Intelligent control system for sewage phosphorus removal treatment dosing

Abstract

The present application provides a sewage phosphorus removal treatment dosing intelligent control system, comprising: sewage inlet end, mixing pool, filter pool and sewage outlet end connected in turn; inlet monitoring module for monitoring the total phosphorus and flow of sewage in the sewage inlet end; outlet monitoring module for monitoring the total phosphorus and flow of sewage in the sewage outlet end; dosing control module for putting phosphorus removal agent in the mixing pool; host computer connected with inlet monitoring module, outlet monitoring module and dosing module respectively, for inputting the total phosphorus, flow, total phosphorus and flow into the pre-trained phosphorus removal agent dosage prediction model, determining the dosage of phosphorus removal agent, and controlling the dosing control module to put the phosphorus removal agent with the dosage of phosphorus removal agent. The sewage phosphorus removal treatment dosing intelligent control system of the present application does not need to determine the dosage of phosphorus removal agent manually, reduces the labor cost, and avoids the inaccurate determination of the dosage of phosphorus removal agent.

Description

Technical Field

[0001] This invention relates to the field of wastewater treatment technology, and in particular to an intelligent control system for phosphorus removal treatment in wastewater. Background Technology

[0002] Currently, wastewater treatment plants often mix phosphorus removal agents with phosphate-treated wastewater for phosphorus removal. When mixing these agents, the dosage needs to be determined. This is typically done manually by staff based on experience, considering the total phosphorus and flow rate of the influent phosphate-treated wastewater, as well as the total phosphorus and flow rate of the effluent after phosphorus removal. This method is labor-intensive. Furthermore, manual determination based on experience can lead to inaccuracies in the dosage. Adding too much agent increases the cost of phosphorus removal, while adding too little results in poor treatment effectiveness, failing to meet standards (according to GB18918-2002 "Discharge Standard of Pollutants for Urban Wastewater Treatment Plants," the total phosphorus (TP) in Class A effluent should be less than 0.5 mg / L. Some local standards require a TP of 0.3 mg / L for treated wastewater).

[0003] Therefore, a solution is urgently needed. Summary of the Invention

[0004] This invention provides an intelligent control system for phosphorus removal treatment in wastewater. It introduces a phosphorus removal agent dosage prediction model, which adaptively determines the phosphorus removal agent dosage based on the total phosphorus in the influent, influent flow rate, total phosphorus in the effluent, and effluent flow rate, and then dispenses the appropriate dosage of phosphorus removal agent to achieve automatic dosing control. This eliminates the need for manual determination of the phosphorus removal agent dosage, reduces labor costs, and avoids the possibility of inaccurate dosage determination due to manual experience.

[0005] This invention provides an intelligent control system for phosphorus removal treatment in wastewater, comprising:

[0006] The wastewater inlet, mixing tank, filtration tank, and wastewater outlet are connected in sequence.

[0007] The influent monitoring module is used to monitor the total phosphorus content and flow rate of the wastewater at the wastewater inlet.

[0008] The effluent monitoring module is used to monitor the total phosphorus and effluent flow rate of the wastewater at the wastewater outlet.

[0009] A dosing control module is used to add phosphorus removal agent into the mixing tank;

[0010] The host computer is connected to the influent monitoring module, the effluent monitoring module, and the dosing module, respectively. It is used to input the total phosphorus in the influent, the influent flow rate, the total phosphorus in the effluent, and the effluent flow rate into a pre-trained phosphorus removal agent dosage prediction model, determine the phosphorus removal agent dosage, and control the dosing control module to dispensing the phosphorus removal agent according to the specified dosage.

[0011] Preferably, the training steps of the phosphorus removal agent dosage prediction model include:

[0012] Obtain multiple sets of historical dosing records, including: historical influent total phosphorus, historical influent flow rate, historical effluent total phosphorus, historical effluent flow rate, and historical phosphorus removal agent dosage;

[0013] Multiple sets of historical pesticide application records were used as training data and input into the CNN-TCN-Attention-GPR model to obtain a phosphorus removal agent dosage prediction model.

[0014] Preferably, the step of inputting multiple sets of historical medication records as training data into the CNN-TCN-Attention-GPR model for training is as follows:

[0015] As the input layer of the first layer: the training data is used as the input independent variable to enter the next layer; the input independent variable X is represented as:

[0016] X = (X1, X2, ..., X n )

[0017] Among them, X n This is the nth input independent variable;

[0018] As the second convolutional layer: it normalizes the input variables and uses convolution operations to generate a feature matrix based on the normalization result.

[0019]

[0020] The `conv` function is the convolution operation in a CNN network. and These are the input and output during the convolution process, where l is the sequence length of the normalized result, and i and j represent the processing positions during the convolution process. These are the weights of the convolutional layer. It is the bias of the convolutional layer;

[0021]

[0022] in, and These are the input and output of the activation function, respectively, where f() is the ReLU non-linear activation function;

[0023] As the third pooling layer: it performs pooling operations on the feature matrix.

[0024]

[0025] The pooling() function represents the pooling operation. This represents the output after the pooling layer;

[0026] The fourth layer, the TCN layer, performs a convolution operation on the input variable X.

[0027]

[0028] F(i) represents the pair (X1,X2,…,X) n The convolution result of the i-th element in ), where h(j) is the filter / convolution kernel, d is the scaling factor, and k is a constant;

[0029] The TCN layer also includes residual blocks for performing residual operations on the convolution result F(i):

[0030] o = Activation(i + F(i))

[0031] Where o is the result of the residual operation, and Activation() is the activation function;

[0032] As the fifth attention layer, it performs attention operations on the data processed by the TCN layer.

[0033] e t =tanh(w t s t +b t )

[0034]

[0035]

[0036] e t The state vector s represents the t-th feature vector of the data processed by the TCN layer. t The required energy value, w t b represents the weight coefficient matrix of the t-th feature vector of the data processed by the TCN layer. t This represents the offset corresponding to the t-th feature vector of the data processed by the TCN layer, where tanh() is the activation function and α is the offset. t Y is the result of the attention operation, and exp is an exponential function with the natural constant e as the base.

[0037] As the sixth layer, the Gaussian process regression model is a nonparametric model that uses Gaussian priors to perform regression analysis on the output data of the attention layer. The model is established as follows:

[0038] y = f(x) + ε

[0039] In the formula: f(x) is the Gaussian process function, y is the observed value, y is the input vector of the attention operation result, and noise is...

[0040] Therefore, the prior distribution of the observed value y is obtained:

[0041]

[0042] The joint prior distribution of the observed and predicted values ​​is as follows:

[0043]

[0044] In the formula: K(X, X) is an n-order positive definite covariance matrix, K(X, X) * K(X) is the n*1 covariance matrix between the test points and the test set; * X * I represents the covariance of the test point itself. n It is an n-dimensional identity matrix;

[0045] The post-validation distribution of the predicted values ​​is calculated as follows:

[0046]

[0047] and

[0048]

[0049]

[0050] In the formula: and cov(f * ) represents the mean and variance of the predicted values ​​corresponding to the test points;

[0051] As the output layer of the seventh layer:

[0052] The prediction results of the training set obtained by training the CNN-TCN-Attention model are denoted as train_X1, and the prediction results of the test set are denoted as test_X1. train_X1 and test_X1 are used as the training set and test set input data of the GPR layer.

[0053] Input test_X1 into the trained GPR model and the final prediction output is Final_predict = GPR.predict(test_X1).

[0054] Preferably, the acquisition of multiple sets of historical dosing records includes:

[0055] Multiple sets of historical dosing records are obtained through preset local acquisition nodes;

[0056] And / or,

[0057] Multiple sets of historical drug dosing records are obtained through multiple preset non-local acquisition nodes.

[0058] Preferably, multiple sets of historical dosing records are obtained through multiple preset non-local acquisition nodes, including:

[0059] Obtain the pre-set equipment information for wastewater phosphorus removal dosing equipment;

[0060] Based on the aforementioned equipment information and preset equipment requirements, a template library is created to formulate appropriate equipment requirements.

[0061] Obtain the node type of the non-locally acquired node, wherein the node type includes: direct node and indirect node;

[0062] When the node type of the non-local acquisition node is a direct node, the first credibility of the node identity of the non-local acquisition node is verified. If the verification is successful, multiple sets of historical dosing records of other wastewater phosphorus removal dosing equipment that meet the requirements of the equipment are obtained through the corresponding non-local acquisition node.

[0063] When the node type of the non-local acquisition node is an indirect node, the second credibility of the node reputation of the non-local acquisition node is verified. If the verification is successful, multiple sets of historical dosing records of other wastewater phosphorus removal dosing equipment that meet the requirements of the equipment are obtained through the corresponding non-local acquisition node.

[0064] Preferably, a template library is developed based on the equipment information and preset equipment requirements to formulate appropriate equipment requirements, including:

[0065] The device information is categorized to obtain first classification information for multiple first information categories;

[0066] Obtain a preset device condition generation template corresponding to the first information category, and generate device conditions based on the condition generation template and the first classification information corresponding to the first information category;

[0067] Obtain a preset equipment condition matching template that corresponds to each of the first information categories, and match each of the equipment conditions based on the equipment condition matching template to obtain suitable equipment requirements.

[0068] Preferably, the first credibility verification of the node identity of the non-locally acquired node includes:

[0069] Obtain the node identity information of the non-local acquisition node;

[0070] The second information category distribution of the node identity information is parsed, and the second information category distribution includes: multiple sets of one-to-one corresponding second information categories and information quantities;

[0071] Based on a preset first feature extraction template, features are extracted from the second information category distribution to obtain multiple distribution feature values, and a first feature expression vector is constructed based on the distribution feature values.

[0072] The first feature expression vector is matched with any second feature expression vector in the preset confidence determination library. If the match is found, the preset first confidence level corresponding to the matched second feature expression vector is obtained.

[0073] Obtain a preset identity trust verification template corresponding to each of the second information categories, and based on the identity trust verification template, perform identity trust verification on the second category information corresponding to the second information category in the node identity information to obtain a verification value;

[0074] The second confidence level is obtained by summing up each of the aforementioned verification values.

[0075] Based on the first confidence level and the second confidence level, the first confidence level of the non-local acquisition node's identity is calculated using the following formula:

[0076] P = J1·K1 + J2·K2

[0077] Where P is the first confidence level, J1 and J2 are preset weight values, K1 is the first confidence level, and K2 is the second confidence level;

[0078] If the first credibility is greater than or equal to the preset first credibility threshold, the first credibility of the node identity of the non-local acquisition node is verified.

[0079] Preferably, the second credibility verification of the node reputation of the non-locally acquired node includes:

[0080] The feedback information is used to retrieve multiple pieces of information that were obtained historically through the non-local acquisition node.

[0081] Based on the preset second feature extraction template, the information is used to extract features using response information to obtain multiple response information features;

[0082] The response information feature is matched with any response trigger feature in the preset response trigger feature library. If the match is found, the response source type corresponding to the information is obtained. The response source type includes: active response and passive response. When the response source type corresponding to the information is passive response, the matching response trigger feature is taken as the target response trigger feature, and the target response trigger feature is recorded as being triggered once.

[0083] The total number of times each target response trigger feature is triggered is counted, and the preset total number of triggers - third confidence level lookup table corresponding to the target response trigger feature is queried to determine the third confidence level corresponding to the total number of triggers.

[0084] Each of the third confidence scores is accumulated to obtain the second confidence score of the node reputation of the non-local acquisition node;

[0085] If the second credibility is greater than or equal to the preset second credibility threshold, the second credibility of the node reputation of the non-local acquisition node is verified.

[0086] Preferably, the host computer is further used for:

[0087] When a user inputs a request for manual intervention in the dosing control, a template is generated based on the preset dosing amount determination interface, and the dosing amount determination interface is generated according to the total phosphorus in the influent, the influent flow rate, the total phosphorus in the effluent, and the effluent flow rate.

[0088] The dosage determination interface is sent to the user's corresponding preset terminal node;

[0089] The terminal node returns the user-inputted amount of phosphorus removal agent, and the dosing control module is controlled to dispense the phosphorus removal agent according to the user-inputted amount.

[0090] Preferably, the host computer is further used for:

[0091] Obtain operational information from the influent monitoring module, effluent monitoring module, and chemical dosing module;

[0092] Based on the operational information, operational anomalies are identified and warnings are issued.

[0093] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings.

[0094] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0095] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0096] Figure 1 This is a flowchart of an intelligent control system for phosphorus removal treatment in wastewater according to an embodiment of the present invention;

[0097] Figure 2 This is a structural diagram of the CNN-TCN-Attention-GPR module in an embodiment of the present invention;

[0098] Figure 3 This is a comparison chart of the predicted and actual values ​​of the phosphorus removal agent dosage during the phosphorus removal agent dosage experiment in this embodiment of the invention. Detailed Implementation

[0099] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0100] This invention provides an intelligent control system for phosphorus removal treatment in wastewater, such as... Figure 1 As shown, it includes:

[0101] The wastewater inlet, mixing tank, filtration tank, and wastewater outlet are connected in sequence.

[0102] The influent monitoring module is used to monitor the total phosphorus content and flow rate of the wastewater at the wastewater inlet.

[0103] The effluent monitoring module is used to monitor the total phosphorus and effluent flow rate of the wastewater at the wastewater outlet.

[0104] A dosing control module is used to add phosphorus removal agent into the mixing tank;

[0105] The host computer is connected to the influent monitoring module, the effluent monitoring module, and the dosing module, respectively. It is used to input the total phosphorus in the influent, the influent flow rate, the total phosphorus in the effluent, and the effluent flow rate into a pre-trained phosphorus removal agent dosage prediction model, determine the phosphorus removal agent dosage, and control the dosing control module to dispensing the phosphorus removal agent according to the specified dosage.

[0106] The working principle and beneficial effects of the above technical solution are as follows:

[0107] Phosphating wastewater enters through the wastewater inlet, flows through a mixing tank where it mixes with a phosphorus removal agent to achieve phosphorus removal, then flows through a filtration tank for secondary filtration, and finally flows out through the wastewater outlet. The filtration equipment in the filtration tank can be configured according to filtration requirements. The influent monitoring module can be a flow meter and a phosphorus content sensor to monitor the total phosphorus and influent flow rate in real time at the wastewater inlet. Similarly, the effluent monitoring module can also be a flow meter and a phosphorus content sensor. The dosing control module can be a processor, a dosing pump, and a flow meter connected in sequence. The processor obtains the required phosphorus removal agent dosage and controls the dosing pump based on the dosage until the flow meter detects that the dosage has been reached and stops the pump. The host computer inputs the influent total phosphorus, influent flow rate, effluent total phosphorus, and effluent flow rate into a pre-trained phosphorus removal agent dosage prediction model to determine the phosphorus removal agent dosage and control the dosing control module to administer the prescribed dosage.

[0108] This application introduces a phosphorus removal agent dosage prediction model, which adaptively determines the phosphorus removal agent dosage based on the total phosphorus in the influent, influent flow rate, total phosphorus in the effluent, and effluent flow rate, and then administers the corresponding dosage of phosphorus removal agent to achieve automatic dosing control. This eliminates the need for manual determination of the phosphorus removal agent dosage, reduces labor costs, and avoids the possibility of inaccurate phosphorus removal agent dosage determination due to manual experience.

[0109] This invention provides an intelligent control system for phosphorus removal treatment in wastewater, wherein the training steps of the phosphorus removal agent dosage prediction model include:

[0110] Obtain multiple sets of historical dosing records, including: historical influent total phosphorus, historical influent flow rate, historical effluent total phosphorus, historical effluent flow rate, and historical phosphorus removal agent dosage;

[0111] Multiple sets of historical pesticide application records were used as training data and input into the CNN-TCN-Attention-GPR model to obtain a phosphorus removal agent dosage prediction model.

[0112] The working principle and beneficial effects of the above technical solution are as follows:

[0113] Historical dosing records are input into a CNN-TCN-Attention-GPR model, which is a convolutional neural network model, for training. The convolutional neural network model can learn from historical dosing records and generate an artificial intelligence model that can replace manual dosing prediction.

[0114] This invention provides an intelligent control system for phosphorus removal treatment in wastewater, such as... Figure 2As shown, multiple sets of historical medication records are input as training data into the CNN-TCN-Attention-GPR model for training:

[0115] As the input layer of the first layer: the training data is used as the input independent variable to enter the next layer; the input independent variable X is represented as:

[0116] X = (X1, X2, ..., X n )

[0117] Among them, X n This is the nth input independent variable;

[0118] As the second convolutional layer: it normalizes the input variables and uses convolution operations to generate a feature matrix based on the normalization result.

[0119]

[0120] The `conv` function is the convolution operation in a CNN network. and These are the input and output during the convolution process, where l is the sequence length of the normalized result, and i and j represent the processing positions during the convolution process. These are the weights of the convolutional layer. It is the bias of the convolutional layer;

[0121]

[0122] in, and These are the input and output of the activation function, respectively, where f() is the ReLU non-linear activation function;

[0123] As the third pooling layer: it performs pooling operations on the feature matrix.

[0124]

[0125] The pooling() function represents the pooling operation. This represents the output after the pooling layer;

[0126] The fourth layer, the TCN layer, performs a convolution operation on the input variable X.

[0127]

[0128] F(i) represents the pair (X1,X2,…,X) n The convolution result of the i-th element in ), where h(j) is the filter / convolution kernel, d is the scaling factor, and k is a constant;

[0129] The TCN layer also includes residual blocks for performing residual operations on the convolution result F(i):

[0130] o = Activation(i + F(i))

[0131] Where o is the result of the residual operation, and Activation() is the activation function;

[0132] As the fifth attention layer, it performs attention operations on the data processed by the TCN layer.

[0133] e t =tanh(w t s t +b t )

[0134]

[0135]

[0136] e t The state vector s represents the t-th feature vector of the data processed by the TCN layer. t The required energy value, w t b represents the weight coefficient matrix of the t-th feature vector of the data processed by the TCN layer. t This represents the offset corresponding to the t-th feature vector of the data processed by the TCN layer, where tanh() is the activation function and α is the offset. t Y is the result of the attention operation, and exp is an exponential function with the natural constant e as the base.

[0137] As the sixth layer, the Gaussian process regression model is a nonparametric model that uses Gaussian priors to perform regression analysis on the output data of the attention layer. The model is established as follows:

[0138] y = f(x) + ε

[0139] In the formula: f(x) is the Gaussian process function, y is the observed value, y is the input vector of the attention operation result, and noise is...

[0140] Therefore, the prior distribution of the observed value y is obtained:

[0141]

[0142] The joint prior distribution of the observed and predicted values ​​is as follows:

[0143]

[0144] In the formula: K(X, X) is an n-order positive definite covariance matrix, K(X, X) * K(X) is the n*1 covariance matrix between the test points and the test set; * X * I represents the covariance of the test point itself. n It is an n-dimensional identity matrix;

[0145] The post-validation distribution of the predicted values ​​is calculated as follows:

[0146]

[0147] and

[0148]

[0149]

[0150] In the formula: and cov(f * ) represents the mean and variance of the predicted values ​​corresponding to the test points;

[0151] As the output layer of the seventh layer:

[0152] The prediction results of the training set obtained by training the CNN-TCN-Attention model are denoted as train_X1, and the prediction results of the test set are denoted as test_X1. train_X1 and test_X1 are used as the training set and test set input data of the GPR layer.

[0153] Input test_X1 into the trained GPR model and the final prediction output is Final_predict = GPR.predict(test_X1).

[0154] The working principle and beneficial effects of the above technical solution are as follows:

[0155] As the input layer for the first layer: 800 sets of historical data on influent flow rate, total effluent phosphorus, and effluent flow rate are used as input variables for the next layer; the input variable X is represented as:

[0156] X = (X1, X2, X3)

[0157] Wherein, X1, X2 and X3 are the influent flow rate, total phosphorus in the effluent and the effluent flow rate, respectively;

[0158] As the second convolutional layer: it normalizes the input variables and uses convolution operations to generate a feature matrix based on the normalization result.

[0159]

[0160] The `conv` function is the convolution operation in a CNN network. and These are the input and output during the convolution process, where l is the sequence length of the normalized result, and i and j represent the processing positions during the convolution process. These are the weights of the convolutional layer. It is the bias of the convolutional layer;

[0161]

[0162] in, and These are the input and output of the activation function, respectively, where f() is the ReLU non-linear activation function;

[0163] As the third pooling layer: it performs pooling operations on the feature matrix.

[0164]

[0165] The pooling() function represents the pooling operation. This represents the output after the pooling layer;

[0166] The fourth layer, TCN, is a novel architecture based on Convolutional Neural Networks (CNNs). Unlike typical CNNs, TCN employs structures such as expanded causal convolutions and residual blocks, enabling it to extract features from large time series samples and achieve predictions. Furthermore, it effectively addresses the performance degradation problem of deep networks during training.

[0167] TCN employs Dilated Causal Convolution (DCC) to increase the receptive field without significantly increasing computational cost. DCC is a convolution operation that performs a step operation on the input sequence, and its expression is:

[0168]

[0169] In the formula, F(i) represents the pair of sequences. The convolution result of the i-th element; h(j) is the filter (also called the convolution kernel, the kernel size of which is K = 1*k for a one-dimensional sequence); d is the scaling factor (when d = 1, it is the standard causal convolution).

[0170] Residual blocks were primarily proposed to address the degradation problem in deep learning networks. Their core idea is to introduce a "skip connection" operation that skips one or more layers. Assuming x is the input to the residual block, the output o of the residual block is as follows:

[0171] o = Activation(x + F(x))

[0172] In the formula, Activation() is the activation function. Since the residual F(x) will not be zero in practice, the stacked layers in a deep learning network can always learn new features, so the learning performance of deep networks will not degrade.

[0173] When modeling TCN, using a network structure that combines residual blocks and DCC can effectively improve the feature learning ability and robustness of the TCN model.

[0174] The fifth layer, the attention layer, generates weight coefficients for different signal segments based on different target values ​​(here, the data processed by the TCN layer). It then combines information from different signal segments in the form of a weighted average, giving greater weight to important parameters and thus improving the generalization performance of the regression prediction model. The output of the attention layer is as follows:

[0175] e t =tanh(w t s t +b t )

[0176]

[0177]

[0178] In the formula: e t The state vector s represents the t-th eigenvector. t The required energy value; w t b represents the weight coefficient matrix of the t-th eigenvector; t This represents the offset corresponding to the t-th feature vector. According to the second formula, it can be used to transform from an initial input state to a new attention state, and then obtain the final output state vector Y. The training set prediction results obtained by training the CNN-TCN-Attention model are denoted as train_X1, and the test set prediction results are denoted as test_X1. train_X1 and test_X1 serve as the training and test set input data for the GPR layer.

[0179] The sixth layer, the GPR layer, is a nonparametric model that uses Gaussian priors to perform regression analysis on the output data of the attention layer. The model is established as follows:

[0180] y = f(x) + ε

[0181] In the formula: f(x) is the Gaussian process function; y is the observed value; x is the input vector; noise

[0182] Therefore, the prior distribution of the observed value y is obtained:

[0183]

[0184] The joint prior distribution of the observed and predicted values ​​was also obtained as follows:

[0185]

[0186] In the formula: K(X,X) is an n-order positive definite covariance matrix; K(X,X) * K(X) is the n*1 covariance matrix between the test points and the test set; * ,X * ) represents the covariance of the test point itself; I n It is an n-dimensional identity matrix.

[0187] The post-validation distribution of the predicted values ​​is calculated as follows:

[0188]

[0189] and

[0190]

[0191]

[0192] In the formula: and cov(f * ) represents the mean and variance of the predicted values ​​corresponding to the test points.

[0193] As the output layer of the seventh layer: it takes test_X1 as input to the trained GPR model and outputs the final prediction.

[0194] Final_predict=GPR.predict(test_X1).

[0195] This invention provides an intelligent control system for phosphorus removal treatment in wastewater, wherein acquiring multiple sets of historical dosing records includes:

[0196] Multiple sets of historical dosing records are obtained through preset local acquisition nodes;

[0197] And / or,

[0198] Multiple sets of historical drug dosing records are obtained through multiple preset non-local acquisition nodes.

[0199] The working principle and beneficial effects of the above technical solution are as follows:

[0200] There are two methods for obtaining multiple sets of historical dosing records: First, multiple sets of historical dosing records are obtained from the wastewater phosphorus removal dosing equipment itself locally. A preset local acquisition node communicates and interfaces with the database storing these historical dosing records. Second, multiple sets of historical dosing records are obtained from other wastewater phosphorus removal dosing equipment outside the local area. A preset non-local acquisition node communicates and interfaces with the storage source containing these historical dosing records. Introducing these two methods improves the comprehensiveness of the training samples obtained for training the phosphorus removal agent dosage prediction model, thereby enhancing the training quality and operational capability of the model.

[0201] This invention provides an intelligent control system for phosphorus removal treatment in wastewater, which acquires multiple sets of historical dosing records through preset non-local acquisition nodes, including:

[0202] Obtain the pre-set equipment information for wastewater phosphorus removal dosing equipment;

[0203] Based on the aforementioned equipment information and preset equipment requirements, a template library is created to formulate appropriate equipment requirements.

[0204] Obtain the node type of the non-locally acquired node, wherein the node type includes: direct node and indirect node;

[0205] When the node type of the non-local acquisition node is a direct node, the first credibility of the node identity of the non-local acquisition node is verified. If the verification is successful, multiple sets of historical dosing records of other wastewater phosphorus removal dosing equipment that meet the requirements of the equipment are obtained through the corresponding non-local acquisition node.

[0206] When the node type of the non-local acquisition node is an indirect node, the second credibility of the node reputation of the non-local acquisition node is verified. If the verification is successful, multiple sets of historical dosing records of other wastewater phosphorus removal dosing equipment that meet the requirements of the equipment are obtained through the corresponding non-local acquisition node.

[0207] The working principle and beneficial effects of the above technical solution are as follows:

[0208] However, when acquiring multiple sets of historical dosing records from other wastewater phosphorus removal treatment dosing equipment outside the local area, the data quality cannot be controlled due to the non-local information source. If these records are directly acquired and used as training samples to train the phosphorus removal agent dosage prediction model, the training quality of the phosphorus removal agent dosage prediction model will be reduced. Therefore, it is necessary to control the quality of multiple sets of historical dosing records from other wastewater phosphorus removal treatment dosing equipment acquired outside the local area.

[0209] First, because other wastewater phosphorus removal dosing equipment differs from the wastewater treatment plant's own equipment in terms of phosphorus removal agent type, inlet pipe size, outlet pipe size, and mixing tank size, the methods for determining phosphorus removal agent dosage have historically varied. Therefore, other wastewater phosphorus removal dosing equipment should be consistent with the wastewater treatment plant's own equipment in terms of hardware. Thus, pre-defined equipment information for wastewater phosphorus removal dosing equipment is obtained. This pre-defined equipment information includes the wastewater treatment plant's own equipment information, such as the phosphorus removal agent type, inlet pipe size, outlet pipe size, and mixing tank size. Based on this equipment information and pre-defined equipment requirements, a template library is developed, and appropriate equipment requirements are established. Other wastewater phosphorus removal dosing equipment should meet these requirements. This ensures the applicability of historical dosing records from other wastewater phosphorus removal dosing equipment to improve the comprehensiveness of training samples for the phosphorus removal agent dosage prediction model, thereby improving the accuracy and efficiency of historical dosing record acquisition from non-local nodes.

[0210] Secondly, non-local acquisition nodes are categorized into two types: First, direct nodes, where the information is directly derived from the node itself, such as other wastewater treatment plants equipped with phosphorus removal dosing equipment. Second, indirect nodes, which collect information before providing it, such as big data platforms that collect confirmed records of phosphorus removal agent dosage from wastewater treatment plants equipped with phosphorus removal dosing equipment. When a node is a direct node, its primary credibility is verified, for example, whether it is a wastewater treatment plant. If verified, the data is acquired. When a node is an indirect node, its secondary credibility is verified, for example, whether it has historically provided reliable data. If verified, the data is acquired. By verifying the credibility of nodes based on their type, and only acquiring information after successful verification, the accuracy and quality of historical dosing record acquisition are further improved.

[0211] This invention provides an intelligent control system for phosphorus removal treatment in wastewater. Based on the equipment information and preset equipment requirements, a template library is created to formulate suitable equipment requirements, including:

[0212] The device information is categorized to obtain first classification information for multiple first information categories;

[0213] Obtain a preset device condition generation template corresponding to the first information category, and generate device conditions based on the condition generation template and the first classification information corresponding to the first information category;

[0214] Obtain a preset equipment condition matching template that corresponds to each of the first information categories, and match each of the equipment conditions based on the equipment condition matching template to obtain suitable equipment requirements.

[0215] The working principle and beneficial effects of the above technical solution are as follows:

[0216] The first information category includes: the type of phosphorus removal agent used, the size of the inlet pipe, the size of the outlet pipe, and the size of the mixing tank. The preset equipment condition generation template corresponding to the first information category is used to generate equipment conditions based on the first category's first classification information. For example, if the first classification information of the first information category is "the type of phosphorus removal agent used is ZX-822," then the generated equipment condition would be "the type of phosphorus removal agent used in other wastewater phosphorus removal treatment dosing equipment should be ZX-822." Another example: if the first classification information of the first information category is "the inlet pipe size is a radius of 0.5 meters," then the generated equipment condition would be "the inlet pipe size used in other wastewater phosphorus removal treatment dosing equipment should be at least 0.5 meters." Based on the condition generation template, equipment conditions are generated according to the first classification information of the corresponding first information category. The preset equipment condition matching templates corresponding to the first information category are templates for matching equipment conditions under each first information category. For example, if the first information category is the type of phosphorus removal agent and the size of the inlet pipe, then the matching equipment requirements would be "The type of phosphorus removal agent used in other wastewater phosphorus removal treatment dosing equipment should be ZX-822, and the size of the inlet pipe used in other wastewater phosphorus removal treatment dosing equipment should be at least 0.5 meters." Based on the equipment condition matching templates, various equipment conditions are matched to obtain suitable equipment requirements. The introduction of equipment condition generation templates and equipment condition matching templates improves the efficiency and accuracy of formulating suitable equipment requirements.

[0217] This invention provides an intelligent control system for phosphorus removal treatment in wastewater, which verifies the first credibility of the node identity of the non-locally acquired node, including:

[0218] Obtain the node identity information of the non-local acquisition node;

[0219] The second information category distribution of the node identity information is parsed, and the second information category distribution includes: multiple sets of one-to-one corresponding second information categories and information quantities;

[0220] Based on a preset first feature extraction template, features are extracted from the second information category distribution to obtain multiple distribution feature values, and a first feature expression vector is constructed based on the distribution feature values.

[0221] The first feature expression vector is matched with any second feature expression vector in the preset confidence determination library. If the match is found, the preset first confidence level corresponding to the matched second feature expression vector is obtained.

[0222] Obtain a preset identity trust verification template corresponding to each of the second information categories, and based on the identity trust verification template, perform identity trust verification on the second category information corresponding to the second information category in the node identity information to obtain a verification value;

[0223] The second confidence level is obtained by summing up each of the aforementioned verification values.

[0224] Based on the first confidence level and the second confidence level, the first confidence level of the non-local acquisition node's identity is calculated using the following formula:

[0225] P = J1·K1 + J2·K2

[0226] Where P is the first confidence level, J1 and J2 are preset weight values, K1 is the first confidence level, and K2 is the second confidence level;

[0227] If the first credibility is greater than or equal to the preset first credibility threshold, the first credibility of the node identity of the non-local acquisition node is verified.

[0228] The working principle and beneficial effects of the above technical solution are as follows:

[0229] The node identity information includes: information that can prove the identity of other wastewater treatment plants equipped with wastewater phosphorus removal and dosing equipment, such as: business license information, legal representative electronic signature information, wastewater phosphorus removal site photos, and legal representative electronic seal information. The second information category includes: information categories that can prove the identity of other wastewater treatment plants equipped with wastewater phosphorus removal and dosing equipment, such as: business license information, legal representative electronic signature information, wastewater phosphorus removal site photos, and legal representative electronic seal information. Distribution feature values ​​include: the information category with the largest information content, the information category with the smallest information content, and the average information content of each information category. A first feature expression vector is constructed in vector form based on the distribution feature values. The second feature expression vector is pre-constructed using distribution feature values ​​reflecting the credibility of different enterprise identities. For example, if the average information content of each information category is small and the information category with the largest information content is the legal representative electronic signature (easily forged), then the credibility of the identity is low, and the corresponding preset first confidence level is low. The preset identity verification template corresponding to the second information category is used to verify the identity credibility of the second category of information. For example, if the second information category is business license information, then the verification will check whether the business scope includes sewage treatment, etc. The higher the verification value, the more credible the identity. The verification values ​​are accumulated to obtain the second confidence level. Based on the first confidence level and the second confidence level, the first credibility level is calculated. If the first credibility level is greater than or equal to the preset first credibility level threshold, the verification passes. The first credibility level of non-locally acquired nodes is determined from two dimensions: the distribution of the second information category of node identity information and the content of node identity information. This improves the accuracy of verifying the first credibility level of non-locally acquired nodes. In addition, the introduction of a preset confidence level determination library and identity verification template improves the accuracy and efficiency of verifying the first credibility level of non-locally acquired nodes.

[0230] This invention provides an intelligent control system for phosphorus removal treatment in wastewater, which verifies the second credibility of the node reputation of the non-locally acquired node, including:

[0231] The feedback information is used to retrieve multiple pieces of information that were obtained historically through the non-local acquisition node.

[0232] Based on the preset second feature extraction template, the information is used to extract features using response information to obtain multiple response information features;

[0233] The response information feature is matched with any response trigger feature in the preset response trigger feature library. If the match is found, the response source type corresponding to the information is obtained. The response source type includes: active response and passive response. When the response source type corresponding to the information is passive response, the matching response trigger feature is taken as the target response trigger feature, and the target response trigger feature is recorded as being triggered once.

[0234] The total number of times each target response trigger feature is triggered is counted, and the preset total number of triggers - third confidence level lookup table corresponding to the target response trigger feature is queried to determine the third confidence level corresponding to the total number of triggers.

[0235] Each of the third confidence scores is accumulated to obtain the second confidence score of the node reputation of the non-local acquisition node;

[0236] If the second credibility is greater than or equal to the preset second credibility threshold, the second credibility of the node reputation of the non-local acquisition node is verified.

[0237] The working principle and beneficial effects of the above technical solution are as follows:

[0238] Information usage feedback includes historical evaluations of information obtained through non-local acquisition nodes, such as "the dosage determination is reasonable" and "the dosage determination is obviously unreasonable." Information usage feedback can be obtained locally or from non-acquiring nodes. Feedback features are extracted from the information usage feedback, which are semantic features of the feedback evaluation. Feedback triggering features are semantic features of poor feedback, such as "the dosage determination is unreasonable" and "the dosage determination is incorrect." Feedback features are matched with feedback triggering features. If a match is found, it indicates a poor feedback evaluation, and the corresponding feedback source type is determined. There are two source types: First, active feedback, where the non-local acquisition node actively reflects the feedback. Second, passive feedback, where the information acquirer discovers the feedback after acquisition. When the feedback source type is passive feedback, the matching feedback triggering feature is used as the target feedback triggering feature, and the target feedback triggering feature is recorded as triggered once. A table showing the preset total number of triggers and third confidence level corresponding to different total number of triggers for the target response trigger feature contains a third confidence level. A negative third confidence level indicates that the more the target response trigger feature is triggered, the more semantic features with poor responses there are. The table is consulted to determine the third confidence level corresponding to the total number of triggers, and these are then summed to obtain the second confidence level of the node reputation of non-locally acquired nodes. If the second confidence level is greater than or equal to a preset second confidence level threshold, the verification is successful. Determining the second confidence level of the node reputation of non-locally acquired nodes based on node response information improves the accuracy of second confidence level determination. Furthermore, the introduction of response trigger features improves the accuracy and efficiency of identifying poor responses.

[0239] This invention provides an intelligent control system for phosphorus removal treatment in wastewater, wherein the host computer is further configured to include:

[0240] When a user inputs a request for manual intervention in the dosing control, a template is generated based on the preset dosing amount determination interface, and the dosing amount determination interface is generated according to the total phosphorus in the influent, the influent flow rate, the total phosphorus in the effluent, and the effluent flow rate.

[0241] The dosage determination interface is sent to the user's corresponding preset terminal node;

[0242] The terminal node returns the user-inputted amount of phosphorus removal agent, and the dosing control module is controlled to dispense the phosphorus removal agent according to the user-inputted amount.

[0243] The working principle and beneficial effects of the above technical solution are as follows:

[0244] The system also provides a function for users to manually determine the dosage of phosphorus removal agent. The preset dosing determination interface template is generated based on the influent total phosphorus, influent flow rate, effluent total phosphorus, and effluent flow rate. For example: "Currently, the influent total phosphorus, influent flow rate, effluent total phosphorus, and effluent flow rate are xxxx. Please enter the phosphorus removal agent dosage." The user-defined terminal node communicates with the user's smart terminal, which can be a mobile phone, tablet, or computer. After the user enters the phosphorus removal agent dosage, the terminal node returns a response, controlling the dosing control module to administer the phosphorus removal agent at the user-input dosage. This user-defined manual phosphorus removal agent dosage determination is more user-friendly and particularly suitable for situations where the phosphorus removal agent dosage prediction model malfunctions, allowing for manual intervention and improving the stability of wastewater phosphorus removal treatment.

[0245] This invention provides an intelligent control system for phosphorus removal treatment in wastewater, wherein the host computer is further configured to include:

[0246] Obtain operational information from the influent monitoring module, effluent monitoring module, and chemical dosing module;

[0247] Based on the operational information, operational anomalies are identified and warnings are issued.

[0248] The working principle and beneficial effects of the above technical solution are as follows:

[0249] The system can also monitor the operation of the inlet water monitoring module, the outlet water monitoring module, and the chemical dosing module for abnormalities, and output an early warning when an abnormality is detected.

[0250] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A sewage phosphorus removal treatment dosing intelligent control system, characterized in that, The sewage treatment device comprises: a sewage inlet end, a mixing pool, a filter pool and a sewage outlet end connected in sequence; an inlet monitoring module for monitoring the total phosphorus and flow rate of sewage in the sewage inlet end; an outlet monitoring module for monitoring the total phosphorus and flow rate of sewage in the sewage outlet end; a dosing control module for adding a phosphorus removal agent in the mixing pool; a host computer connected with the inlet monitoring module, the outlet monitoring module and the dosing module, for inputting the total phosphorus, flow rate, total phosphorus and flow rate of sewage into a pre-trained phosphorus removal agent dosage prediction model, determining the dosage of the phosphorus removal agent, and controlling the dosing control module to add the phosphorus removal agent in the dosage; the training steps of the phosphorus removal agent dosage prediction model comprise: obtaining a plurality of groups of historical dosing records, wherein the historical dosing records comprise historical total phosphorus, historical flow rate, historical total phosphorus, historical flow rate and historical dosage of the phosphorus removal agent; inputting the plurality of groups of historical dosing records as training data into a CNN-TCN-Attention-GPR model for training to obtain the phosphorus removal agent dosage prediction model; the inputting of the plurality of groups of historical dosing records as training data into the CNN-TCN-Attention-GPR model for training comprises: the input layer as the first layer: inputting the training data as an independent variable into the next layer; the independent variable X is represented as: X = (X1, X2,..., X n ) wherein X n is the nth input argument; As the second layer, a convolution layer: normalizes the input argument, and generates a feature matrix from the normalized result using a convolution operation wherein the conv function is a convolution operation in the CNN network, and is an input and an output in the convolution process, wherein / is a sequence length of the normalization processing result, and i and j represent processing positions in the convolution process, is a weight of the convolution layer, is a bias of the convolution layer; wherein, and are the input and output of the activation function, respectively, and f() is the non-linear activation function Relu. the pooling layer as the third layer: performing a pooling operation on the feature matrix: wherein the pooling() pooling function represents a pooling operation, represents the output after the pooling layer. the TCN layer as the fourth layer: performing a convolution operation on the independent variable X: F(i) is the convolution result of the i-th element in (X1, X2, …, X n ), h(j) is the filter / convolution kernel, d is the expansion factor, and k is a constant. the TCN layer further comprises a residual block for performing a residual operation on the convolution result F(i): o=Activation(i+F(i)) wherein o is the residual operation result, and Activation() is an activation function; the attention layer as the fifth layer: performing an attention operation on the data processed by the TCN layer: e t = tanh(w t s t + b t ) e t State vector s representing the t-th feature vector of the data processed by the TCN layer t Energy value to be determined, w t Weight coefficient matrix b representing the t-th feature vector of the data processed by the TCN layer t Offset corresponding to the t-th feature vector of the data processed by the TCN layer, tanh() is an activation function, and α t Intermediate variable, Y is the result of attention operation, and exp is the exponential function with natural constant e as the base the GPR layer as the sixth layer: the Gaussian process regression model is a non-parametric model for performing regression analysis on the output data of the attention layer using a Gaussian prior, and the model is established as follows: y=f(x)+ε where: f(x) is a Gaussian process function, y is an observation, y is an input vector of the attention operation result, and is a noise Therefore, the prior distribution about the observation value y is obtained as follows: the joint prior distribution of the observation value and the predicted value is obtained as follows: where: K(X,X) is an n x n positive definite covariance matrix, K(X * ) is an n x 1 covariance vector between the test point and the test set; K(X * ,X * ) is the covariance of the test point itself, I n is an n-dimensional identity matrix; the posterior distribution of the predicted value is calculated as follows: and wherein: and cov(f * ) is the mean and variance of the predicted values corresponding to the test points; the output layer as the seventh layer: the training set prediction result obtained by training the CNN-TCN-Attention model is recorded as train_X1, and the test set prediction result is recorded as test_X1; train_X1 and test_X1 are used as the training set and test set input data of the GPR layer; the test_X1 is input into the trained GPR model, and the final prediction output Final_predict=GPR.predict(test_X1).

2. The intelligent dosing control system for phosphorus removal treatment of sewage according to claim 1, characterized in that, the obtaining of the plurality of groups of historical dosing records comprises: obtaining the plurality of groups of historical dosing records through a preset local acquisition node; and / or obtaining the plurality of groups of historical dosing records through a plurality of preset non-local acquisition nodes.

3. The intelligent dosing control system for phosphorus removal in sewage treatment according to claim 2, characterized in that, the obtaining of the plurality of groups of historical dosing records through a plurality of preset non-local acquisition nodes comprises: obtaining device information of a preset sewage phosphorus removal treatment dosing device; Formulate a template library based on the device information and preset device requirements, formulate appropriate device requirements, Obtain the node type of the non-local acquisition node, the node type including: direct node and indirect node; When the node type of the non-local acquisition node is a direct node, verify the first credibility of the node identity of the non-local acquisition node, if the verification is passed, obtain multiple sets of historical dosing records of other sewage phosphorus removal treatment dosing equipment that meet the device requirements through the non-local acquisition node; When the node type of the non-local acquisition node is an indirect node, verify the second credibility of the node reputation of the non-local acquisition node, if the verification is passed, obtain multiple sets of historical dosing records of other sewage phosphorus removal treatment dosing equipment that meet the device requirements through the non-local acquisition node.

4. The intelligent dosing control system for phosphorus removal in sewage treatment according to claim 3, characterized in that, Formulate a template library based on the device information and preset device requirements, formulate appropriate device requirements, including: Classify the device information to obtain first classification information of multiple first information categories; Obtain a preset device condition generation template corresponding to the first information category, and generate device conditions based on the condition generation template and the first classification information corresponding to the first information category; Obtain a preset device condition collocation template commonly corresponding to each of the first information categories, and collocate each of the device conditions based on the device condition collocation template to obtain appropriate device requirements.

5. The intelligent dosing control system for phosphorus removal treatment of sewage according to claim 3, characterized in that, Verify the first credibility of the node identity of the non-local acquisition node, including: Obtain the node identity information of the non-local acquisition node; Parse the second information category distribution of the node identity information, the second information category distribution including: multiple sets of one-to-one second information categories and information amounts; Based on a preset first feature extraction template, extract features from the second information category distribution to obtain multiple distribution feature values, and based on the distribution feature values, construct a first feature expression vector; Perform feature vector matching between the first feature expression vector and any second feature expression vector in a preset confidence determination library, if the matching is consistent, obtain the preset first confidence corresponding to the second feature expression vector that matches the first feature expression vector; Obtain a preset identity credibility verification template corresponding to each of the second information categories, and based on the identity credibility verification template, perform identity credibility verification on the second classification information corresponding to the second information category in the node identity information to obtain a verification value; Cumulatively calculate each of the verification values to obtain a second confidence; Based on the first confidence and the second confidence, calculate the first credibility of the node identity of the non-local acquisition node, the calculation formula being as follows: P=J1·K1+J2·K2 Wherein, P is the first credibility, J1 and J2 are preset weight values, K1 is the first confidence, and K2 is the second confidence; If the first credibility is greater than or equal to a preset first credibility threshold, the first credibility of the node identity of the non-local acquisition node is verified.

6. The intelligent dosing control system for phosphorus removal in sewage treatment according to claim 3, characterized in that, Verify the second credibility of the node reputation of the non-local acquisition node, including: Obtain a plurality of information use feedback information after the information is obtained by the non-local acquisition node in history; Based on the preset second feature extraction template, the information use feedback information is extracted, and a plurality of feedback information features are obtained; The feedback information features are matched with any feedback trigger feature in the preset feedback trigger feature library. If the match is consistent, the feedback source type corresponding to the information use feedback information is obtained. The feedback source type includes active feedback and passive feedback. When the feedback source type corresponding to the information use feedback information is passive feedback, the matching consistent feedback trigger feature is taken as the target feedback trigger feature, and the target feedback trigger feature is triggered once. The total number of triggers of each target feedback trigger feature is counted, and the preset trigger total number-third confidence level table corresponding to the target feedback trigger feature is queried to determine the third confidence level corresponding to the trigger total number. The third confidence level is accumulated to obtain the second credibility of the node reputation of the non-local acquisition node; If the second credibility is greater than or equal to the preset second credibility threshold, the second credibility of the node reputation of the non-local acquisition node is verified.

7. The intelligent dosing control system for phosphorus removal in sewage treatment according to claim 1, characterized in that, The host computer is also used to include: When the user inputs the dosing control manual intervention request, the dosing amount determination basis interface is generated based on the preset dosing amount determination basis interface generation template according to the total phosphorus in the water, the water flow, the total phosphorus in the water, and the water flow. The dosing amount determination basis interface is delivered to the user's corresponding preset terminal node; The terminal node returns the user input phosphorus removal agent dosage, and controls the dosing control module to put the user input phosphorus removal agent dosage.

8. The intelligent dosing control system for phosphorus removal in sewage treatment according to claim 1, characterized in that, The host computer is also used to include: Obtain the operation information of the water inlet monitoring module, the water outlet monitoring module and the dosing module; Based on the operation information, determine the operation exception and output the warning.

Images