Methods for sewage discharge from concentrated and contiguous aquaculture households
By using the LSTM-AdaBoost model to calculate pollution weights in centralized continuous seawater aquaculture, the problems of poor monitoring of aquaculture wastewater and indistinguishable responsibilities are solved, and effective wastewater management and environmental supervision are achieved.
Patent Information
- Application Number
- CN202411201321.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-29
- Publication Date
- 2025-05-13
- Estimated Expiration
- 2044-08-29
AI Technical Summary
The poor monitoring of aquaculture wastewater caused by concentrated and continuous seawater aquaculture is difficult to distinguish between the responsibility for environmental pollution, which brings inconvenience to supervision work.
The water quality data prediction model based on LSTM-AdaBoost and the AHP hierarchical analysis method were used to predict water quality monitoring data and pollution weight analysis to control the sewage discharge of the aquaculture pond.
Timely monitoring and management of aquaculture wastewater has been achieved, ensuring that emissions meet standards, avoid exceeding the standard emissions, and facilitate accountability for pollution and improve environmental supervision efficiency.
Smart Images
Figure CN118916615B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of sewage treatment, and in particular relates to a method for discharging sewage from concentrated and contiguous farmers. Background Art
[0002] Concentrated and contiguous seawater aquaculture is an important form of pond aquaculture in my country's coastal areas. It is characterized by large scale and concentrated distribution. When aquaculture wastewater is discharged, a large amount of suspended solids, nitrogen, phosphorus and other pollutants will be discharged into the sea, affecting the coastal marine environment. In addition, concentrated and contiguous seawater aquaculture has concentrated distribution of farmers and relatively concentrated drainage time, which makes it difficult to distinguish the responsibility after environmental pollution, bringing inconvenience to supervision.
[0003] At present, small-scale pond-type marine aquaculture wastewater monitoring adopts manual sampling and laboratory analysis. The timeliness of data acquisition is poor, which is not conducive to timely regulation of tailwater discharge. Factory-scale marine aquaculture wastewater monitoring is often built simultaneously with the circulating water system, which requires a large investment. There is an urgent need for a method that can monitor aquaculture wastewater in a timely manner and distinguish responsibilities after causing environmental pollution. Summary of the invention
[0004] In view of the above-mentioned deficiencies in the prior art, the concentrated and contiguous aquaculture farmers wastewater discharge method provided by the present invention solves the problems of poor timeliness and inconvenient supervision of aquaculture wastewater monitoring of concentrated and contiguous aquaculture farmers.
[0005] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a method for discharging sewage from concentrated and continuous farmers, comprising the following steps:
[0006] S1: Determine the aquaculture ponds and sewage outlets in the concentrated and contiguous aquaculture areas, and set up water quality monitoring points at the sewage outlets;
[0007] S2: Perform water quality monitoring at water quality monitoring points to obtain water quality monitoring data of each water quality monitoring point;
[0008] S3: According to the water quality monitoring data, the trained water quality data prediction model based on LSTM-AdaBoost is used to process and obtain the predicted water quality data;
[0009] S4: Based on the predicted water quality data, the pollution weight of the predicted water quality data is calculated using the AHP hierarchical analysis method;
[0010] S5: According to the pollution weight, control the sewage discharge from the aquaculture ponds corresponding to each water quality monitoring point.
[0011] The beneficial effects of the present invention are as follows: the present invention trains a number of different long short-term memory models LSTM for different aquaculture wastewater pollutants, and based on the AdaBoost adaptive algorithm, uses a number of long short-term memory models LSTM as weak classifiers and combines them into a strong classifier, which can more accurately predict the water quality monitoring data of aquaculture wastewater of each farmer, and adopts the AHP hierarchical analysis method to analyze the pollution weight of each farmer, and determines the discharge of aquaculture wastewater according to the pollution weight, so as to realize the wastewater discharge management of concentrated and contiguous farmers, ensure that the aquaculture wastewater discharged by concentrated and contiguous farmers is within the wastewater discharge standard, effectively avoid the problem of excessive discharge of aquaculture wastewater, and facilitate pollution accountability and improve the efficiency of daily environmental supervision.
[0012] Further: the specific steps of S1 are as follows:
[0013] S101: Divide the breeding ponds and sewage discharge channels of each breeder in the concentrated and contiguous breeding area;
[0014] S102: Determine the sewage outlet of the breeding pond according to the breeding pond and sewage channel of each breeding household, wherein the sewage outlet includes a sewage branch outlet and a sewage main outlet of the sewage channel;
[0015] S103: Set up water quality monitoring points at the sewage outlets of the aquaculture pond and the main sewage outlet of the sewage discharge channel.
[0016] The beneficial effect of the above further scheme is: the breeding areas of concentrated and contiguous breeders are divided to facilitate the setting of water quality monitoring points and the acquisition of water quality monitoring data.
[0017] Further: The specific training steps of the trained water quality data prediction model based on LSTM-AdaBoost in S3 are as follows:
[0018] A1: Obtain historical water quality monitoring data and the types of pollutants in the water quality monitoring data;
[0019] A2: pre-process the historical water quality monitoring data to obtain pre-processed historical water quality monitoring data;
[0020] A3: Divide the pre-processed historical water quality monitoring data into a training set and a test set, and select one of the pollutant types in the water quality monitoring data as the target pollutant;
[0021] A4: Based on the training set and target pollutants, the long short-term memory network model LSTM is used to process and obtain the predicted water quality monitoring data and target pollutant prediction data;
[0022] A5: Calculate the error between the predicted water quality monitoring data and the test set water quality monitoring data, and determine whether it is less than the set first threshold. If so, obtain the trained long short-term memory network model LSTM and proceed to A6. Otherwise, adjust and optimize the parameters of the long short-term memory network model LSTM and return to A4.
[0023] A6: Calculate the error between the target pollutant prediction data and the target pollutant data of the test set, and determine whether it is less than the set second threshold. If so, obtain the long short-term memory network model LSTM for the target pollutant prediction and proceed to A7. Otherwise, adjust and optimize the parameters of the long short-term memory network model LSTM and return to A4.
[0024] A7: Repeat A3 to A6 to obtain several LSTM network models for predicting different target pollutants;
[0025] A8: Define several LSTM network models for predicting different target pollutants as several weak classifiers;
[0026] A9: Based on the historical water quality monitoring data and the AdaBoost adaptive algorithm, several weak classifiers are combined into a strong classifier to obtain a trained water quality data prediction model based on LSTM-AdaBoost.
[0027] The beneficial effects of the above further scheme are as follows: different long short-term memory models LSTM are trained for different aquaculture wastewater pollutants, and based on the AdaBoost adaptive algorithm, the long short-term memory model LSTM is combined into a strong classifier, which further improves the prediction accuracy of the target pollutants while ensuring the prediction accuracy of the strong classifier.
[0028] Further: In A9, several weak classifiers are combined into a strong classifier according to historical water quality monitoring data and the AdaBoost adaptive algorithm, and the specific implementation method is as follows:
[0029] B1: Preprocess the historical water quality monitoring data, use the preprocessed historical water quality detection data as the training set, and initialize the sample weights of the training set;
[0030] B2: According to the training set and the sample weights of the training set, several weak classifiers are trained respectively, and the error rate of each weak classifier is calculated;
[0031] B3: Calculate the weight of each weak classifier according to the error rate of each weak classifier;
[0032] B4: According to the weight of each weak classifier, update the sample weight of the training set corresponding to each weak classifier;
[0033] B5: Determine whether the number of iterations of the AdaBoost adaptive algorithm reaches the set number. If so, complete the iterative training and enter B6. Otherwise, return to B2.
[0034] B6: According to the weight of each weak classifier, all weak classifiers are transformed into a strong classifier to obtain a trained water quality data prediction model based on LSTM-AdaBoost.
[0035] The mathematical expression of the training set in B1 is as follows:
[0036] D={(x1,y1),(x2,y2),…,(x N ,y N )}
[0037] Among them, D is the training set, N is the total number of samples, x N is the feature vector of the Nth sample, y N is the feature vector x N The corresponding true label, x2 is the feature vector of the second sample, y2 is the true label corresponding to the feature vector x2, x1 is the feature vector of the first sample, y1 is the true label corresponding to the feature vector x1;
[0038] The sample weight of the initial training set is expressed as follows:
[0039]
[0040] in, The sample weight initialized for the i-th sample, i is the i-th sample.
[0041] According to the training set and the sample weight of the training set, several weak classifiers are trained respectively, and the error rate of each weak classifier is calculated, and its mathematical expression is as follows:
[0042]
[0043] E k =max|y i -h k (x i )|
[0044] i=1,2,3,4,…,N
[0045] k=1,2,3,4,…,K
[0046] in, is the error rate of the kth weak classifier in the tth round of iterative training, is the square error of the classifier for the i-th sample in the t-th round of iterative training, is the sample weight of the k-th weak classifier corresponding to the i-th sample in the t-1-th round of iterative training, h k (x i ) is the kth weak classifier for the feature vector x i The predicted value, y i is the feature vector x i The corresponding true label, E k is the maximum error rate of the kth weak classifier, max is the maximum value, i is the sequence number of the sample in the training set, k is the sequence number of the weak classifier, N is the total number of samples, K is the total number of weak classifiers, and t is the round of iterative training.
[0047] In B3, the weight of each weak classifier is calculated according to the error rate of each weak classifier, and its mathematical expression is as follows:
[0048]
[0049] Among them, α tk is the weight of the kth weak classifier in the tth round of iterative training, is the error rate of the kth weak classifier in the tth round of iterative training.
[0050] In B4, the sample weight of the training set corresponding to each weak classifier is updated according to the weight of each weak classifier, and its mathematical expression is as follows:
[0051]
[0052] in, is the sample weight of the k-th weak classifier corresponding to the i-th sample in the t-th round of iterative training, is the sample weight of the k-th weak classifier corresponding to the i-th sample in the t-1-th round of iterative training, α tk is the weight of the kth weak classifier in the tth round of iterative training, is the square error of the classifier for the i-th sample in the t-th round of iterative training, Z K is the normalization factor, i is the serial number of the sample in the training set, k is the serial number of the weak classifier, and t is the round of iterative training.
[0053] In B6, according to the weight of each weak classifier, all weak classifiers are transformed into a strong classifier, and a trained water quality data prediction model based on LSTM-AdaBoost is obtained, and its mathematical expression is as follows:
[0054]
[0055] Among them, H(x) is a strong classifier, α tkis the weight of the k-th weak classifier in the t-th round of iterative training, indicating the weight of the k-th weak classifier after completing t rounds of iterative training, h k (x) is the kth weak classifier, k is the sequence number of the weak classifier, and K is the total number of weak classifiers.
[0056] The beneficial effect of the above further scheme is: through the AdaBoost adaptive algorithm, several long short-term memory models LSTM are combined according to different weights, so that the long short-term memory model LSTM with a lower error rate obtains a higher weight, thereby improving the prediction accuracy of the strong classifier and ensuring the reliability of the predicted water quality monitoring data. BRIEF DESCRIPTION OF THE DRAWINGS
[0057] Figure 1 This is a flow chart of a method for concentrated and contiguous aquaculture farmers to discharge wastewater;
[0058] Figure 2 This is a plan view of concentrated and contiguous breeding areas. DETAILED DESCRIPTION
[0059] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.
[0060] like Figure 1 As shown, a flow chart of a method for concentrated and contiguous wastewater discharge from farmers includes the following steps:
[0061] S1: Determine the aquaculture ponds and sewage outlets in the concentrated and contiguous aquaculture areas, and set up water quality monitoring points at the sewage outlets;
[0062] S2: Perform water quality monitoring at water quality monitoring points to obtain water quality monitoring data of each water quality monitoring point;
[0063] S3: According to the water quality monitoring data, the trained water quality data prediction model based on LSTM-AdaBoost is used to process and obtain the predicted water quality data;
[0064] S4: Based on the predicted water quality data, the pollution weight of the predicted water quality data is calculated using the AHP hierarchical analysis method;
[0065] S5: According to the pollution weight, control the sewage discharge from the aquaculture ponds corresponding to each water quality monitoring point.
[0066] In one embodiment of the present invention, Figure 2The figure shows a schematic diagram of a concentrated breeding area. According to S1, the breeding ponds and sewage outlets in the concentrated breeding area are determined, and water quality monitoring points are set at the sewage outlets. The specific steps are as follows:
[0067] S101: Divide the breeding ponds and sewage discharge channels of each breeder in the concentrated and contiguous breeding area, including: breeder A, breeder B, breeder C and breeder D, wherein breeder A and breeder D each have 8 breeding ponds, and breeder B and breeder C each have 4 breeding ponds;
[0068] S102: Determine the sewage outlet of the breeding pond according to the breeding pond and sewage channel of each breeding household, wherein the sewage outlet includes a sewage branch outlet and a sewage main outlet of the sewage channel, the breeding pond of each breeding household is connected to the sewage channel through the sewage branch outlet, and the sewage channel is connected to the ocean through the sewage main outlet;
[0069] S103: Set up water quality monitoring points at the sewage outlets of the aquaculture pond and the main sewage outlet of the sewage discharge channel.
[0070] In one embodiment of the present invention, based on the long short-term memory neural network model LSTM and the AdaBoost adaptive algorithm, the long short-term memory neural network model LSTM is used as a weak classifier, and is combined into a strong classifier through the AdaBoost adaptive algorithm to construct a water quality data prediction model, and a trained water quality data prediction model based on LSTM-AdaBoost is obtained. The specific training steps are as follows:
[0071] A1: Obtain historical water quality monitoring data and the types of pollutants in the water quality monitoring data, including ammonia nitrogen, chemical oxygen demand, organic pollutants, phosphorus, fouling organisms and other aquaculture wastewater pollutants;
[0072] A2: Preprocessing the historical water quality monitoring data to obtain preprocessed historical water quality monitoring data, wherein the preprocessing methods include: data cleaning, normalization processing and conventional preprocessing methods;
[0073] A3: Divide the pre-processed historical water quality monitoring data into a training set and a test set, and select one of the pollutant types in the water quality monitoring data as the target pollutant. The technicians can select different target pollutants according to the specific types of breeding ponds and livestock and poultry;
[0074] A4: Based on the training set and target pollutants, the long short-term memory network model LSTM is used to process and obtain the predicted water quality monitoring data and target pollutant prediction data;
[0075] A5: Calculate the error between the predicted water quality monitoring data and the test set water quality monitoring data, and determine whether it is less than the set first threshold. If so, obtain the trained long short-term memory network model LSTM and proceed to A6. Otherwise, adjust and optimize the parameters of the long short-term memory network model LSTM and return to A4.
[0076] A6: Calculate the error between the target pollutant prediction data and the target pollutant data of the test set, and determine whether it is less than the set second threshold. If so, obtain the long short-term memory network model LSTM for the target pollutant prediction and proceed to A7. Otherwise, adjust and optimize the parameters of the long short-term memory network model LSTM and return to A4.
[0077] The long short-term memory network model LSTM includes: input gate, forget gate and output gate;
[0078] The forget gate determines whether the previous state information is forgotten at the current time step, which can be expressed as:
[0079] f t =σ(W f ·[h t-1 ,x t ]+b f )
[0080] Among them, f t is the output of the forget gate, σ() is the sigmoid activation function, W f is the weight of the forget gate, b f is the bias term of the forget gate, x t is the input at time t, h t-1 is the hidden state of the unit at time t-1;
[0081] The input gate determines the updated state that should be stored at the current time step, which can be expressed as:
[0082] i t =σ(W i ·[h t-1 ,x t ]+b i )
[0083]
[0084] Among them, i t is the output of the input gate, W i is the weight of the input gate, b i is the bias term of the input gate, W c is the weight of the candidate state, b c is the bias term of the candidate state, tanh() is the hyperbolic tangent function, is the candidate state at time t;
[0085] The output gate outputs the result based on the current state, the forget gate, and the result of the input gate, which can be expressed as:
[0086] o t =σ(W o ·[h t-1 ,x t ]+b o )
[0087] h t =o t *tanh(C t )
[0088] Among them, t is the output of the output gate, h t is the hidden state of the unit at time t, C t is the state at time t, W o is the weight of the output gate, b o is the bias term of the output gate.
[0089] In one embodiment of the present invention, the error between the predicted water quality monitoring data and the test set water quality monitoring data is calculated, and the root mean square error between the predicted water quality monitoring data and the test set water quality monitoring data can be selected, and its expression is as follows:
[0090]
[0091] Among them, M is the total number of water quality monitoring data in the test set, g is the sequence number of the water quality monitoring data in the test set, and Y g is the actual value of the g-th water quality monitoring data in the test set, is the predicted value of the g-th water quality monitoring data.
[0092] A7: Repeat A3 to A6 to obtain several LSTM network models for predicting different target pollutants;
[0093] According to the target pollutant selection by the technicians, the long short-term memory network model LSTM for chemical oxygen demand prediction, the long short-term memory network model LSTM for organic pollutant prediction, the long short-term memory network model LSTM for phosphorus prediction and the long short-term memory network model LSTM for other pollutants prediction can be obtained;
[0094] A8: Define several LSTM network models for predicting different target pollutants as several weak classifiers;
[0095] A9: Based on historical water quality monitoring data and the AdaBoost adaptive algorithm, several weak classifiers are combined into a strong classifier to obtain a trained water quality data prediction model based on LSTM-AdaBoost;
[0096] These long short-term memory network models LSTM for predicting different pollutants are combined as weak classifiers to obtain a strong classifier. When predicting the target pollutant, the long short-term memory network model LSTM for predicting the target pollutant has a higher weight, which can improve the prediction accuracy of the strong classifier for the target pollutant while ensuring the prediction accuracy of other water quality monitoring data.
[0097] In one embodiment of the present invention, several weak classifiers are combined into a strong classifier based on historical water quality monitoring data and the AdaBoost adaptive algorithm, and the specific implementation method is as follows:
[0098] B1: Preprocess the historical water quality monitoring data, use the preprocessed historical water quality detection data as the training set, and initialize the sample weight of the training set. The mathematical expression of the training set is as follows:
[0099] D={(x1,y1),(x2,y2),…,(x N ,y N )}
[0100] Among them, D is the training set, N is the total number of samples, and x N is the feature vector of the Nth sample, y N is the feature vector x N The corresponding true label, x2 is the feature vector of the second sample, y2 is the true label corresponding to the feature vector x2, x1 is the feature vector of the first sample, y1 is the true label corresponding to the feature vector x1;
[0101] Initialize the sample weights of the training set, and its mathematical expression is as follows:
[0102]
[0103] in, The sample weight initialized for the i-th sample, i is the i-th sample;
[0104] B2: According to the training set and the sample weights of the training set, several weak classifiers are trained respectively, and the error rate of each weak classifier is calculated. The mathematical expression is as follows:
[0105]
[0106] E k =max|y i -h k(x i )|
[0107] i=1,2,3,4,…,N
[0108] k=1,2,3,4,…,K
[0109] in, is the error rate of the kth weak classifier in the tth round of iterative training, is the square error of the classifier for the i-th sample in the t-th round of iterative training, is the sample weight of the k-th weak classifier corresponding to the i-th sample in the t-1-th round of iterative training, h k (x i ) is the kth weak classifier for the feature vector x i The predicted value, y i is the feature vector x i The corresponding true label, E k is the maximum error rate of the kth weak classifier, max is the maximum value, i is the serial number of the sample in the training set, k is the serial number of the weak classifier, N is the total number of samples, K is the total number of weak classifiers, and t is the round of iterative training;
[0110] B3: According to the error rate of each weak classifier, the weight of each weak classifier is calculated, and its mathematical expression is as follows:
[0111]
[0112] Among them, α tk is the weight of the kth weak classifier in the tth round of iterative training, is the error rate of the kth weak classifier in the tth round of iterative training;
[0113] B4: According to the weight of each weak classifier, update the sample weight of the training set corresponding to each weak classifier. Its mathematical expression is as follows:
[0114]
[0115] in, is the sample weight of the k-th weak classifier corresponding to the i-th sample in the t-th round of iterative training, is the sample weight of the k-th weak classifier corresponding to the i-th sample in the t-1-th round of iterative training, α tk is the weight of the kth weak classifier in the tth round of iterative training, is the square error of the classifier for the i-th sample in the t-th round of iterative training, Z K is the normalization factor, i is the serial number of the sample in the training set, k is the serial number of the weak classifier, and t is the round of iterative training;
[0116] B5: Determine whether the number of iterations of the AdaBoost adaptive algorithm reaches the set number. If so, complete the iterative training and enter B6. Otherwise, return to B2.
[0117] B6: According to the weight of each weak classifier, all weak classifiers are transformed into a strong classifier, and the trained water quality data prediction model based on LSTM-AdaBoost is obtained. Its mathematical expression is as follows:
[0118]
[0119] Among them, H(x) is a strong classifier, α tk is the weight of the k-th weak classifier in the t-th round of iterative training, indicating the weight of the k-th weak classifier after completing t rounds of iterative training, h k (x) is the kth weak classifier, k is the sequence number of the weak classifier, and K is the total number of weak classifiers.
[0120] In one embodiment of the present invention, during the discharge of aquaculture wastewater by farmers, water quality information is monitored every 15 minutes, and the acquired water quality monitoring data is analyzed and processed using a water quality data prediction model to predict the water quality data of each branch outlet and the total outlet, and the AHP hierarchical analysis method is used for weight analysis to obtain the pollutant weights of each farmer during the discharge of aquaculture wastewater, and to formulate the farmer's discharge plan based on this; the AHP analysis method is an existing conventional technology, which decomposes complex problems into multiple components by establishing a hierarchical model, and compares and scores each factor in pairs to determine the relative importance weight of each factor, and finally conducts a comprehensive evaluation.
[0121] In one embodiment of the present invention, when the weight of farmer A is the highest, the aquaculture wastewater discharge of farmer A has the greatest impact on the environment. At this time, more stringent emission monitoring and aquaculture wastewater treatment measures are implemented on farmer A, its emission time is restricted, and further aquaculture wastewater treatment is carried out; when the weight of farmer C is the lowest, the aquaculture wastewater discharge of farmer C has a smaller impact on the environment. At this time, a more relaxed emission monitoring can be adopted for farmer C.
[0122] The beneficial effects of the present invention are as follows: the present invention trains a number of different long short-term memory models LSTM for different aquaculture wastewater pollutants, and based on the AdaBoost adaptive algorithm, uses a number of long short-term memory models LSTM as weak classifiers to combine into a strong classifier, which can more accurately predict the water quality monitoring data of aquaculture wastewater. At the same time, the AHP hierarchical analysis method is used to analyze the pollution weight of each breeder, and the discharge of aquaculture wastewater is determined according to the pollution weight. In addition, the breeders are promptly and quickly differentiated after environmental pollution occurs, so as to realize the wastewater discharge management of concentrated and contiguous breeders, ensure that the aquaculture wastewater discharged by concentrated and contiguous breeders is within the wastewater discharge standard, effectively avoid the problem of excessive discharge of aquaculture wastewater, and improve the efficiency of daily environmental supervision.
Claims
1. A method for discharging sewage from concentrated and contiguous farmers, characterized in that: The following steps are involved: S1: Determine the aquaculture ponds and sewage outlets in the concentrated and contiguous aquaculture areas, and set up water quality monitoring points at the sewage outlets; S2: Perform water quality monitoring at water quality monitoring points to obtain water quality monitoring data of each water quality monitoring point; S3: According to the water quality monitoring data, the trained water quality data prediction model based on LSTM-AdaBoost is used to process and obtain the predicted water quality data; S4: Based on the predicted water quality data, the pollution weight of the predicted water quality data is calculated using the AHP hierarchical analysis method; S5: According to the pollution weight, control the sewage discharge of the aquaculture ponds corresponding to each water quality monitoring point; The specific training steps of the trained water quality data prediction model based on LSTM-AdaBoost in S3 are as follows: A1: Obtain historical water quality monitoring data and the types of pollutants in the water quality monitoring data; A2: pre-process the historical water quality monitoring data to obtain pre-processed historical water quality monitoring data; A3: Divide the pre-processed historical water quality monitoring data into a training set and a test set, and select one of the pollutant types in the water quality monitoring data as the target pollutant; A4: Based on the training set and target pollutants, the long short-term memory network model LSTM is used to process and obtain the predicted water quality monitoring data and target pollutant prediction data; A5: Calculate the error between the predicted water quality monitoring data and the test set water quality monitoring data, and determine whether it is less than the set first threshold. If so, obtain the trained long short-term memory network model LSTM and proceed to A6. Otherwise, adjust and optimize the parameters of the long short-term memory network model LSTM and return to A4. A6: Calculate the error between the target pollutant prediction data and the target pollutant data of the test set, and determine whether it is less than the set second threshold. If so, obtain the long short-term memory network model LSTM for the target pollutant prediction and proceed to A7. Otherwise, adjust and optimize the parameters of the long short-term memory network model LSTM and return to A4. A7: Repeat A3 to A6 to obtain several LSTM network models for predicting different target pollutants; A8: Define several LSTM network models for predicting different target pollutants as several weak classifiers; A9: Based on the historical water quality monitoring data and the AdaBoost adaptive algorithm, several weak classifiers are combined into a strong classifier to obtain a trained water quality data prediction model based on LSTM-AdaBoost.
2. The method for discharging sewage from concentrated and continuous aquaculture households according to claim 1, characterized in that: The specific steps of S1 are as follows: S101: Divide the breeding ponds and sewage discharge channels of each breeder in the concentrated and contiguous breeding area; S102: Determine the sewage outlet of the breeding pond according to the breeding pond and sewage channel of each breeding household, wherein the sewage outlet includes a sewage branch outlet and a sewage main outlet of the sewage channel; S103: Set up water quality monitoring points at the sewage outlets of the aquaculture pond and the main sewage outlet of the sewage discharge channel.
3. The method for discharging sewage from concentrated and contiguous farmers according to claim 1, characterized in that: In A9, several weak classifiers are combined into a strong classifier according to historical water quality monitoring data and the AdaBoost adaptive algorithm. The specific implementation method is as follows: B1: Preprocess the historical water quality monitoring data, use the preprocessed historical water quality monitoring data as a training set, and initialize the sample weights of the training set; B2: According to the training set and the sample weights of the training set, several weak classifiers are trained respectively, and the error rate of each weak classifier is calculated; B3: Calculate the weight of each weak classifier according to the error rate of each weak classifier; B4: According to the weight of each weak classifier, update the sample weight of the training set corresponding to each weak classifier; B5: Determine whether the number of iterations of the AdaBoost adaptive algorithm reaches the set number. If so, complete the iterative training and enter B6. Otherwise, return to B2. B6: According to the weight of each weak classifier, all weak classifiers are combined into a strong classifier to obtain a trained water quality data prediction model based on LSTM-AdaBoost.
4. The method for discharging sewage from concentrated and contiguous farmers according to claim 3, characterized in that: The mathematical expression of the training set in B1 is as follows: D={(x1,y1),(x2,y2),…,(x N ,y N )} Among them, D is the training set, N is the total number of samples, and x N is the feature vector of the Nth sample, y N is the feature vector x N The corresponding true label, x2 is the feature vector of the second sample, y2 is the true label corresponding to the feature vector x2, x1 is the feature vector of the first sample, y1 is the true label corresponding to the feature vector x1; The sample weight of the initial training set is expressed as follows: in, The sample weight initialized for the i-th sample, i is the i-th sample.
5. The method for discharging sewage from concentrated and contiguous farmers according to claim 3, characterized in that: According to the training set and the sample weight of the training set, several weak classifiers are trained respectively, and the error rate of each weak classifier is calculated, and its mathematical expression is as follows: E k =max|y i -h k (x i )| i=1,2,3,4,…,N k=1,2,3,4,…,K in, is the error rate of the kth weak classifier in the tth round of iterative training, is the square error of the classifier for the i-th sample in the t-th round of iterative training, is the sample weight of the k-th weak classifier corresponding to the i-th sample in the t-1-th round of iterative training, h k (x i ) is the kth weak classifier for the feature vector x i The predicted value, y i is the feature vector x i The corresponding true label, E k is the maximum error rate of the kth weak classifier, max is the maximum value, i is the sequence number of the sample in the training set, k is the sequence number of the weak classifier, N is the total number of samples, K is the total number of weak classifiers, and t is the round of iterative training.
6. The method for discharging sewage from concentrated and contiguous farmers according to claim 5, characterized in that: In B3, the weight of each weak classifier is calculated according to the error rate of each weak classifier, and its mathematical expression is as follows: Among them, α tk is the weight of the kth weak classifier in the tth round of iterative training, is the error rate of the kth weak classifier in the tth round of iterative training.
7. The method for discharging sewage from concentrated and contiguous farmers according to claim 6, characterized in that: In B4, the sample weight of the training set corresponding to each weak classifier is updated according to the weight of each weak classifier, and its mathematical expression is as follows: in, is the sample weight of the k-th weak classifier corresponding to the i-th sample in the t-th round of iterative training, is the sample weight of the k-th weak classifier corresponding to the i-th sample in the t-1-th round of iterative training, α tk is the weight of the kth weak classifier in the tth round of iterative training, is the square error of the classifier for the i-th sample in the t-th round of iterative training, Z K is the normalization factor, i is the serial number of the sample in the training set, k is the serial number of the weak classifier, and t is the round of iterative training.
8. The method for concentrated and continuous sewage discharge of farmers according to claim 7, characterized in that: In B6, according to the weight of each weak classifier, all weak classifiers are combined into a strong classifier to obtain a trained water quality data prediction model based on LSTM-AdaBoost, and its mathematical expression is as follows: Among them, H(x) is a strong classifier, α tk is the weight of the k-th weak classifier in the t-th round of iterative training, indicating the weight of the k-th weak classifier after completing t rounds of iterative training, h k (x) is the kth weak classifier, k is the sequence number of the weak classifier, and K is the total number of weak classifiers.