i ′(t i ′) = [x i (1), x i (1) + x i (n i ) - x i (1) / N, x i (1) + 2[x i (n i ) - x i (1)] / N, …, x i (1) + (N - 1)[x i (n i ) - x i (1)] / N];

[0069] ③ Set the number of feature extraction times L1 = 1;

[0070] ④ When j = L1, according to the obtained sampling sample vectors, merge them into a sample matrix X(t) i = [x1(t), x2(t), …, x h×o (t), …, x i (t)] in ascending order of n o (t), where x i (t) = x i ′(t i ′); h and o respectively represent the total number of samples and the number of variables of the sample matrix X(t) h×o , and h ≤ N, o ≤ 14; perform standardization transformation on the sample matrix X(t) h×o to obtain a standardized matrix X std (t) h×o = [z1(t), z2(t), …, z2(t), …, z o (t)], where:

[0071]

[0072] Where b = 1, 2, ..., o, x b (t) = [x b (1),x b (2),…,x b (t)] is the matrix X(t) h×o The sample vector of the bth variable in z b (t) = [z b (1),z b (2),…,z b (t)] is the matrix X std (t) h×o The standardized sample vector of the bth variable in , μ b and θ b The matrix x b The mean and standard deviation of (t) are calculated as follows:

[0073]

[0074]

[0075] ⑤According to the standardized matrix X std (t) h×o , calculate the corresponding covariance matrix R(t) o×o =[r1(t),r2(t),…,r o (t)], where the calculation formula is as follows:

[0076]

[0077] Among them, r b (t) = [r b (1),r b (2),…,r b (t)] is the matrix R(t) o×o The sample vector of the bth variable in , represents the correlation coefficient vector between the bth variable and the cth variable, z c (t) = [z c (1),z c (2),…,z c (t)] is the matrix X std (t) h×o The sample vector of the cth variable in , c = 1, 2, ..., o, T is the transpose of the formula;

[0078] ⑥ Solve the covariance matrix R(t) o×o The eigenvalues ​​and corresponding eigenvectors are λ1,λ2,…,λ o and p1,p2,…,p o ;

[0079] ⑦ Arrange the eigenvalues ​​in descending order and calculate the cumulative contribution rate η(e) of the first e related variables. The calculation formula is as follows:

[0080]

[0081] Among them, e∈[1,o], λ c Represents the matrix R(t) o×o The cth eigenvalue is selected, and the first e related variables with cumulative contribution rate η(e) exceeding 85% are selected as the characteristic variables of sewage biodegradability. The extracted characteristic variable set X e (t) is expressed as follows:

[0082] X e (t)=[x1(t),x2(t),…,x j (t),...,x e (t)]; (20)

[0083] Among them, x j (t) = [x j (1),x j (2),…,x j (t)], j = 1, 2, ..., e, is the set X e The jth sample vector in (t);

[0084] ⑧The number of feature extractions L1 increases by 1. If the number of extractions L1 ≤ N, go to step ④ to continue training, otherwise stop feature extraction;

[0085] After extracting features using a multi-scale feature extraction algorithm, the characteristic variables related to the biodegradability of petrochemical wastewater are finally determined to be redox potential ORP, influent pH, influent ammonia nitrogen NH4-N, influent chemical oxygen demand COD, influent total phosphorus TP, influent suspended solids SS, and influent biochemical oxygen demand BOD, which are recorded as x1(t), x2(t),…,x7(t) respectively;

[0086] (3) Design of intelligent evaluation model for biodegradability of sewage: A fuzzy neural network is used to establish a biodegradability evaluation model for sewage. The topological structure of the model is divided into four layers: input layer, RBF layer, normalization layer, and output layer. The connection mode of the neural network is 7-14-14-4, that is, the number of neurons in the input layer is 7, the number of neurons in the RBF layer is 14, the number of neurons in the normalization layer is 14, and the number of neurons in the output layer is 4. The weights of the neural network are randomly assigned. The input of the fuzzy neural network is expressed as X(t) = [x1(t), x2(t), …, x7(t)], and the expected output of the neural network is expressed as y d (t) = [y d 1(t),y d 2 (t),y d 3 (t),y d 4 (t)], the actual output is expressed as y(t) = [y1(t), y2(t), y3(t), y4(t)], y d q (t) and y q (t) represents the expected output and actual output of the network corresponding to the qth category, q = 1, 2, 3, 4, the expected outputs of the 1st to 4th categories are defined as [1, 0, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0] and [0, 0, 0, 1] respectively. The input of the fuzzy neural network at the tth time is expressed as x1(t), x2(t), …, x7(t). The calculation function of the fuzzy neural network is:

[0087] ① Input layer: This layer consists of 7 neurons, and the output of each neuron is:

[0088] u I (t) = x I (t),I=1,2,…,7; (21)

[0089] Among them, u I (t) represents the output of the Ith neuron in the input layer at time t, x I (t) represents the input of the Ith neuron in the input layer at time t;

[0090] ②RBF layer: The RBF layer consists of 14 neurons, and the output of each neuron is:

[0091]

[0092] in, is the output of the Jth RBF neuron, J = 1, 2, ..., 14, c IJ (t) and σ IJ (t) are the center value and Gaussian width of the I-th membership function in the J-th neuron of the RBF layer, c IJ (t)∈[-2,2],σ IJ (t)∈[0.01,1];

[0093] ③ Normalization layer: The normalization layer consists of 14 neurons, and its output calculation formula is:

[0094]

[0095] Among them, v z (t) is the output of the zth neuron in the normalization layer;

[0096] ④Output layer: The output of the output layer is:

[0097]

[0098] Among them, y q (t) is the output of the qth neuron in the output layer, indicating the output probability of the network on the qth category, q = 1, 2, 3, 4, w q (t) represents the connection weight between the normalization layer neuron and the qth neuron in the output layer, w z q (t) represents the connection weight between the zth neuron in the normalization layer and the qth neuron in the output layer, v(t) = [v1(t), v2(t), …, v p (t)] is the output vector of the neurons in the normalization layer;

[0099] ⑤Define the error function as:

[0100]

[0101] Among them, y d q (t) is the expected output of the qth neuron in the output layer;

[0102] (5) Training the neural network, specifically:

[0103] ① Given that the number of neurons in the RBF layer and the normalization layer of the fuzzy neural network is 14, the model input is X(1), X(2), ..., X(t), ..., X(P), P = 200 is the number of training samples, training is performed and the calculation step L2 = 1 is designed;

[0104] ②t=L2, calculate the output y(t) of the model, and use the gradient descent method to adjust the parameters of the dynamic evaluation model;

[0105]

[0106]

[0107]

[0108] Among them, c IJ (t),σ IJ (t) and w z q (t) are the center value, width and weight before model adjustment, c IJ (t+1),σ IJ (t+1) and w z q(t + 1) are the center value, width, and weight after model adjustment, and η ∈ (0, 0.1] represents the learning rate;

[0109] ③ Increase the learning step L2 by 1. If the step L2 < P, go to step ② for continued training. If L2 = P, stop the calculation;

[0110] (6) Use the test sample data as the input of the trained fuzzy neural network. The category corresponding to the maximum output of the fuzzy neural network is the evaluation result of the biodegradability of the sewage; Figure 2 Display the confusion matrix diagram for the evaluation of the biodegradability of the sewage. X-axis: The actual categories of the biodegradability of the sewage, which are difficult to degrade, partially degradable, relatively easy to degrade, and easy to degrade, respectively marked as 1 - 4; Y-axis: The categories of the biodegradability of the sewage predicted by the model output. The four predicted output categories are difficult to degrade, partially degradable, relatively easy to degrade, and easy to degrade, respectively marked as 1 - 4; The black squares on the diagonal of the confusion matrix are the number of correct evaluations and their precision rates. The square e(q1, q2) is the number of mis-evaluations of the q2-th sewage biodegradability category to the q1-th category and its proportion, where q1 = 1, 2, 3, 4 and q2 = 1, 2, 3, 4.

Claims

1. A data-driven intelligent evaluation method for biodegradability of petrochemical wastewater. Features The following steps are involved: (1) Determine the predictor variables and related variables: Intelligently evaluate the biodegradability of wastewater during petrochemical wastewater treatment to determine the biochemical oxygen demand (BOD), the chemical oxygen demand (COD) of the primary sedimentation tank, and the chemical oxygen demand (COD) of the influent. in , total phosphorus TP, ammonia nitrogen NH 4 -N, pH, dissolved oxygen DO, temperature T, suspended solids concentration SS, redox potential ORP, inlet flow Q in , turbidity, nitrate nitrogen NO 3 -N, sludge concentration MLSS, a total of 14 variables are wastewater biodegradability-related variables, with four categories of wastewater biodegradability, namely, difficult to degrade, partially degradable, relatively degradable, and easily degradable, as predictor variables, and the four categories are defined as categories 1 to 4; (2) Design a multi-scale feature extraction algorithm for selecting biodegradability characteristic variables of petrochemical wastewater: ① According to the sampling period of relevant variables, including dissolved oxygen DO, nitrate nitrogen NO 3 The sampling period of -N and ORP is 5 minutes, denoted as τ 1 ,τ 2 ,τ 3 The obtained sampling vectors are x 1 (t 1 ),x 2 (t 2 ),x 3 (t 3 ), where t 1 ,t 2 ,t 3 They are sampling time, sludge concentration MLSS, turbidity, temperature T, pH and influent flow rate Q in The sampling period is 15 minutes, denoted as τ 4 ,τ 5 ,τ 6 ,τ 7 ,τ 8 The obtained sampling vectors are x 4 (t 4 ),x 5 (t 5 ),x 6 (t 6 ),x 7 (t 7 ),x 8 (t 8 ), where t 4 ,t 5 ,t 6 ,t 7 ,t 8 are the sampling time, ammonia nitrogen NH 4 -N, chemical oxygen demand (COD) in primary sedimentation tank, chemical oxygen demand (COD) in influent in The sampling period is 2 hours, denoted as τ 9 ,τ 10 ,τ 11 The obtained sampling vectors are x 9 (t 9 ),x 10 (t 10 ),x 11 (t 11 ), where t 9 ,t 10 ,t 11 are the sampling time, biochemical oxygen demand BOD, suspended solids concentration SS, and total phosphorus TP. The sampling period is 1 day, expressed as τ 12 ,τ 13 ,τ 14 , the obtained sampling sample vectors are x 12 (t 12 ),x 13 (t 13 ),x 14 (t 14 ), where t 12 ,t 13 ,t 14 are sampling moments respectively, and the sampling scale vector of the variable is defined as τ=[τ 1 ,…,τ i ,…,τ 14 ], where τ i is a positive integer, i=1,2,…,14, and x i (t i )=[x i (1),x i (2),…,x i (t i ),…,x i (n i )] is an element in the vector; define the sampling time corresponding to each variable as t i =1+kτ i ,k=0,1,…,n i -1,n i is the number of samples of the i-th variable, and is a positive integer; ② Define the step size of time-scale fusion as m, where m 0 ≤m≤m 1 , m, m 0 and m 1 are positive integers and m 0 <m 1 ≤14. Define the number of variable samples after fusion as N, i.e., n i =N; when τ i =m, the sampling vector of each variable scale after fusion is x i ′(t i ′)=x i (t i ), t i ′ is the sampling time of the fused variable; when τ i <m, the sampling time of each variable becomes t′ i =1 + km, and the sampling vector of each variable scale after fusion is x i ′(t i ′)=x i (1 + km); when τ i >m, the sampling vector of each variable scale after fusion is x i ′(t i ′)=[x i (1),x i (1)+x i (n i )-x i (1) / N,x i (1)+2[x i (n i )-x i (1)] / N,…,x i (1)+(N-1)[x i (n i )-x i (1)] / N]; ③Set the number of feature extraction times L 1 =1; ④j=L 1 , according to the obtained sampling sample vector according to n i Merge from small to large to form the sample matrix X(t) h×o =[x 1 (t),x 2 (t),…,x i (t),…,x o (t)], where x i (t) = x i ′(t i ′); h and o represent the sample matrix X(t) h×o The total number of samples and the number of variables, and h≤N, o≤14; the sample matrix X(t) h×o Perform standardization transformation to obtain the standardized matrix X std (t) h×o =[z 1 (t),z 2 (t),…,z 2 (t),…,z o (t)], where: z b (t)=(x b (t)-μ b ) / θ b ; (1) Where b = 1, 2, ..., o, x b (t) = [x b (1),x b (2),…,x b (t)] is the matrix X(t) h×o The sample vector of the bth variable in z b (t) = [z b (1),z b (2),…,z b (t)] is the matrix X std (t) h×o The standardized sample vector of the bth variable in , μ b and θ b The matrix x b The mean and standard deviation of (t) are calculated as follows: ⑤According to the standardized matrix X std (t) h×o , calculate the corresponding covariance matrix R(t) o×o =[r 1 (t),r 2 (t),…,r o (t)], where the calculation formula is as follows: Among them, r b (t) = [r b (1),r b (2),…,r b (t)] is the matrix R(t) o×o The sample vector of the bth variable in , represents the correlation coefficient vector between the bth variable and the cth variable, z c (t) = [z c (1),z c (2),…,z c (t)] is the matrix X std (t) h×o The sample vector of the cth variable in , c = 1, 2, ..., o, T is the transpose of the formula; ⑥ Solve the covariance matrix R(t) o×o The eigenvalues ​​and corresponding eigenvectors of are λ 1 ,λ 2 ,…,λ o and p 1 ,p 2 ,…,p o ; ⑦ Arrange the eigenvalues ​​in descending order and calculate the cumulative contribution rate η(e) of the first e related variables. The calculation formula is as follows: Among them, e∈[1,o], λ c Represents the matrix R(t) o×o The cth eigenvalue is selected, and the first e related variables with cumulative contribution rate η(e) exceeding 85% are selected as the characteristic variables of sewage biodegradability. The extracted characteristic variable set X e (t) is expressed as follows: X e (t)=[x 1 (t),x 2 (t),…,x j (t),...,x e (t)]; (6) Among them, x j (t) = [x j (1),x j (2),…,x j (t)], j = 1, 2, ..., e, is the set X e The jth sample vector in (t); ⑧Number of feature extractions L 1 Increase by 1 if the number of extractions is L 1 ≤N, go to step ④ to continue training, otherwise stop feature extraction; After the features are extracted by the multi-scale feature extraction algorithm, the first e variables that meet the contribution rate η(e)>85% are selected in descending order from the eigenvalue as the feature variables; (3) Design of intelligent evaluation model for biodegradability of sewage: A fuzzy neural network is used to establish a sewage biodegradability evaluation model. The topological structure of the model is divided into four layers: input layer, RBF layer, normalization layer, and output layer. The connection mode of the neural network epp-4 is determined, that is, the number of neurons in the input layer is e, the number of neurons in the RBF layer is p, the number of neurons in the normalization layer is p, and the number of neurons in the output layer is 4. The weights of the neural network are randomly assigned. The input of the fuzzy neural network is expressed as X(t)=[x 1 (t),x 2 (t),…,x e (t)], the expected output of the neural network is denoted as y d (t) = [y d 1 (t),y d 2 (t),y d 3 (t),y d 4 (t)], the actual output is expressed as y(t) = [y 1 (t),y 2 (t),y 3 (t),y 4 (t)],y d q (t) and y q (t) represents the expected output and actual output of the network corresponding to the qth category, q = 1, 2, 3, 4, the expected outputs of the 1st to 4th categories are defined as [1, 0, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0] and [0, 0, 0, 1] respectively. The input of the fuzzy neural network at the tth time is represented by x 1 (t),x 2 (t),…,x a (t), the computational function of the fuzzy neural network is: ① Input layer: This layer consists of e neurons, and the output of each neuron is: u I (t)=x I (t),I=1,2,…,e; (7) Among them, u I (t) represents the output of the Ith neuron in the input layer at time t, x I (t) represents the input of the Ith neuron in the input layer at time t; ②RBF layer: The RBF layer consists of p neurons, and the output of each neuron is: in, is the output of the Jth RBF neuron, J = 1, 2, ..., p, c IJ (t) and σ IJ (t) are the center value and Gaussian width of the I-th membership function in the J-th neuron of the RBF layer, c IJ (t)∈[-2,2],σ IJ (t)∈[0.01,1]; ③ Normalization layer: The normalization layer consists of p neurons, and its output calculation formula is: Among them, v z (t) is the output of the zth neuron in the normalization layer; ④Output layer: The output of the output layer is: Among them, y q (t) is the output of the qth neuron in the output layer, indicating the output probability of the network on the qth category, q = 1, 2, 3, 4, w q (t) represents the connection weight between the normalization layer neuron and the qth neuron in the output layer, w z q (t) represents the connection weight between the zth neuron in the normalization layer and the qth neuron in the output layer, v(t) = [v 1 (t),v 2 (t),…,v p (t)] is the output vector of the neurons in the normalization layer; ⑤Define the error function as: Among them, y d q (t) is the expected output of the qth neuron in the output layer; (5) Training the neural network, specifically: ① Given the number of neurons in the RBF layer and normalization layer of the fuzzy neural network is p, p is a natural number, p∈[10,20], the model input is X(1), X(2), …, X(t), …, X(P), P is the number of training samples, conduct training and design the calculation steps L 2 =1; ②t=L 2 , calculate the output y(t) of the model, and use the gradient descent method to adjust the parameters of the dynamic evaluation model; Among them, c IJ (t),σ IJ (t) and w z q (t) are the center value, width and weight before model adjustment, c IJ (t+1),σ IJ (t+1) and w z q (t+1) are the center value, width and weight of the model after adjustment, η∈(0,0.1] represents the learning rate; ③Learning step count L 2 Increment by 1 if the step count L 2 <P, go to step ② for continued training, if L 2 = P, stop the calculation; (6) Use the y(t) obtained by the trained neural network to evaluate the biodegradability of the sewage, that is, select the y(t) with the maximum output probability in the output layer of the fuzzy neural network. δ The category corresponding to the node of (t) is taken as the final classification result; δ (t) = max{y 1 (t),y 2 (t),y 3 (t),y 4 (t)}, δ = 1, 2, 3, 4, max represents the symbol of taking the maximum value in the set; if y 1 (t) is the maximum output probability, that is, δ = 1, then the evaluation result is difficult to degrade. 2 (t) is the maximum output probability, that is, δ = 2, then the evaluation result is partially degradable. If y 3 (t) is the maximum output probability, that is, δ = 3, then the evaluation result is relatively easy to degrade. 4 (t) is the maximum output probability, that is, δ=4, and the evaluation result is easy to degrade.

Images