Prediction method for coagulation dosing of sewage treatment high-density pool

Abstract

The invention discloses a sewage treatment high-density pool coagulation dosing prediction method, which accurately and quickly predicts the optimal time and the optimal dosage of coagulation dosing by relying on a resampling technology and an AO algorithm on the basis of collecting real-time and non-real-time sewage treatment data and further utilizing a data processing algorithm to carry out dimension reduction on the data. Important reference information is provided for real-time management and control of a sewage treatment plant, for example, application services such as resource management, water quality monitoring and sustainable water treatment can be realized.

Description

Technical Field

[0001] The invention belongs to the field of industrial sewage treatment, and in particular relates to a method for predicting coagulation and dosing in a high-density sewage treatment tank. Background Art

[0002] The dosing data of industrial wastewater has the characteristics of nonlinearity and weak periodicity. In its actual delivery process, it is necessary to comprehensively consider multiple factors. Through the collected data, some data mining algorithms can be used to discover the changing rules of dosing data, so as to achieve the correct, timely and efficient delivery of dosing, thereby saving resources and costs. In recent years, the research on the prediction of coagulation dosing in high-density tanks for industrial wastewater treatment mainly includes:

[0003] The Chinese patent "Dosage Optimization Control Method Based on VSM-GRU Total Phosphorus Prediction Model" (CN117763352A) uses deep learning to construct a VSM-GRU model to accurately and efficiently predict the ammonia nitrogen concentration in sewage, and combines the DDPG algorithm to obtain the optimal dosing strategy, thereby reducing the amount of coagulant added while ensuring that the effluent quality meets the discharge standard; the Chinese patent "High-density Clarification Tank Dosage Prediction and Control Method Based on Multidimensional Scoring Model" (CN114781249A) uses historical alum flower image data to train a deep convolutional neural network recognition model, and through a multidimensional scoring method, a coagulant and coagulant dosage prediction model is established to achieve accurate dosage control; the Chinese patent "Coagulant Dosage Prediction Method Based on BR-NARX Neural Network" (CN115481720A) uses the BR algorithm to optimize the NARX neural network weights, solving the problem that other models ignore the lag effect of input variables, and provides an optimization control method for the coagulant dosing system in the sewage treatment process.

[0004] The above inventions all provide a method for predicting the dosage of sewage treatment, aiming to provide guidance for the coagulant dosing system in the sewage treatment process. However, in the case of large data volume and complex characteristics, there are still the following defects in quickly predicting and realizing the dosage of the dosage: 1) How to further reduce the dimensionality of a large number of multi-source heterogeneous data samples and optimize the relevant parameters of the model when the dosage prediction model is feasible, thereby improving the efficiency of the coagulation dosage prediction model and promoting the timely and appropriate dosage of the dosage; 2) How to comprehensively evaluate and verify the model prediction knowledge, improve the reliability and effectiveness of the prediction results, and realize the application service of dosage prediction in sewage treatment. Summary of the invention

[0005] The purpose of the present invention is to provide a method for predicting coagulation and dosing in a high-density sewage treatment tank, so as to accurately predict the optimal time and optimal dosage of coagulation and dosing, and provide important reference information for real-time control of sewage treatment plants.

[0006] The object of the present invention is achieved by the following technical means: a method for predicting coagulation and dosing in a high-density sewage treatment tank, comprising the following steps:

[0007] Step 1: Obtain the sewage treatment data of the high-density pool, and aggregate various sewage treatment data into a data set containing multiple features;

[0008] Step 2: Data cleaning: remove duplicates and missing items from the data set and process outliers;

[0009] Step 3: data resampling: resampling and normalizing the processed data;

[0010] Step 4: Data division: divide the data into training set and test set;

[0011] Step 5: Model training. Establish an ELM prediction model, initialize the input weights from the input layer to the hidden layer and the hidden layer node bias of the ELM model, train the ELM model using the training set data, input the test set data into the trained ELM model for prediction, and use the population fitness function to evaluate the ELM model. If the population fitness value does not meet the design requirements, use the AO optimization algorithm to optimize the iterative input weights and hidden layer node biases.

[0012] After each optimization is completed, the new input weights and hidden layer node biases are input into the ELM model, and the training set training, test set prediction, and population fitness function evaluation are repeated until the population fitness function evaluation meets the requirements. At this time, the model obtained is the AO-ELM model for prediction;

[0013] Step six, dosage prediction, through the AO-ELM model, input the actual collected high-density pool sewage treatment data to predict the dosage.

[0014] The outlier processing is to find the lower limit, upper limit and median of the data set through the box plot method, and for each data point in the data set, if its value is less than the lower limit or greater than the upper limit, it is replaced by the median.

[0015] Before the step three, the types of sewage treatment data are also screened, and through feature selection, sewage treatment data corresponding to features with a high correlation with the prediction index are screened out as data for subsequent use.

[0016] The data resampling is to further downsample the data to 10 minutes, 20 minutes and 30 minutes granularity.

[0017] The data division is specifically to divide the data into multiple groups, one of which is a test set and the rest are training sets.

[0018] The ELM model is,

[0019]

[0020] For N training samples (x i ,y i ), (i=1,2,...,N), x i =[x1,x2,...,x n ] T ∈R n represents the i-th input sample, y i =[y1,y2,...,y m ] T ∈R m represents the expected output corresponding to the i-th sample, where n represents the number of neurons in the input layer of the model, m represents the number of neurons in the output layer, the network has L hidden layer nodes, and the input weight from the input layer to the hidden layer is ω i =[ω1,ω2,...,ω n ] T , the output weight from the hidden layer to the output layer is β = β i =[β1,β2,...,β m ] T , b i represents the bias of the i-th hidden layer node, and g(·) is the activation function;

[0021] The population fitness function is:

[0022]

[0023] where predict i is the predicted value of the model on the test set; true i is the true value of the model on the test set; p is the number of test samples; i is each individual in the population.

[0024] The specific method of using the AO optimization algorithm to optimize the input weights from the input layer to the hidden layer and the hidden layer node bias in the ELM model is:

[0025] Set the population size, maximum number of iterations, exploration and development parameters, and set the current number of iterations to 1;

[0026] Initialize the population position of the AO algorithm;

[0027] The input weight ω from the input layer to the hidden layer in the ELM model i and the hidden layer node bias b i As the location information of the Sky Eagle, the population fitness is iteratively updated through four optimization strategies, and the new Sky Eagle best individual location information and fitness are obtained;

[0028] Until the preset maximum number of iterations is reached or the preset conditions are met, the input weight ω corresponding to the best individual position information of the Eagle is output i and the hidden layer node bias b i into the ELM model.

[0029] When t≤2 / 3T and rand≤0.5, use method 1 to update the position of the Eagle;

[0030] When t≤2 / 3T and rand>0.5, use method 2 to update the position of the Eagle;

[0031] When t>2 / 3T and rand≤0.5, use method 3 to update the position of the Eagle;

[0032] When t>2 / 3T and rand>0.5, use method 4 to update the position of the Eagle;

[0033] t is the current iteration number, T is the maximum iteration number, and rand represents a value in the range [0,1]

[0034] Randomly generated numbers;

[0035] Methods one to four are respectively expanding the search range, narrowing the search range, expanding the development range and narrowing the development range in the AO algorithm.

[0036] The beneficial effects of the present invention are: based on the real-time monitoring of key parameters in the sewage treatment process by combining IoT devices, the dimension reduction of massive data is achieved through data preprocessing and feature screening methods, thereby improving the operation efficiency and calculation speed of the model. With the application of resampling and optimization algorithms, the accuracy and robustness of the prediction model are effectively improved, so that it can better adapt to changes and fluctuations in actual operation, thereby promoting efficiency and quality in the actual sewage treatment process;

[0037] Based on the advantages of resampling technology to smooth data and the powerful global search capability of Aquila Optimizer (AO), the design and implementation of a predictive plan for coagulation and dosing can be achieved through data collection, data processing, and data mining. This method can not only improve the efficiency and reliability of the industrial wastewater treatment production process, but also provide a sustainable solution for environmental protection and resource utilization;

[0038] The Eagle Optimization Algorithm is used to determine the optimal parameters of the Extreme Learning Machine (ELM) prediction model to improve the accuracy and reliability of the prediction results; the best time and optimal dosage of coagulation dosing can be accurately and quickly predicted, providing important reference information for real-time management and control of sewage treatment plants. For example, it can realize application services such as resource management, water quality monitoring and sustainable water treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A flow chart of the prediction method for coagulation and dosing in high-density sewage treatment tanks;

[0040] Figure 2 It is the flow chart of AO algorithm;

[0041] Figure 3 This is a schematic diagram of expanding the search range in the AO algorithm;

[0042] Figure 4 This is a schematic diagram of narrowing the search range in the AO algorithm;

[0043] Figure 5 Schematic diagram of expanding the development scope in AO algorithm;

[0044] Figure 6 This is a schematic diagram of narrowing the development scope in the AO algorithm;

[0045] Figure 7 is the fitness curve of the AO algorithm;

[0046] Figure 8 This is the effect comparison between the original value and the predicted value at a 1-minute granularity;

[0047] Fig. 9 This is the effect comparison between the original value and the predicted value at a 10-minute granularity;

[0048] Fig.10 This is the effect comparison between the original value and the predicted value at a granularity of 20 minutes;

[0049] Fig.11 This is the effect comparison between the original value and the predicted value at a 30-minute granularity;

[0050] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. DETAILED DESCRIPTION

[0051] [Example 1]

[0052] like Figure 1 As shown, a method for predicting coagulation and dosing in a high-density sewage treatment tank comprises the following steps:

[0053] Step 1: Obtain the sewage treatment data of the high-density pool, and aggregate various sewage treatment data into a data set containing multiple features;

[0054] Based on the Internet of Things technology, IoT devices such as smart meters, pH sensors, temperature sensors and humidity sensors are deployed at each process end of the high-density pool. Wastewater treatment data such as influent flow (L / h), influent turbidity (NTU), influent pH, influent temperature (℃), raw water conductivity (μs / cm), raw water dissolved oxygen (mg / L), raw water COD (mg / L), raw water ammonia nitrogen (mg / L) and dosage (L / h) are collected every 1 minute. The collected data are then uploaded to the enterprise database for further processing. Through the Internet of Things technology, a large amount of data is collected to lay a data foundation for the prediction of coagulation dosage.

[0055] Step 2: Data cleaning: remove duplicates and missing items from the data set and process outliers;

[0056] In order to ensure the validity of the data, data cleaning operations are required.

[0057] Duplicates and missing items were removed from the huge data set, and appropriate and stable sewage treatment data for ten months were selected.

[0058] Next, it is necessary to remove the outliers in the data set. The outlier processing is to find the lower limit, upper limit and median of the data set through the box plot method. For each data point in the data set, if its value is less than the lower limit or greater than the upper limit, it is replaced by the median.

[0059] Using the box plot method, we can find the lower limit Min, interquartile range IQR, lower quartile Q1, median Q2, upper quartile Q3 and upper limit Max of the data set. The relevant calculation methods are as follows:

[0060] Q i =i(n+1) / 4 (5)

[0061] Where i = 1, 2, 3, and n represents the number of items contained in the sequence.

[0062] IQR=Q3-Q1 (6)

[0063] Max=Q3+1.5IQR (7)

[0064] Min=Q1-1.5IQR (8)

[0065] After drawing the box plot, for each data point X i , if its value is less than the lower limit Min or greater than the upper limit Max, replace it with the median:

[0066]

[0067] Before the step three, the types of sewage treatment data are also screened, and through feature selection, sewage treatment data corresponding to features with a high correlation with the prediction index are screened out as data for subsequent use.

[0068] Since there are too many types of collected data, in order to simplify the calculation and improve the convergence speed of the model, the recursive feature elimination (RFE) algorithm is integrated to perform feature selection. The number of features to be selected is specified. In this embodiment, 5 are selected. The RFE algorithm will iteratively train the model. After each training, the importance of the features is ranked, and the least important features are selected and eliminated until the specified number of features is reached. Finally, five features with a high correlation with the prediction index are screened out, namely, influent flow, influent temperature, raw water COD, influent turbidity, and influent pH.

[0069] As shown in the table below, raw water conductivity, raw water dissolved oxygen, and raw water ammonia nitrogen have little correlation with the prediction indicators and are removed from the data set. The data of these three items will no longer be used in subsequent model training.

[0070]

[0071] Step 3: data resampling: resampling and normalizing the processed data;

[0072] The data resampling is to further downsample the data to 10 minutes, 20 minutes and 30 minutes granularity.

[0073] Based on the processed multi-source heterogeneous real-time and non-real-time data, resampling is performed on them.

[0074] The data is further downsampled to 10 minutes, 20 minutes, and 30 minutes granularity to smooth the data and improve prediction efficiency. The formula for mean resampling can be expressed as:

[0075]

[0076] In the formula, X new Indicates the new record value, X i It represents the value of the i-th time period in the original data record, and n represents the number of original data records in each time scale.

[0077] Step 2.2: To eliminate the impact of different dimensions and ensure the analyzability and confidentiality of the data, normalize the data:

[0078]

[0079] Where X represents the original data, X max Indicates the maximum value of the data, X min represents the minimum value of the data, and X' represents the normalized data.

[0080] Step 4: Data division: divide the data into training set and test set;

[0081] The data division is specifically to divide the data into multiple groups, one of which is a test set and the rest are training sets.

[0082] The ten-month sewage treatment data is divided into a training set and a test set, where the training set takes the data from the first nine months and the test set takes the data from the last month, that is, the dosage data from the first nine months is used to predict the dosage data for the last month.

[0083] Step 5: Model training. Establish an ELM prediction model, initialize the input weights from the input layer to the hidden layer and the hidden layer node bias of the ELM model, train the ELM model using the training set data, input the test set data into the trained ELM model for prediction, and use the population fitness function to evaluate the ELM model. If the population fitness value does not meet the design requirements, use the AO optimization algorithm to optimize the iterative input weights and hidden layer node biases.

[0084] After each optimization is completed, the new input weights and hidden layer node biases are input into the ELM model, and the training set training, test set prediction, and population fitness function evaluation are repeated until the population fitness function evaluation meets the requirements. At this time, the model obtained is the AO-ELM model for prediction;

[0085] The ELM model is,

[0086]

[0087] For N training samples (x i ,y i ), (i=1,2,...,N), x i =[x1,x2,...,x n ] T ∈R n represents the i-th input sample, y i =[y1,y2,...,y m ] T ∈R m represents the expected output corresponding to the i-th sample, where n represents the number of neurons in the input layer of the model, m represents the number of neurons in the output layer, the network has L hidden layer nodes, and the input weight from the input layer to the hidden layer is ω i =[ω1,ω2,...,ω n ] T, the output weight from the hidden layer to the output layer is β = β i =[β1,β2,...,β m ] T , b i represents the bias of the i-th hidden layer node, and g(·) is the activation function;

[0088] The above equation (8) is converted into matrix form:

[0089] Hβ=Y (13)

[0090] Where: H is the output matrix of the ELM hidden layer nodes, Y is the model expected output matrix, and

[0091]

[0092] In order to achieve the training effect, we need to get w i , b i , β i , so that:

[0093]

[0094] In the ELM model learning process, the input weight ω i and the hidden layer bias b i is randomly generated. Therefore, when the number of hidden layer nodes and the activation function are determined, only the output weight β is unknown. The only solution can be obtained by Moore-Penrose generalized inverse calculation:

[0095] (17)

[0096] in, It represents the generalized inverse matrix of the expected output matrix Y, so the essence of model training is to find the least squares solution of Hβ=Y process.

[0097] Each time you train, first set the input weight ω i and the hidden layer bias b i The specific value of (the first one is the initial value, and the subsequent ones are optimized and iterated by the AO algorithm) is used to train the ELM model through the training set data to obtain the output weight β. At this time, the three parameters in the ELM model have been determined and can be used for prediction;

[0098] Next, the accuracy of the ELM model needs to be evaluated. By inputting the water flow, inlet water temperature, raw water COD, inlet water turbidity and inlet water pH in the test set data, the corresponding dosage prediction value is obtained;

[0099] Then, according to the actual value of the dosage in the test set, the population fitness is calculated. If the population fitness meets the requirements, it is considered that the input weight ω used at this time i and the hidden layer bias b i is the optimal value, and the trained ELM model can be used for dosing prediction, which is called AO-ELM model;

[0100] If the population fitness does not meet the requirements, then the input weight ω i and the hidden layer bias b i Based on this, the AO algorithm is used to adjust the input weight ω i and the hidden layer bias b i Optimize and iterate once more; then train again with the training set, test with the test set, and evaluate with the population fitness function until the population fitness value meets the requirements.

[0101] The population fitness function is to evaluate the trained ELM model by the root mean square error on the test set.

[0102]

[0103] where predict i is the predicted value of the model on the test set (i.e., the predicted dosage based on water flow, inlet water temperature, raw water COD, inlet water turbidity and inlet water pH); true i is the true value of the model on the test set (corresponding to water flow, inlet temperature, raw water COD, inlet turbidity and inlet pH, actual dosage); p is the number of test samples; i is each individual in the population.

[0104] The specific method of using the AO optimization algorithm to optimize the input weights from the input layer to the hidden layer and the hidden layer node bias in the ELM model is:

[0105] Set the population size, maximum number of iterations, exploration and development parameters, and set the current number of iterations to 1;

[0106] Initialize the population position of the AO algorithm;

[0107] The input weight ω from the input layer to the hidden layer in the ELM model i and the hidden layer node bias b i As the location information of the Sky Eagle, the population fitness is iteratively updated through four optimization strategies, and the new Sky Eagle best individual location information and fitness are obtained;

[0108] Until the preset maximum number of iterations is reached or the preset conditions are met, the input weight ω corresponding to the best individual position information of the Eagle is output i and the hidden layer node bias b i into the ELM model.

[0109] like Figure 2 As shown, when t≤2 / 3T and rand≤0.5, method 1 is used to update the position of the Eagle;

[0110] When t≤2 / 3T and rand>0.5, use method 2 to update the position of the Eagle;

[0111] When t>2 / 3T and rand≤0.5, use method 3 to update the position of the Eagle;

[0112] When t>2 / 3T and rand>0.5, use method 4 to update the position of the Eagle;

[0113] t is the current iteration number, T is the maximum iteration number, and rand represents a value in the range [0,1]

[0114] Randomly generated numbers;

[0115] Methods one to four are respectively expanding the search range, narrowing the search range, expanding the development range and narrowing the development range in the AO algorithm.

[0116] Set the population size, maximum number of iterations, and exploration and exploitation parameters, and set the current number of iterations to 1.

[0117] Initialize the population position of the AO algorithm. The AO algorithm is a population optimization technique. Its optimization process starts from the initial population and randomly generates individuals within the predefined upper and lower bounds, as follows:

[0118]

[0119] x i,j =rand*(UB j -LB j )+LB j , i=1,2,…,N; j=1,2,…,Dim (20)

[0120] Where i represents the index in the population and j represents the index of the eagle. i,j represents the position of the jth eagle in the i-th population; rand represents a randomly generated number in the range [0,1]; N represents the number of eagles in the population; Dim represents the dimension of the eagle population; UB and LB represent the upper and lower bounds of the problem to be solved respectively.

[0121] Input weight ω into the ELM model i and the hidden layer bias b i As the location information of the Skyhawk.

[0122] According to the idea of ​​AO algorithm, operations such as expanding the exploration phase, reducing the exploration phase, expanding the development phase, and reducing the development phase are carried out. The population position is continuously updated to find the optimal model parameters.

[0123] The AO algorithm imitates the specific behavioral strategies of the eagle during its hunting process, including: a) expanding the search range by soaring vertically to hunt birds; b) attacking prey at low altitude close to the ground through short-distance gliding attacks; c) gradually approaching prey by flying at low altitude and slowly descending; d) completing the entire process by diving down, then walking on the ground, and finally capturing prey.

[0124] (1) Expand the search scope

[0125] The search space is selected by the strategy of high flying and vertical descent. In this process, the eagle will quickly rise from the ground, and once it finds prey, it will glide at a nearly vertical angle. As the distance shortens, the speed also increases accordingly, as shown in the following figure. Figure 3 The corresponding formula is as follows:

[0126]

[0127] Where X(t) and X(t+1) represent the individual positions of the Tianying optimization algorithm in the tth and t+1th iterations, respectively. best (t) represents the best individual position obtained by the algorithm up to the tth iteration, X M (t) represents the average position of the population at the tth iteration, and T represents the maximum number of iterations of the algorithm.

[0128] (2) Narrow your search

[0129] like Figure 4 As shown in the figure, the individual eagle will gradually approach the target prey in a spiraling manner in the air to create favorable conditions for landing and attacking. This strategy aims to find the best attack angle and ensure the success rate of the attack while maintaining flight stability. The relevant formula is as follows:

[0130] X2(t+1)=X best (t)*Levy+X R (t)+(yx)*rand (23)

[0131]

[0132] Where Levy(D) represents the Levy flight distribution strategy, s is a constant with a value of 0.01, and u and v are random numbers between 0 and 1.

[0133]

[0134] τ(x) is the gamma function, and β is a constant with a fixed value of 1.5.

[0135] y and x represent the shape of the spiral flight in the search,

[0136] y=r*cos(θ) (26)

[0137] x=r*sin(θ) (27)

[0138] r represents the search step size, i.e. the spiral radius, and D1 is an integer matrix from 1 to the length of the search space.

[0139] r=r1+U*D1 (28)

[0140] θ represents the helix angle,

[0141] θ=-ω*D1+θ1 (29)

[0142] θ1 represents the starting spiral angle,

[0143]

[0144] Among them, r1 ranges from 1 to 20 search cycles, U takes a value of 0.00565, and ω takes a value of 0.005.

[0145] (3) Expand the scope of development

[0146] When a hawk is about to land and attack in a hunting area, it usually descends vertically, such as Figure 5 As shown, the mathematical formula is as follows:

[0147] X3(t+1)=[X best (t)-X M (t)]*α-rand+[(U B -L B )*rand+L B ]*δ (31)

[0148] Where α represents the development adjustment parameter, and its fixed value is 0.1.

[0149] (4) Narrowing the scope of development

[0150] The eagle gradually approaches its prey and launches a random attack pattern, then walks and captures the prey, such as Figure 6 As shown, the mathematical formula is as follows:

[0151] X4(t+1)=QF*X best (t)-(G1*X(t)*rand)-G2*Levy(D)+rand*G1 (32)

[0152] Where QF represents the average search strategy value of the quality function:

[0153]

[0154] G1 represents the various movement trajectories of the eagle during the escape of the prey, and the value is a random number between [-1,1]:

[0155] G1=2*rand-1 (34)

[0156] G2 is a decreasing value ranging from 2 to 0, which describes the change in the following slope from the starting position to the end position when the eagle is chasing its prey:

[0157]

[0158] The synergy of these strategies enables a single Skyhawk to conduct a more comprehensive search, thus having better local and global optimization capabilities. Therefore, using the AO algorithm to optimize the weights and biases in the ELM model can improve prediction accuracy and reduce the risk of overfitting.

[0159] Step six, dosage prediction, through the AO-ELM model, input the actual collected high-density pool sewage treatment data to predict the dosage.

[0160] The AO-ELM model is used to predict the resampled data of different particle sizes, and the fitness curve of the AO algorithm is plotted, and the predicted value of the dosage is printed.

[0161] The fitness curve of the AO algorithm can intuitively show the performance changes of the AO algorithm in the optimization process, proving the effectiveness of the proposed method and the rationality of the optimization process. At the same time, it shows how the solution is gradually improved as the number of iterations increases until it reaches the optimal state or meets the stopping condition.

[0162] Analyze the prediction results at different time granularities, that is, evaluate the model from two aspects: the fitness curve of the AO algorithm and the comparison between the original value and the predicted value. The specific steps include:

[0163] Reference Figure 7 ,In terms of the fitness curve of the AO algorithm, the AO algorithm at each granularity has a rapid convergence trend at the beginning, and the number of iterations at each granularity is distributed between 15 and 65 times, among which the 10-minute granularity converges fastest and the 30-minute granularity converges slowly.

[0164] Reference Figure 8-11, which are the comparisons of the original values ​​and predicted values ​​at 1 minute, 10 minutes, 20 minutes and 30 minutes granularity. Since the number of data samples at each granularity is inconsistent after mean resampling, when drawing the comparison chart between the predicted value and the true value of the AO-ELM model at each granularity, it is necessary to divide the horizontal axis at each granularity by a factor to scale the predicted value to within 30 days. It can be found that the AO-ELM predicted values ​​at each granularity have well captured the large volatility and weak periodicity of the coagulation dosage. In this process, the AO-ELM algorithm shows relatively good stability and accuracy at each granularity, and its predicted results are highly consistent with the true values. Based on the knowledge mined, the predictive plan of coagulation dosing in high-density tanks for industrial wastewater treatment can be guided to realize application services such as resource management, water quality monitoring and sustainable water treatment.

Claims

1. A method for predicting coagulation and dosing in a high-density sewage treatment tank, characterized in that: The following steps are involved: Step 1: Obtain the sewage treatment data of the high-density pool, and aggregate various sewage treatment data into a data set containing multiple features; Step 2: Data cleaning: remove duplicates and missing items from the data set and process outliers; Step 3: data resampling: resampling and normalizing the processed data; Step 4: Data division: divide the data into training set and test set; Step 5: Model training. Establish an ELM prediction model, initialize the input weights from the input layer to the hidden layer and the hidden layer node bias of the ELM model, train the ELM model using the training set data, input the test set data into the trained ELM model for prediction, and use the population fitness function to evaluate the ELM model. If the population fitness value does not meet the design requirements, use the AO optimization algorithm to optimize the iterative input weights and hidden layer node biases. After each optimization is completed, the new input weights and hidden layer node biases are input into the ELM model, and the training set training, test set prediction, and population fitness function evaluation are repeated until the population fitness function evaluation meets the requirements. At this time, the model obtained is the AO-ELM model for prediction; Step six, dosage prediction, through the AO-ELM model, input the actual collected high-density pool sewage treatment data to predict the dosage.

2. A method for predicting coagulation and dosing in a high-density sewage treatment tank according to claim 1, characterized in that: The outlier processing is to find the lower limit, upper limit and median of the data set through the box plot method, and for each data point in the data set, if its value is less than the lower limit or greater than the upper limit, it is replaced by the median.

3. A method for predicting coagulation and dosing in a high-density sewage treatment tank according to claim 1, characterized in that: Before the step three, the types of sewage treatment data are also screened, and through feature selection, sewage treatment data corresponding to features with a high correlation with the prediction index are screened out as data for subsequent use.

4. A method for predicting coagulation and dosing in a high-density sewage treatment tank according to claim 1, characterized in that: The data resampling is to further downsample the data to 10 minutes, 20 minutes and 30 minutes granularity.

5. A method for predicting coagulation and dosing in a high-density sewage treatment tank according to claim 1, characterized in that: The data division is specifically to divide the data into multiple groups, one of which is a test set and the rest are training sets.

6. A method for predicting coagulation and dosing in a high-density sewage treatment tank according to claim 1, characterized in that: The ELM model is, For N training samples (x i ,y i ), (i=1,2,...,N), x i =[x1,x2,...,x n ] T ∈R n represents the i-th input sample, y i =[y1,y2,...,y m ] T ∈R m represents the expected output corresponding to the i-th sample, where n represents the number of neurons in the input layer of the model, m represents the number of neurons in the output layer, the network has L hidden layer nodes, and the input weight from the input layer to the hidden layer is ω i =[ω1,ω2,...,ω n ] T , the output weight from the hidden layer to the output layer is β = β i =[β1,β2,...,β m ] T , b i represents the bias of the i-th hidden layer node, and g(·) is the activation function; The population fitness function is: where predict i is the predicted value of the model on the test set; true i is the true value of the model on the test set; p is the number of test samples; i is each individual in the population.

7. A method for predicting coagulation and dosing in a high-density sewage treatment tank according to claim 1, characterized in that: The specific method of using the AO optimization algorithm to optimize the input weights from the input layer to the hidden layer and the hidden layer node bias in the ELM model is: Set the population size, maximum number of iterations, exploration and development parameters, and set the current number of iterations to 1; Initialize the population position of the AO algorithm; The input weight ω from the input layer to the hidden layer in the ELM model i and the hidden layer node bias b i As The location information of the Sky Eagle is used to iteratively update the population fitness through four optimization strategies, and obtain the new Sky Eagle's best individual location information and fitness; Until the preset maximum number of iterations is reached or the preset conditions are met, the input weight ω corresponding to the best individual position information of the Eagle is output i and the hidden layer node bias b i into the ELM model.

8. A method for predicting coagulation and dosing in a high-density sewage treatment tank according to claim 7, characterized in that: When t≤2 / 3T and rand≤0.5, use method 1 to update the position of the Eagle; When t≤2 / 3T and rand>0.5, use method 2 to update the position of the Eagle; When t>2 / 3T and rand≤0.5, use method 3 to update the position of the Eagle; When t>2 / 3T and rand>0.5, use method 4 to update the position of the Eagle; t is the current iteration number, T is the maximum iteration number, and rand represents a randomly generated number in the range [0,1]; Methods one to four are respectively expanding the search range, narrowing the search range, expanding the development range and narrowing the development range in the AO algorithm.

Images