Self-organizing modular neural network effluent ammonia nitrogen detection method and system based on memetic algorithm

Abstract

The invention relates to the technical field of sewage treatment and detection, in particular to a self-organizing modular neural network effluent ammonia nitrogen detection method and system based on a memetic algorithm, and the method comprises the steps: collecting actual operation data of an urban sewage treatment plant to construct a data set; dividing a data segment of the training sample into a plurality of different operation conditions by adopting a multivariate segmentation method based on a memetic algorithm; designing a self-organizing sub-network based on variance sensitivity-mutual information, and training each sub-network by using different operation condition data; and inputting a test data sample into the model, integrating output results of each sub-network, carrying out reverse normalization on model output, and outputting an effluent ammonia nitrogen predicted value. Real-time detection of effluent ammonia nitrogen in the sewage treatment process is achieved through the effluent ammonia nitrogen soft measurement model, and the problem that the effluent ammonia nitrogen concentration in the sewage treatment process is difficult to detect in real time is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of sewage treatment detection, and in particular to a method and system for detecting effluent ammonia nitrogen using a self-organizing modular neural network based on a memetic algorithm. Background Art

[0002] Real-time detection of effluent ammonia nitrogen concentration is particularly important for the stable operation of the sewage treatment process. However, the urban sewage treatment process is a complex, highly nonlinear physical and chemical reaction, microbial reaction and other processes, with large water flow, large fluctuations, and strong coupling between various water quality parameters. In the sewage treatment process, microbial reactions are complex, biochemical reaction theory is not yet mature, the mechanism is unclear, and it is easily disturbed by external factors. It is difficult to establish a mathematical model of the reaction process, and it is difficult to accurately detect effluent ammonia nitrogen in real time.

[0003] The main methods are:

[0004] (1) Most sewage treatment plants use biochemical analysis methods or instruments such as Nessler's reagent colorimetry and ion chromatography to quickly detect the ammonia nitrogen concentration in effluent water in real time. However, these chemical or biological analysis methods are cumbersome, require a certain reaction time, have a large hysteresis, and the detection accuracy is easily affected by the external environment and human operation factors, making it difficult to promote and apply them on a large scale in practical applications; and the maintenance cost of instrument detection methods is high, and they are easily affected by the harsh environment of the sewage treatment process, causing instrument pollution and reduced detection accuracy. Therefore, the study of efficient real-time detection methods for ammonia nitrogen concentration in effluent from sewage treatment processes is of great significance to the intelligentization of urban sewage treatment processes.

[0005] (2) Existing public technology: A soft measurement method for effluent ammonia nitrogen based on modular neural network of EMD algorithm. This invention patent uses the task decomposition algorithm based on EMD to decompose the effluent ammonia nitrogen time series into multiple simple, single-frequency, independent sub-time series; sample entropy and Euclidean distance are used to calculate the complexity and similarity of the sub-time series, merge the sub-time series with small complexity and large similarity, and establish the corresponding sub-network module for learning; the results show that the prediction model can automatically adjust the network structure and accurately predict the effluent ammonia nitrogen concentration in the sewage treatment process.

[0006] The current technical defects are: 1. The sewage treatment process has the characteristics of large water flow, large fluctuations, and strong coupling between various water quality parameters. This technology is a single time series prediction, which only uses historical effluent ammonia nitrogen data to predict its concentration. It is difficult to obtain rich information about the sewage treatment process, reducing the accuracy of the prediction model. 2. When the EMD algorithm decomposes the effluent ammonia nitrogen time series, the processing capacity is limited and it is easily affected by data mutations and abnormal points, resulting in unstable decomposition results. Summary of the invention

[0007] The present invention aims to solve the problem that the concentration of ammonia nitrogen in effluent is difficult to detect in real time during sewage treatment, and proposes a self-organizing modular neural network effluent ammonia nitrogen detection method based on a memetic algorithm, which realizes the real-time detection of ammonia nitrogen in effluent during sewage treatment.

[0008] To achieve the above purpose, the technical solution adopted is:

[0009] A method for detecting ammonia nitrogen in effluent water by a self-organizing modular neural network based on a memetic algorithm, comprising:

[0010] Collect actual operation data of urban sewage treatment plants to build a data set;

[0011] A multivariate segmentation method based on a meme algorithm is used to divide the data segments of the training samples into multiple different operating conditions;

[0012] Design a self-organizing sub-network based on variance sensitivity-mutual information, and train each sub-network using data from different operating conditions;

[0013] The test data samples are input into the model, the output results of each sub-network are integrated, and the model output is denormalized to output the predicted value of effluent ammonia nitrogen.

[0014] According to the effluent ammonia nitrogen detection method of the self-organizing modular neural network based on the meme algorithm of the present invention, further, the data set X is normalized to obtain X; five water quality variables are selected as the input of the model: influent total phosphorus TP, anaerobic terminal redox potential ORP, aerobic front end dissolved oxygen DO, aerobic terminal total suspended solids TSS and effluent pH.

[0015] According to the method for detecting ammonia nitrogen in effluent using a self-organizing modular neural network based on a meme algorithm of the present invention, further, using a genetic algorithm to divide the training samples into data segments includes:

[0016] Randomly generate an initial population containing a set of segmentation breakpoints 1,…,k as chromosomes;

[0017] The fitness value of each segmentation breakpoint is evaluated through the fitness function;

[0018] According to the fitness value, the better segmentation breakpoints are selected, and chromosome crossover and gene mutation are performed to generate the next generation population for global search;

[0019] If the set termination condition is met, the training sample is divided into k data segments with breakpoints of N L =[n1,…,n k+1 ], then each data segment is

[0020] According to the self-organizing modular neural network effluent ammonia nitrogen detection method based on the memetic algorithm of the present invention, further, the data segment is divided into a plurality of different operating conditions including:

[0021] ① Use dynamic time warping algorithm to calculate the similarity of data segments and calculate the local density of data segments;

[0022] ②Select the data segment with local density greater than the set threshold as the operating condition center;

[0023] ③ Use the dynamic time warping algorithm to calculate the distance from the data segment to the center of each operating condition, and divide the data segment into the operating condition with the closest distance;

[0024] ④ After the new data segment is included in the operating condition, the local density of the new data segment is calculated and compared with the local density of the center of the original operating condition, and the data segment with large local density is selected as the operating center of the operating condition;

[0025] ⑤ Set the operating condition S i Assign it as a subtask to the corresponding sub-network for training and learning.

[0026] According to the self-organizing modular neural network effluent ammonia nitrogen detection method based on the memetic algorithm of the present invention, further, a self-organizing sub-network based on variance sensitivity-mutual information is designed, and each sub-network is trained using different operating condition data, including:

[0027] Design a sub-network topology for the sub-task: The sub-network structure includes an input layer, a hidden layer, and an output layer. The input layer has 5 neuron nodes, the hidden layer has p neuron nodes, and the output layer has 1 neuron node;

[0028] The sub-network is trained using the gradient descent learning algorithm. During the training process, when the network reaches a local minimum at time t, the learning state of the network is evaluated;

[0029] If the network error E(t)>E d When E d is the expected error, indicating that the sub-network cannot learn the current sample at this time, so a hidden layer neuron is added;

[0030] If the network error E(t)≤E d When , the hidden layer neurons with smaller contribution in the sub-network are deleted.

[0031] According to the self-organizing modular neural network effluent ammonia nitrogen detection method based on the memetic algorithm of the present invention, further, when E(t)>E d When , the variance sensitive analysis method is used to calculate the contribution of hidden layer neurons to the sub-network output, and the neurons with the largest contribution to the sub-network output are found and added.

[0032] According to the method for detecting ammonia nitrogen in effluent using a self-organizing modular neural network based on a meme algorithm of the present invention, further, a mutual information algorithm is used to calculate the correlation between neurons in each hidden layer respectively; if neuron i has a smaller contribution and a smaller correlation with other neurons, the neuron is deleted; and the connection weight of neuron j having the greatest correlation with neuron i is compensated.

[0033] According to the self-organizing modular neural network effluent ammonia nitrogen detection method based on the meme algorithm of the present invention, further, after the test data sample is input into the model, the test data sample is divided into multiple test operating conditions, and the dynamic time warping algorithm is used to calculate the similarity between the test operating conditions and each sub-network, and the sub-network with the greatest similarity between each test operating condition is activated for testing.

[0034] Furthermore, the present invention also proposes a self-organizing modular neural network effluent ammonia nitrogen detection system based on a memetic algorithm, comprising:

[0035] The data set construction module is used to collect the actual operation data of the urban sewage treatment plant to construct the data set;

[0036] An operating condition division module is used to divide the data segments of the training samples into a plurality of different operating conditions by using a multivariate segmentation method based on a meme algorithm;

[0037] Model building module, used to design self-organizing sub-networks based on variance sensitivity-mutual information, and train each sub-network using data from different operating conditions;

[0038] The test module is used to input the test data samples into the model, integrate the output results of each sub-network, denormalize the model output, and output the predicted value of ammonia nitrogen in the outlet water.

[0039] The beneficial effects achieved by adopting the above technical solution are:

[0040] The present invention provides a method for detecting ammonia nitrogen in effluent of a self-organizing modular neural network based on a memetic algorithm. First, a task decomposition algorithm based on a memetic algorithm is used to divide a complex sewage treatment process into multiple independent operating conditions. A feedforward neural network self-organizing method based on variance sensitivity-mutual information is designed. According to the different operating conditions divided, a sub-network model with structural self-organization is constructed to realize real-time detection of ammonia nitrogen concentration in effluent of the sewage treatment process, thereby solving the problem that it is difficult to detect ammonia nitrogen concentration in effluent of the sewage treatment process in real time, and promoting the development of key technologies for urban sewage treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings of the embodiments of the present invention, wherein the drawings are only used to illustrate some embodiments of the present invention, but not to limit all embodiments of the present invention thereto.

[0042] Figure 1 It is a framework diagram of a method for detecting ammonia nitrogen in effluent using a self-organizing modular neural network based on a memetic algorithm according to an embodiment of the present invention;

[0043] Figure 2 is a schematic diagram of data segment segmentation according to an embodiment of the present invention;

[0044] Figure 3 It is a graph showing the training RMSE and the number of neurons in the hidden layer of each sub-network of the effluent ammonia nitrogen soft-sensing model according to an embodiment of the present invention;

[0045] Figure 4 It is a graph of test results of each sub-network of the effluent ammonia nitrogen soft sensor model according to an embodiment of the present invention;

[0046] Figure 5 This is a test result diagram of the effluent ammonia nitrogen soft sensor model according to an embodiment of the present invention;

[0047] Figure 6 This is a test error diagram of the effluent ammonia nitrogen soft sensor model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings of specific embodiments of the present invention to clearly and completely describe the exemplary scheme of the embodiment of the present invention. Unless otherwise defined, the technical terms or scientific terms used in the present invention should be the common meanings understood by people with ordinary skills in the field.

[0049] The experimental data of the present invention are water quality analysis data of a sewage treatment plant collected in 2014, and the data are collected every 15 minutes; according to the mechanism of the sewage treatment process and prior knowledge, the influent total phosphorus TP, the anaerobic terminal redox potential ORP, the aerobic front dissolved oxygen DO, the aerobic terminal total solid suspended solids TSS and the effluent pH are selected as the input of the model, and the effluent ammonia nitrogen is used as the output of the model; after data preprocessing, there are 400 groups of data samples, 300 groups of samples are selected as training samples of the model, and the remaining 100 groups of samples are used as test samples of the model.

[0050] like Figure 1 As shown, this embodiment discloses a method for detecting ammonia nitrogen in effluent water using a self-organizing modular neural network based on a memetic algorithm, comprising the following steps:

[0051] Step S101: Collect actual operation data of urban sewage treatment plants to construct a data set.

[0052] Water quality variables with a high correlation with effluent ammonia nitrogen were selected as the input of the soft sensor model: influent total phosphorus TP, anaerobic terminal redox potential ORP, aerobic front end dissolved oxygen DO, aerobic terminal total suspended solids TSS and effluent pH.

[0053] Normalize the data set X to get Mapping to the [0 1] range:

[0054]

[0055] Where, X min is the minimum value of the data set, X max is the maximum value of the data set.

[0056] Step S102: using a multivariate segmentation method based on a meme algorithm to divide the data segments of the training samples into a plurality of different operating conditions.

[0057] First, the genetic algorithm is used to divide the training samples into data segments. The specific steps are as follows:

[0058] Step S1021, randomly generate an initialization population, which includes a set of segmentation breakpoints 1, ..., k as chromosomes, and use them as candidate solutions.

[0059] Step S1022: Evaluate the fitness value of each segmentation breakpoint using a fitness function.

[0060]

[0061] In the formula, is the i-th segment, Σ i is the covariance, and λ is a positive constant.

[0062]

[0063] In the formula, Cov i is the empirical covariance, and I represents the identity matrix.

[0064]

[0065] In the formula, μ i is the average value of the ith segment, x t Represents X n i The data of the data segment.

[0066]

[0067] Step S1023, select the optimal segmentation breakpoint according to the fitness value, perform chromosome crossover and gene mutation to generate the next generation population for global search. And according to the time points distributed in two different segments, the local likelihood is low, use Evaluate it and perform local optimization search on the group.

[0068] Step S1024, evaluate the population through the fitness function to determine whether the set termination condition is met. If so, obtain the optimal solution; otherwise, return to step S1023.

[0069] Step S1025: Finally, Can be divided into 10 data segments, such as Figure 2 As shown, the breakpoint is N L =[29,47,99,123,151,166,196,218,252], then each data segment is i=[1,…,10].

[0070] Then divide the data segment into multiple different operating conditions. The specific steps are as follows:

[0071] ① The dynamic time warping algorithm is used to calculate the similarity of the data segments and the local density of the data segments. The formula is as follows:

[0072]

[0073] In the formula, γ ij for and The similarity between c is the local density threshold.

[0074] ② Create an operating condition center: If the data segment If the local density is greater than the set threshold (select local density 2-4 as the threshold), a new operating condition S is created. new , Operation Condition Center

[0075] ③Dividing the operating conditions S i :Use the dynamic time warping algorithm to calculate the distance from the data segment to the center of each operating condition, and divide the data segment into the operating condition S with the closest distance i .

[0076] ④ Update the operating condition center: the new data segment is included in the operating condition S i After that, the local density of the new data segment is calculated and compared with the local density of the center of the original operating condition, and the data segment with the large local density is selected as the S of the operating condition. i Operations center.

[0077] ⑤ Set the operating condition S i As subtasks, they are assigned to the corresponding sub-networks for training and learning. The RMSE of each sub-network training is as follows: Figure 3 shown.

[0078] Step S103: designing a self-organizing subnetwork based on variance sensitivity-mutual information, specifically including the following contents:

[0079] Step S1031, design the sub-network topology for the sub-task: the sub-network structure includes an input layer, a hidden layer and an output layer; determine the sub-network structure to be a 5-p-1 connection mode, that is, the input layer has 5 neuron nodes, the number of feature variables is input in step S101, the hidden layer has p neuron nodes, and the output layer has 1 neuron node. Suppose at time t the kth sub-task training sample in They are the influent total phosphorus TP, anaerobic terminal redox potential ORP, aerobic front end dissolved oxygen DO, aerobic terminal total suspended solids TSS and effluent pH screened in step S101.

[0080] The output of each layer is:

[0081] Input layer: This layer has 5 input neurons and its output is: u k (t)=[x1(t),x2(t),x3(t),x4(t),x5(t)] T .

[0082] Hidden layer: This layer has p neurons and its output is:

[0083]

[0084] In the formula, H j (t) is the input of the jth hidden layer neuron, v ij is the connection weight between the ith input neuron and the jth hidden layer neuron, b h is the hidden layer bias, is the activation function of the jth hidden layer neuron, which is defined as:

[0085] Output layer: This layer has 1 output neuron, and its output is:

[0086]

[0087] In the formula, w j is the connection weight between the jth hidden layer neuron and the output neuron, b o is the bias for the output layer.

[0088] The root mean square error (RMSE) of subnetwork training is defined as:

[0089]

[0090] In the formula, y d is the expected value, y out is the network output value, and N is the number of training samples.

[0091] Step S1032: The initial number p of neurons in the hidden layer of the sub-network is 2, and the weights v and w are randomly set in the range of [-1 1].

[0092] Step S1033: Use the gradient descent learning algorithm to train the sub-network; during the training process, when the network reaches the minimum local error at time t, the learning state of the network is evaluated, that is, it satisfies:

[0093]

[0094] Where ε is the iteration delay and δ is the smaller threshold.

[0095] Step S1034: If the network error E(t)>E d When E d is the expected error, indicating that the sub-network cannot learn the current sample at this time, so a hidden layer neuron is added to ensure the learning ability of the sub-network, specifically:

[0096] ① Use variance sensitivity analysis to calculate the contribution of hidden layer neurons to model output; if a model output is f(x), the model input is x; for x 0 x + When , the model change Δy is expanded through Taylor series as:

[0097]

[0098] In the formula, o(||x + –x 0 || 2 ) is a high-order infinitesimal, when Δx i Small enough that it can be ignored:

[0099]

[0100] Therefore, the variable x i The sensitivity to the model output is defined as:

[0101]

[0102] In the sub-network structure design, the contribution of the i-th hidden layer neuron to the network output is defined as:

[0103]

[0104] ② Calculate the contribution of each hidden layer neuron, find the neuron j that contributes the most to the sub-network output, and set the initial weight of the newly added hidden layer neuron to:

[0105] v :new (t) = v :j (t)

[0106]

[0107] Where, e(t) = y d (t)-y out (t), v new (t) is the connection weight between the newly added neuron and the input layer, w new (t) is the connection weight between the newly added neuron and the output layer.

[0108] Step S1034: If the network error E(t)≤E d When , the hidden layer neurons with smaller contribution in the sub-network are deleted to keep the network structure compact. Specifically:

[0109] ① The mutual information algorithm is used to calculate the correlation between neurons in each hidden layer to obtain rich local network information. The correlation between two neurons m and n is defined as:

[0110]

[0111] In the formula, H(Ho m (t)) and H(Ho n (t)) are the entropies of neurons m and n, respectively, H(Ho m (t),Ho n (t)) is the joint entropy of two neurons m and n; its formula is as follows:

[0112]

[0113] in,

[0114] ② If neuron i has a small contribution and a small correlation with other neurons, delete the neuron to keep the network compact; to ensure the stability of the network, compensate the connection weight of neuron j that has the greatest correlation with neuron i:

[0115] v new (t) = v j (t)

[0116]

[0117] Step S104: The sewage treatment process is divided into five different operating conditions. The five sub-networks learn the data of each operating condition. The number of neurons in the training RSME and hidden layer of each sub-network is changed as follows: Figure 3 shown.

[0118] Step S105: After training the soft sensor model for effluent ammonia nitrogen, the test data samples are divided into three test operation conditions. The dynamic time warping algorithm is used to calculate the similarity between the test operation conditions and each sub-network module. The sub-network (module 1, module 2 and module 5) with the greatest similarity to the three test operation conditions is activated for testing. The test results of each sub-network module are as follows: Figure 4 shown.

[0119] Step S106: Integrate the output results y of each sub-network module out , and denormalize the model output to output the predicted value of ammonia nitrogen in the effluent. out In fact, it is the output of a sub-network.

[0120] In this embodiment, the measurement results of the effluent ammonia nitrogen concentration are as follows: Figure 5 As shown in the figure, the X-axis is the test sample, in units of pieces; the Y-axis is the effluent ammonia nitrogen concentration, in units of mg / L, the solid line is the expected value of the effluent ammonia nitrogen concentration, and the dotted line is the model output value. Figure 6 As shown, X-axis: expected value, unit is mg / L; Y-axis: network output, unit is mg / L. The results show the effectiveness of the effluent ammonia nitrogen detection method based on the self-organizing modular neural network of the memetic algorithm.

[0121] Tables 1-13 are experimental data of the present invention, wherein Table 1-6 is a training sample: the measured values ​​of influent total phosphorus concentration, anaerobic terminal redox potential, aerobic front-stage dissolved oxygen, aerobic terminal total solid suspended solids, effluent pH, and effluent ammonia nitrogen concentration; Tables 7-12 are test samples: the measured values ​​of influent total phosphorus concentration, anaerobic terminal redox potential, aerobic front-stage dissolved oxygen, aerobic terminal total solid suspended solids, effluent pH, and effluent ammonia nitrogen concentration; Table 13 is the measured value of effluent ammonia nitrogen concentration of the present invention.

[0122] Training samples:

[0123] Table 1. Total phosphorus concentration in influent (mg / L)

[0124]

[0125]

[0126] Table 2. Anaerobic terminal redox potential (mg / L)

[0127] -96.3949 -90.3702 -198.045 -190.547 -201.827 -96.5231 -163.307 -183.56 -209.454 -80.4359 -94.4721 -408.076 -87.0374 -93.0621 -196.443 -188.88 -199.904 -102.42 -164.14 -186.06 -210.159 -75.2444 -265.15 -421.792 -83.256 -97.9331 -195.93 -187.598 -197.597 -110.623 -165.358 -186.381 -196.635 -73.5139 -275.789 -437.751 -73.2576 -103.83 -196.058 -186.957 -194.072 -114.533 -169.012 -186.509 -185.355 -73.0012 -280.084 -457.619 -73.8985 -108.252 -194.328 -186.701 -188.111 -118.25 -171.832 -187.855 -175.998 -75.1162 -282.134 -482.038 -74.0908 -113.315 -193.174 -186.957 -183.432 -121.904 -91.9725 -189.457 -165.038 -78.2568 -284.378 -506.137 -70.4375 -117.225 -192.277 -187.598 -179.266 -125.172 -100.689 -190.867 -157.026 -81.205 -292.517 -521.583 -68.2583 -120.558 -191.957 -187.919 -174.844 -128.249 -111.328 -193.623 -154.206 -81.3973 -305.08 -542.99 -64.0923 -123.57 -191.828 -188.111 -169.396 -131.133 -114.341 -197.917 -143.503 -81.5255 -314.117 -570.422 -58.7086 -126.711 -191.636 -188.175 -164.205 -133.825 -120.558 -199.263 -137.67 -82.1023 -324.051 -579.395 -54.0299 -130.812 -191.444 -189.521 -159.59 -136.581 -115.238 -200.545 -126.006 -82.0382 -329.691 -584.65 -50.633 -135.043 -191.059 -190.482 -155.103 -139.721 -125.877 -200.096 -113.636 -82.1023 -338.408 -588.24 -48.5179 -139.401 -190.675 -191.252 -150.553 -142.477 -131.069 -199.904 -102.035 -81.3973 -340.202 -401.539 -47.8129 -144.016 -190.162 -191.957 -140.042 -145.874 -128.762 -199.584 -95.6258 -76.6544 -344.56 -380.452 -50.7611 -149.015 -189.777 -192.149 -142.157 -149.976 -107.547 -199.712 -93.8312 -76.5904 -360.391 -383.913 -53.1326 -153.565 -189.201 -192.085 -142.862 -153.373 -111.777 -200.353 -91.2675 -76.2058 -363.404 -375.453 -60.5032 -157.475 -188.88 -191.957 -143.887 -156.257 -111.969 -201.186 -87.2297 -75.9494 -365.839 -372.441 -64.0923 -160.872 -188.752 -192.213 -144.849 -158.629 -107.611 -201.955 -86.4606 -77.3595 -367.313 -364.814 -69.9248 -164.012 -193.174 -192.662 -145.81 -160.615 -117.289 -202.468 -84.2173 -78.449 -368.851 -357.956 -75.5649 -167.73 -193.238 -193.11 -146.131 -162.795 -116.328 -202.211 -81.91 -79.9872 -370.966 -349.88 -81.9741 -182.022 -191.508 -192.533 -144.913 -164.653 -120.173 -201.955 -79.3463 -94.6644 -374.043 -350.201 -88.2552 -190.162 -191.38 -191.7 -106.137 -169.204 -128.954 -202.211 -76.334 -94.5362 -378.529 -456.402 -82.0382 -193.815 -191.764 -191.7 -84.0892 -168.691 -139.849 -201.891 -76.0776 -94.3439 -383.913 -508.573 -84.7301 -196.379 -191.7 -200.353 -76.7826 -165.551 -153.053 -201.635 -78.7054 -93.7671 -390.771 -544.4 -86.6529 -198.045 -191.572 -202.404 -85.2428 -164.076 -178.049 -206.954 -80.5 -94.4721 -398.206 -556.834

[0128] Table 3. Dissolved oxygen in the aerobic front section (mg / L)

[0129] 4.8361 5.0619 0.2091 0.2175 0.1947 5.2903 7.2738 0.2021 5.9736 0.1623 4.0795 5.0250 3.9512 0.1695 2.0236 4.6679 3.9797 0.2093 0.1740 0.1950 5.7441 5.6345 0.2181 5.6802 0.1924 3.9646 4.9868 4.2286 0.1912 2.2966 5.5601 3.7773 0.1992 0.1948 0.1740 6.1131 3.3980 0.2096 5.4310 0.1925 4.0803 4.8941 4.4811 0.6876 2.7585 5.4760 3.6212 0.2182 0.1968 0.1932 6.2332 3.2327 0.2017 4.5563 0.1978 4.2508 4.7089 4.7399 1.7019 2.9850 5.3806 3.5188 0.2270 0.1863 0.1926 6.4145 3.0442 0.1980 2.8577 0.2009 4.3996 4.4551 4.9622 2.1296 3.1355 5.5256 3.3626 0.2051 0.1815 0.1867 6.3871 2.9185 0.1944 0.3806 0.1841 4.4104 4.3142 4.4737 2.1379 3.2544 5.5780 2.9670 0.2176 0.2260 0.2013 6.4397 2.7346 0.2184 0.2260 0.2074 4.4186 4.4260 1.9691 2.0820 3.2124 5.7257 1.8388 0.2254 0.2426 0.1991 6.6182 2.2825 0.2572 0.2470 0.1668 4.5772 4.4090 0.1658 2.0489 3.0557 5.7635 0.3539 0.1934 0.2119 0.2185 6.7046 1.8575 0.2073 0.2177 0.2026 4.6793 4.3582 0.1660 2.0375 2.6762 5.7912 0.2411 0.2091 0.1877 0.2228 6.8805 1.1170 0.2449 0.2113 0.1681 4.8446 4.2685 0.1817 2.0237 2.0382 5.7960 0.1959 0.1858 0.2056 0.2176 7.0088 0.2966 3.6240 0.2291 0.1725 5.0582 4.5372 0.2485 2.1228 1.3534 5.8353 0.2211 0.1885 0.2063 0.2207 7.1219 0.2294 4.7762 0.2105 0.1910 5.1176 4.7394 4.5762 2.2464 0.7728 5.9259 0.2105 0.2342 0.1936 0.1923 7.1444 0.1948 5.3974 0.2127 0.2055 5.0474 4.5786 4.5091 2.2857 0.4342 5.9917 0.2158 0.1734 0.1956 0.3682 7.1830 0.2041 5.6035 0.2679 0.1935 4.9798 4.1746 4.3764 2.2877 0.3595 6.0288 0.2403 0.1998 0.2040 0.3479 7.1801 0.2003 5.7539 0.2642 0.1941 4.9498 3.7994 4.1352 2.1766 0.2741 6.0610 0.2542 0.2151 0.1812 0.3516 7.2633 0.1856 6.6685 0.4310 0.4266 4.9328 3.8032 3.9453 1.8995 0.2368 6.1357 0.2090 0.2079 0.1693 0.3502 7.2801 0.2046 6.7041 0.2653 1.1853 4.9357 3.6959 3.2627 1.7168 0.2646 6.1091 0.1953 0.2045 0.1925 0.4392 7.2830 0.1997 6.6429 0.2138 2.0981 4.9835 3.6381 2.6406 1.9951 0.6562 6.0891 0.1905 0.1954 0.1978 2.0988 7.2811 0.1908 6.4968 0.1781 2.6009 5.0008 3.5979 1.5146 2.2138 1.2076 6.0312 0.2168 0.1873 0.1764 4.2810 7.3009 0.2186 6.2672 0.1586 4.1318 5.0560 3.7153 0.1792 2.0540 1.5664

[0130] Table 4. Total suspended solids at the aerobic end (mg / L)

[0131] 2.7774 2.9311 2.7895 2.7396 2.6340 2.7073 3.0457 2.8810 2.7427 2.6340 2.6707 2.6663 2.6874 2.6466 2.5933 2.8052 2.8900 2.8075 2.7467 2.6351 2.8241 2.8940 2.8391 2.7544 2.6127 2.6638 2.7102 2.6890 2.6608 2.6241 2.8471 2.8817 2.7625 2.7052 2.6454 2.8362 2.8624 2.8353 2.7177 2.6521 2.6475 2.7018 2.6994 2.6394 2.5943 2.8279 2.8369 2.7509 2.7137 2.6698 2.9598 2.8843 2.7970 2.6925 2.6122 2.6359 2.7142 2.7121 2.6191 2.6319 2.8052 2.8273 2.7431 2.7205 2.6530 2.8498 2.8920 2.7583 2.6356 2.6436 2.6194 2.6969 2.7304 2.6963 2.6138 2.8218 2.7808 2.7460 2.7211 2.6906 2.8814 2.8422 2.7846 2.6066 2.6160 2.6624 2.6882 2.7117 2.6246 2.6417 2.8097 2.7630 2.7266 2.6907 2.6319 2.8752 2.8917 2.8411 2.6159 2.6351 2.6563 2.6725 2.7074 2.6221 2.6434 2.8109 2.7180 2.7543 2.6955 2.6154 2.8878 2.9100 2.7935 2.6488 2.6484 2.6537 2.6624 2.6985 2.6120 2.6207 2.7951 2.6886 2.7632 2.7496 2.6774 2.9040 2.8675 2.8059 2.6493 2.6528 2.6284 2.6386 2.6871 2.6432 2.6314 2.8411 2.7318 2.7306 2.6741 2.6497 2.8966 2.9315 2.8153 2.6683 2.6691 2.6069 2.6055 2.7049 2.6546 2.6300 2.8942 2.7250 2.7665 2.6996 2.6591 2.9303 2.9153 2.8296 2.6760 2.6481 2.6311 2.5903 2.7110 2.6314 2.6305 2.9344 2.7395 2.7716 2.6351 2.6643 2.9265 2.8921 2.7904 2.6394 2.6571 2.6284 2.5968 2.7084 2.6547 2.6236 2.9005 2.7311 2.7859 2.6550 2.7107 2.9545 2.8856 2.8209 2.6153 2.6329 2.6391 2.5981 2.7448 2.6262 2.6315 2.8991 2.7212 2.8112 2.6217 2.6965 2.9348 2.8582 2.7826 2.6427 2.6485 2.6475 2.6528 2.7125 2.6328 2.6148 2.8871 2.7386 2.7645 2.6318 2.6595 2.9286 2.9216 2.8256 2.6464 2.6913 2.6348 2.7113 2.6958 2.6725 2.6162 2.8783 2.7682 2.7598 2.6262 2.6635 2.9601 2.8805 2.7359 2.6310 2.6633 2.6260 2.7118 2.7086 2.6161 2.6198 2.9275 2.7679 2.7438 2.6473 2.6945 2.9476 2.8769 2.6982 2.6093 2.6480 2.6436 2.6655 2.7099 2.6181 2.6660 2.9116 2.7964 2.7361 2.6600 2.7443 2.9443 2.8998 2.6837 2.6000 2.6609 2.6525 2.6498 2.6586 2.6574 2.6633 2.8838 2.8094 2.7211 2.6637 2.7999 2.9877 2.8622 2.6910 2.6062 2.6616 2.6417 2.6559 2.6691 2.5944 2.6252 2.9486 2.8202 2.7946 2.6306 2.7805 2.9994 2.8355 2.7303 2.6596 2.6604 2.6893 2.6642 2.6626 2.6173 2.6292

[0132] Table 5. Outlet pH5 (mg / L)

[0133] 7.7259 7.8032 7.6578 7.6520 7.7047 7.6573 7.8311 7.7217 7.6993 7.7240 7.7500 7.8054 7.7842 7.8431 7.8166 7.7357 7.8051 7.6580 7.6567 7.6924 7.6720 7.8343 7.7204 7.7166 7.7228 7.7553 7.8191 7.7843 7.8374 7.8150 7.7455 7.8116 7.6558 7.6614 7.6789 7.6890 7.8341 7.7215 7.7294 7.7217 7.7595 7.8237 7.7833 7.8316 7.8193 7.7521 7.8216 7.6524 7.6650 7.6733 7.7062 7.8316 7.7258 7.7330 7.7251 7.7682 7.8168 7.7835 7.8259 7.8270 7.7554 7.8261 7.6501 7.6651 7.6586 7.7199 7.8224 7.7223 7.7250 7.7363 7.7780 7.8210 7.7870 7.8330 7.8338 7.7486 7.8011 7.6545 7.6653 7.6462 7.7325 7.8042 7.7176 7.7180 7.7499 7.7869 7.8343 7.7939 7.8354 7.8395 7.7328 7.7670 7.6497 7.6654 7.6202 7.7462 7.7873 7.7150 7.7168 7.7578 7.7901 7.8430 7.8058 7.8344 7.8385 7.7329 7.7397 7.6426 7.6642 7.6147 7.7589 7.7795 7.7479 7.7315 7.7590 7.7788 7.8449 7.8136 7.8334 7.8421 7.7442 7.7160 7.6456 7.6630 7.6159 7.7648 7.7718 7.7559 7.7429 7.7580 7.7710 7.8402 7.7876 7.8313 7.8469 7.7601 7.6958 7.6555 7.6628 7.6184 7.7685 7.7516 7.7412 7.7475 7.7570 7.7814 7.7804 7.7583 7.8292 7.8494 7.7760 7.6870 7.6643 7.6625 7.6185 7.7756 7.7211 7.6880 7.7442 7.7514 7.7928 7.7374 7.7404 7.8293 7.8450 7.7841 7.6804 7.6732 7.6643 7.6185 7.7804 7.7143 7.6619 7.7340 7.7470 7.8043 7.7327 7.7990 7.8272 7.8361 7.7878 7.6738 7.6743 7.6673 7.6196 7.7931 7.7200 7.6382 7.7170 7.7426 7.8123 7.7578 7.7900 7.8251 7.8112 7.7914 7.6603 7.6664 7.6725 7.6171 7.8092 7.7269 7.6144 7.7033 7.7416 7.8158 7.7819 7.7878 7.8241 7.8102 7.7916 7.6557 7.6575 7.6745 7.6043 7.7888 7.7304 7.5849 7.6895 7.7394 7.8136 7.7935 7.7924 7.8242 7.8183 7.7861 7.6523 7.6509 7.6766 7.6108 7.7672 7.7340 7.6551 7.6699 7.7395 7.8081 7.7983 7.7993 7.8265 7.8241 7.7727 7.6400 7.6467 7.6915 7.6150 7.7786 7.7295 7.6424 7.6638 7.7362 7.8094 7.7996 7.8095 7.8286 7.8276 7.7751 7.6413 7.6458 7.7110 7.6215 7.8000 7.7285 7.6502 7.6872 7.7316 7.8050 7.7952 7.8242 7.8282 7.8289 7.7853 7.6506 7.6437 7.7191 7.6315 7.8169 7.7263 7.6683 7.7108 7.7291 7.7961 7.7908 7.8388 7.8221 7.8279 7.7966 7.6553 7.6484 7.7158 7.6449 7.8268 7.7240 7.6821 7.7219 7.7367 7.7962 7.7852 7.8455 7.8193 7.8267

[0134] Table 6. Effluent ammonia nitrogen concentration (mg / L)

[0135]

[0136]

[0137] Test sample:

[0138] Table 7. Total phosphorus concentration in influent (mg / L)

[0139] 2.3200 1.8700 2.0600 2.4700 1.9000 1.9500 2.6000 1.7600 1.2800 1.5500 2.0500 2.1800 1.9100 2.2200 2.7900 2.0500 2.2300 2.2300 2.4100 2.3900 1.8200 2.1500 1.2800 1.3600 1.8600 1.8600 1.7200 1.9100 2.0700 2.2200 2.3300 2.2100 2.1900 1.9000 2.4400 1.9000 2.0200 1.3800 1.3000 1.9000 1.9200 1.6700 2.0600 2.7400 2.1300 2.1800 2.2800 2.2600 2.3500 1.9500 2.2600 1.3700 1.5400 2.0200 1.8000 1.7300 2.5400 2.4900 2.3300 2.1300 2.0700 2.4700 2.3800 2.4300 2.4400 1.4500 1.1200 1.8200 1.7100 1.7300 2.0800 2.6900 1.8400 2.5100 2.7000 2.4000 1.7800 2.0400 2.7900 1.3300 1.5100 2.0500 1.8700 1.8900 2.5800 2.4600 1.9800 2.7100 2.5500 2.0200 1.8700 2.0100 1.7900 1.5000 1.4600 1.6800 2.0300 2.0400 2.3700 1.8200

[0140] Table 8. Anaerobic terminal redox potential (mg / L)

[0141] -429.611 -290.595 -251.37 -241.5 -234.834 -232.207 -228.169 -213.235 -214.645 -139.144 -328.858 -274.187 -377.312 -282.327 -249.319 -240.603 -234.706 -232.207 -227.72 -216.183 -214.453 -351.418 -330.524 -358.725 -272.777 -247.268 -239.641 -234.578 -231.822 -227.207 -216.44 -216.248 -376.35 -328.089 -347.893 -268.355 -245.602 -239 -234.386 -231.886 -226.566 -213.748 -208.685 -352.316 -319.564 -338.728 -264.701 -244.384 -238.808 -234.514 -230.668 -223.041 -210.351 -184.522 -341.612 -316.744 -325.333 -260.92 -243.551 -238.808 -234.514 -229.899 -219.58 -208.236 -158.629 -338.344 -309.502 -316.68 -258.549 -242.205 -238.744 -234.258 -227.784 -216.953 -209.326 -145.874 -335.523 -300.978 -302.836 -256.049 -242.269 -232.014 -233.937 -226.246 -212.338 -209.838 -155.744 -331.293 -286.557 -297.132 -253.549 -241.692 -235.091 -233.36 -227.592 -208.813 -211.953 -147.22 -322.128 -282.455

[0142] Table 9. Dissolved oxygen in the aerobic front section (mg / L)

[0143] 1.7937 3.2710 0.1354 0.1542 0.1376 0.1698 0.1352 2.6789 3.2877 4.2062 4.5882 4.8141 5.4726 6.5653 0.1745 1.9559 3.0170 0.1461 0.1489 0.1422 0.1610 0.1360 3.1839 3.3306 4.2008 4.8684 4.6739 6.2470 6.5912 0.1242 2.0548 2.6681 0.1538 0.1324 0.1345 0.1655 0.1545 3.6199 3.6333 4.1892 5.0281 4.1697 6.4235 6.6234 2.1917 1.8654 0.1367 0.1318 0.1107 0.1462 0.2233 3.6803 4.0109 4.1971 5.1361 4.0194 6.4872 6.5860 2.4289 0.3985 0.1478 0.1499 0.1312 0.1431 1.1550 3.6033 4.2858 4.1724 5.0532 4.0652 6.5529 6.6861 2.7839 0.1593 0.1596 0.1370 0.1300 0.1447 1.7068 3.4768 4.3714 4.1575 5.0539 4.1195 6.5679 5.8157 3.1233 0.1247 0.1494 0.1336 0.1394 0.1373 2.0151 3.3889 4.3110 4.2920 4.9227 3.9931 6.5468 2.2767

[0144] Table 10. Total suspended solids at the aerobic end (mg / L)

[0145]

[0146]

[0147] Table 11. Outlet pH (mg / L)

[0148] 7.8197 7.8171 7.8332 7.7970 7.7878 7.8050 7.8287 7.8209 7.8377 7.8461 7.8400 7.8616 7.8605 7.8690 7.8148 7.8182 7.7809 7.8311 7.7903 7.7920 7.8075 7.8277 7.8206 7.8398 7.8179 7.8388 7.8637 7.8494 7.8736 7.8046 7.8145 7.7856 7.8279 7.7835 7.7940 7.8099 7.8278 7.8213 7.8421 7.8022 7.8353 7.8636 7.8396 7.8713 7.8128 7.8029 7.8212 7.7812 7.7950 7.8157 7.8279 7.8265 7.8433 7.8090 7.8340 7.8601 7.8433 7.8646 7.8134 7.8157 7.8146 7.7821 7.7971 7.8181 7.8235 7.8318 7.8457 7.8216 7.8372 7.8590 7.8561 7.8522 7.8153 7.8261 7.8091 7.7818 7.7982 7.8239 7.8212 7.8359 7.8482 7.8307 7.8472 7.8591 7.8745 7.8409 7.8173 7.8308 7.8047 7.7849 7.8016 7.8263 7.8200 7.8356 7.8505 7.8365 7.8573 7.8604 7.8724 7.8262

[0149] Table 12. Measured values ​​of effluent ammonia nitrogen concentration (mg / L)

[0150] 4.0006 4.8176 4.8551 7.2663 10.6769 11.5408 11.2822 8.5093 5.0477 4.9363 5.1594 5.0469 4.9203 5.3920 4.3928 4.0011 4.4525 5.0883 7.7022 11.0306 11.5664 11.2261 8.5309 4.9432 5.0221 5.4592 4.9953 4.9580 5.3116 4.3719 4.0722 5.7752 5.2269 8.1355 11.2830 11.5661 10.8974 7.1227 4.7679 5.2491 5.3132 4.9507 5.2008 5.0914 4.1822 5.6392 5.5607 8.7810 11.7954 11.6090 10.1795 6.1419 4.7237 5.3015 5.3929 4.8598 5.1238 4.9048 4.1545 5.1019 5.8562 9.1073 12.3217 11.6395 9.8213 6.1444 4.8317 5.2565 5.2453 4.8432 5.1987 4.8820 4.0025 4.8437 5.9135 9.4816 11.7101 11.7039 9.9373 6.0269 4.8050 5.2096 5.2067 5.1041 5.5725 4.7802 4.2193 4.7245 6.5018 10.2384 11.5128 11.4367 9.5053 5.3591 4.7314 5.2869 5.0340 4.9531 5.4274 4.4173

[0151] Table 13. Measured values ​​of effluent ammonia nitrogen concentration (mg / L)

[0152] 4.2644 4.2768 4.7189 6.3673 10.4021 11.3811 11.2474 9.4679 5.2191 4.8397 5.2706 4.9907 4.9462 5.4152 4.4929 4.1528 4.7241 4.9826 7.1181 10.7802 11.4216 11.1925 8.7156 4.9733 4.9614 5.1487 5.0076 4.8853 5.3655 4.5195 4.2170 4.4476 5.2580 7.6805 11.0408 11.4082 11.1259 8.7186 4.8975 4.9912 5.3857 4.9423 4.9188 5.2907 4.1780 5.5096 5.3331 8.2151 11.2858 11.4164 10.8396 7.1561 4.7902 5.1625 5.2238 4.9136 5.1426 5.0704 4.2474 5.5090 5.4782 8.8496 11.6416 11.4290 10.2436 5.9980 4.7707 5.2355 5.2928 4.8223 5.1177 4.8997 4.2241 5.2315 5.7947 9.1995 11.9726 11.4983 9.6530 5.9967 4.8542 5.2312 5.1527 4.8006 5.1923 4.8668 4.2157 4.8104 6.0177 9.5767 11.5440 11.4884 9.6980 5.8687 4.8667 5.2128 5.1861 5.0467 5.5473 4.7793

[0153] Corresponding to the above method, this embodiment also proposes a self-organizing modular neural network effluent ammonia nitrogen detection system based on a memetic algorithm, comprising:

[0154] The data set construction module is used to collect the actual operation data of urban sewage treatment plants to construct data sets.

[0155] The operating condition division module is used to divide the data segment of the training sample into multiple different operating conditions.

[0156] Model building module, used to design self-organizing sub-networks based on variance sensitivity-mutual information, which is conducive to the use of multivariate segmentation method based on memetic algorithm

[0157] Each sub-network is trained using data from different operating conditions.

[0158] The test module is used to input the test data samples into the model, integrate the output results of each sub-network, denormalize the model output, and output the predicted value of ammonia nitrogen in the outlet water.

[0159] In view of the complex environment of the current urban sewage treatment process, the long cycle, high instrument cost, low precision and other problems in real-time detection of effluent ammonia nitrogen concentration, the present invention adopts a self-organizing modular neural network based on a meme algorithm to construct a soft measurement model of effluent ammonia nitrogen concentration in the sewage treatment process, realizes real-time detection of effluent ammonia nitrogen concentration in the sewage treatment process, and solves the problems of high cost, long cycle and low detection precision of effluent ammonia nitrogen concentration detection in complex environments.

[0160] The present invention uses a self-organizing modular neural network based on a memetic algorithm to construct an intelligent prediction model for NH4-N in the effluent of the urban sewage treatment process, and realizes the real-time detection of effluent ammonia nitrogen in the sewage treatment process. The effluent ammonia nitrogen parameter is one of the important indicators for evaluating the stable and efficient operation of the sewage treatment process and whether the treated wastewater meets the established standards. Real-time detection of ammonia nitrogen concentration is of great significance to ensure the stable and efficient operation of the urban sewage treatment process. It is an important branch of the field of advanced manufacturing technology, belonging to both the field of automatic control and the field of water treatment.

[0161] Unless otherwise specifically stated, the components, steps, numerical expressions and values ​​set forth in these embodiments do not limit the scope of the present invention.

[0162] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0163] The units and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A person of ordinary skill in the art may use different methods to implement the described functions for each specific application, but such implementation is not considered to be beyond the scope of the present invention.

[0164] Those skilled in the art will appreciate that all or part of the steps in the above method can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, such as a read-only memory, a disk or an optical disk. Optionally, all or part of the steps in the above embodiment can also be implemented using one or more integrated circuits, and accordingly, each module / unit in the above embodiment can be implemented in the form of hardware or in the form of software function modules. The present invention is not limited to any specific form of combination of hardware and software.

[0165] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed by the present invention, or replace some of the technical features therein by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A method for detecting ammonia nitrogen in effluent water using a self-organizing modular neural network based on a memetic algorithm, characterized in that: Include: Collect actual operation data of urban sewage treatment plants to build a data set; A multivariate segmentation method based on a meme algorithm is used to divide the data segments of the training samples into multiple different operating conditions; Design a self-organizing sub-network based on variance sensitivity-mutual information, and train each sub-network using data from different operating conditions; The test data samples are input into the model, the output results of each sub-network are integrated, and the model output is denormalized to output the predicted value of effluent ammonia nitrogen.

2. The method for detecting ammonia nitrogen in effluent using a self-organizing modular neural network based on a meme algorithm according to claim 1, characterized in that: Normalize the data set X to obtain X; Five water quality variables were selected as the input of the model: influent total phosphorus TP, anaerobic terminal redox potential ORP, aerobic front end dissolved oxygen DO, aerobic terminal total suspended solids TSS and effluent pH.

3. The method for detecting ammonia nitrogen in effluent using a self-organizing modular neural network based on a meme algorithm according to claim 1, characterized in that: The genetic algorithm is used to divide the training samples into data segments including: Randomly generate an initial population containing a set of segmentation breakpoints 1,…,k as chromosomes; The fitness value of each segmentation breakpoint is evaluated through the fitness function; According to the fitness value, the better segmentation breakpoints are selected, and chromosome crossover and gene mutation are performed to generate the next generation population for global search; If the set termination condition is met, the training sample is divided into k data segments with breakpoints of N L =[n1,…,n k+1 ], then each data segment is i=[1,…,k].

4. The method for detecting ammonia nitrogen in effluent using a self-organizing modular neural network based on a meme algorithm according to claim 1, characterized in that: Dividing the data segments into different operating conditions includes: ① Use dynamic time warping algorithm to calculate the similarity of data segments and calculate the local density of data segments; ②Select the data segment with local density greater than the set threshold as the operating condition center; ③ Use the dynamic time warping algorithm to calculate the distance from the data segment to the center of each operating condition, and divide the data segment into the operating condition with the closest distance; ④ After the new data segment is included in the operating condition, the local density of the new data segment is calculated and compared with the local density of the center of the original operating condition, and the data segment with large local density is selected as the operating center of the operating condition; ⑤ Set the operating condition S i Assign it as a subtask to the corresponding sub-network for training and learning.

5. The method for detecting ammonia nitrogen in effluent using a self-organizing modular neural network based on a meme algorithm according to claim 1, characterized in that: A self-organizing sub-network based on variance sensitivity-mutual information is designed, and each sub-network is trained using different operating condition data, including: Design the sub-network topology for the sub-task: the sub-network structure includes an input layer, a hidden layer, and an output layer. The input layer has 5 neuron nodes, the hidden layer has p neuron nodes, and the output layer has 1 neuron node; The sub-network is trained using the gradient descent learning algorithm. During the training process, when the network reaches a local minimum at time t, the learning state of the network is evaluated; If the network error E(t)>E d When E d is the expected error, indicating that the sub-network cannot learn the current sample at this time, so a hidden layer neuron is added; If the network error E(t)≤E d When , the hidden layer neurons with smaller contribution in the sub-network are deleted.

6. The method for detecting ammonia nitrogen in effluent using a self-organizing modular neural network based on a meme algorithm according to claim 5, characterized in that: When E(t)>E d When , the variance sensitive analysis method is used to calculate the contribution of hidden layer neurons to the sub-network output, and the neurons with the largest contribution to the sub-network output are found and added.

7. The method for detecting ammonia nitrogen in effluent using a self-organizing modular neural network based on a meme algorithm according to claim 6, characterized in that: The mutual information algorithm is used to calculate the correlation between neurons in each hidden layer. If neuron i has a small contribution and a small correlation with other neurons, the neuron is deleted. The connection weight of neuron j, which has the greatest correlation with neuron i, is compensated.

8. The method for detecting ammonia nitrogen in effluent water by a self-organizing modular neural network based on a meme algorithm according to claim 1, characterized in that: After the test data samples are input into the model, the test data samples are divided into multiple test operating conditions. The dynamic time warping algorithm is used to calculate the similarity between the test operating conditions and each sub-network, and the sub-network with the greatest similarity among the test operating conditions is activated for testing.

9. A self-organizing modular neural network effluent ammonia nitrogen detection system based on memetic algorithm, characterized in that: include: The data set construction module is used to collect the actual operation data of the urban sewage treatment plant to construct the data set; An operating condition division module is used to divide the data segments of the training samples into a plurality of different operating conditions by using a multivariate segmentation method based on a meme algorithm; Model building module, used to design self-organizing sub-networks based on variance sensitivity-mutual information, and train each sub-network using data from different operating conditions; The test module is used to input the test data samples into the model, integrate the output results of each sub-network, denormalize the model output, and output the predicted value of ammonia nitrogen in the outlet water.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images