CN115758125A - Industrial sewage treatment soft measurement method based on feature structure optimization and deep learning - Google Patents
Industrial sewage treatment soft measurement method based on feature structure optimization and deep learning Download PDFInfo
- Publication number
- CN115758125A CN115758125A CN202211565490.0A CN202211565490A CN115758125A CN 115758125 A CN115758125 A CN 115758125A CN 202211565490 A CN202211565490 A CN 202211565490A CN 115758125 A CN115758125 A CN 115758125A
- Authority
- CN
- China
- Prior art keywords
- feature
- sewage treatment
- industrial sewage
- deep learning
- gru
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Activated Sludge Processes (AREA)
Abstract
The invention discloses an industrial sewage treatment soft measurement method based on feature structure optimization and deep learning; the invention firstly eliminates the overlapping characteristic of the data collected by the industrial sewage treatment plant. In addition, a full set empirical mode decomposition method is adopted to decompose the input characteristic sequence and historical data. The input sequence and the historical data are decomposed into respective intrinsic mode functions, and then the IMFs are subjected to feature screening by using a feature selection Relief F method. Then, the invention applies a new mixed deep learning model to predict four water-yielding indexes, and the model combines a convolution neural network and a gate control circulation unit network to carry out optimization through an attention mechanism. Compared with a soft measurement algorithm without feature optimization, the method provided by the invention has the advantage that the prediction effect is improved.
Description
Technical Field
The invention belongs to the field of water quality soft measurement in control of a sewage treatment system, and particularly relates to an industrial sewage soft measurement method based on feature structure optimization and deep learning.
Background
Industrial wastewater refers to wastewater, sewage and waste liquid produced in industrial production process, and has various kinds, complex components and various toxic substances. With the rapid development of industry, the pollution of industrial wastewater to water bodies is becoming wide and serious, so that the pollution of soil and air is caused, and the health and safety of human beings are threatened. It follows that the treatment of industrial waste water is more important than the treatment of municipal sewage. How to effectively adopt corresponding purification measures to carry out comprehensive utilization and treatment according to pollutant components and concentration in the wastewater is one of the core problems to be solved for ensuring sustainable development. Although an industrial wastewater treatment plant (IETP) can effectively treat wastewater through physical, chemical and biological methods, the treatment effect of the IETP is often weakened due to the fluctuation of interference factors such as the water quality and the water quantity of the wastewater. In addition, the coupling relation between other process variables and a certain index of effluent in a sewage treatment plant in the time domain and the frequency domain is rarely proposed.
Disclosure of Invention
Aiming at the limitation of the existing industrial sewage treatment soft measurement method, the invention provides an industrial sewage treatment soft measurement method based on feature structure optimization and deep learning.
According to the invention, the collected data characteristics in the industrial sewage treatment plant are subjected to Pearson correlation analysis (PCC) to remove the characteristics of information overlapping. In addition, the input signature sequence and historical data are decomposed using a full-set empirical mode decomposition (CEEMD) method. The decomposed IMFs components include main feature information, and some IMFs may represent noise components in the original signal, and these spurious components may become an obstacle to subsequent processing. And selecting main process variables in the IMFs as an input sequence by adopting a Relief (recursive F) feature selection method. Then, a new hybrid deep learning model is appliedCombining a Convolutional Neural Network (CNN) and a gated cyclic unit network (GRU), and finally, optimizing through an Attention Mechanism (AM) to output effluent water qualities CODcr and NH 3 -the strong predictions of N, TN and TP.
The technical scheme adopted by the invention for solving the technical problem comprises the following steps:
step 2, performing empirical mode decomposition on the data in the industrial sewage treatment plant with the overlapped features removed;
step 3, performing characteristic screening on the decomposed IMFs;
step 4, designing a CNN-GRU-AM hybrid network model;
and 5, training and verifying the proposed model network, and outputting a prediction index.
Preferably, the elimination of overlapping features in step 1 is specifically as follows:
before a model is constructed, pearson correlation analysis is carried out on data collected by an industrial sewage treatment plant, and the strength of the correlation indicates whether input information is overlapped; correlation coefficient ρ of signals x (t) and y (t) of PCC xy Is defined as:
in the formula, σ xy Is the covariance, σ, of the signals x (t) and y (t) x ，σ y Standard deviations of the signals x (t) and y (t), respectively; finally, normalizing the calculated correlation coefficient, rho xy ∈[-1，1]，ρ xy The value is more close to 1, the correlation between the two signals is stronger, a threshold value is set, and when rho xy If the input signal is larger than the set threshold value, the input signal is determined as an overlapped signal.
Preferably, the empirical mode decomposition is performed on the features in step 2, specifically as follows:
processing the original data by adopting a complementary integrated empirical mode decomposition method, and eliminating high-frequency noise interference and false components by combining a characteristic screening algorithm to screen out components capable of representing effective information. The complementary integrated empirical mode decomposition not only keeps the good adaptivity of the empirical mode decomposition, but also combines the advantage that the integrated empirical mode decomposition effectively inhibits the mode aliasing phenomenon, and effectively overcomes the residual problem of Gaussian white noise. Therefore, the integral model has better noise reduction performance, and the specific steps are as follows:
adding n positive and negative white noise omega to an original signal s (t) i (t)：
And (2) decomposing the signal after the noise is added again according to the EMD decomposition process to obtain two groups of corresponding intrinsic mode functions IMFs and residual errors r i (t), averaging:
in the formula, C ij + (t)、C ij - (t) denotes the jth IMF component, r i + (t)、r i - (t) represents the i-th calculated residual.
Data acquisition in industrial sewage treatment plants is often in a complex environment, and the acquired data is often mixed with a large amount of noise. The CEEMD decomposition process is to decompose the original signal at different time scales to obtain IMFs components with frequencies from high to low, and the process is an adaptive filtering process similar to quadrature. The IMFs components include main feature information, and some IMFs represent noise components in the original signal, and these spurious components may become an obstacle to subsequent processing. The invention will select the effective IMF component by setting the screening threshold of IMF in the next step.
Preferably, the step 3 of performing feature screening on the decomposed IMFs specifically includes:
performing feature screening on the decomposed IMFs through a Relieff algorithm, specifically: and selecting k nearest neighbor samples from the same type of samples and different types of samples, and averaging the k nearest neighbor samples to obtain each feature weight, thereby obtaining the correlation between each feature and each class in each sample instance. Then, sorting the features according to the weight values, judging whether the features are effective or ineffective by setting a threshold, selecting n features with the maximum weight values, and removing other features to select the features;
the method comprises the following steps:
and D is set as a feature set, m is the repeated sampling times of the samples, delta is a feature statistic index, k is the number of nearest neighbor samples, and T is a statistic index for outputting each feature.
(1) Set 0 all feature weights
(2)For i＝1 to m do
(3) R is sample data arbitrarily extracted from the feature set D; h j (j =1,2.. K) is the k nearest neighbors H found from the same class of sample set of R j (j =1, 2.. K), finding k nearest neighbors M in the feature set inconsistent from each class j (C)(j＝1，2，...k)。
(4)for i＝1 to N do
The statistics for a feature are labeled as:
where p (C) is the proportion of the class. p (Class (R)) is the proportion of the Class of a sample chosen at random.
(6) Ordering W
After the feature weight is calculated, the larger the weight is, the stronger the distinguishing capability of the feature on the sample is, and a new feature subset is selected by setting a threshold.
Preferably, the CNN-GRU-AM network model designed in step 4 is specifically as follows:
the convolutional neural network CNN structure comprises a convolutional layer, a pooling layer, a full-link layer and an output layer; the CNN model uses gradient descent method training parameters, and the trained model can learn features in the time series data.
The gated loop unit GRU neural network comprises an updating gate and a resetting gate; the data propagation process in the GRU unit is described as follows: first, the evaluation state h is transmitted by the last one t-1 And input x of the current node t To obtain a reset gate R t And a refresh door Z t The state of (2):
r t ＝σ(x t W xr +h t-1 W hr +b r )
z t ＝σ(x t W xr +h t-1 W hr +b z )
then, at present
The current time state memorized in the set is represented as:
the GRU model then updates the state by:
finally, the output of the forward propagation is:
y t ＝σ(W o ·h t )
wherein h is t Is the output of the GRU, W is the weight vector, b is the offset vector of the GRU.
By last transmitted state H t-1 And input x of the current node t σ is a sigmoid function by which data is transformed into a value in the range of 0-1 to serve as a gate signal. Wherein H t-1 Containing past information, indicates a Hadamard product.
Note that the implementation steps of the mechanism AM are shown below.
(1) Calculating a given h j Value and target state S t-1 Degree of similarity, i.e. state h at each time t j The weight of (c):
e tj ＝a(S t-1 ，h j )
(2) Normalizing the weight coefficients:
(3) For state h j Carrying out weighted average:
where T is the total number of time steps in the input sequence.
The invention has the beneficial effects that:
1. the invention provides a characteristic structure optimization method aiming at data of an industrial sewage treatment plant, and compared with a soft measurement algorithm without characteristic optimization, the method provided by the invention improves the prediction effect.
2. In order to evaluate the prediction accuracy of the proposed CEEMD-Relieff-CNN-GRU-AM model, the effluent CODcr and NH of a real industrial sewage treatment plant are used 3 And the efficiency and the stability of the proposed hybrid model on an actual data set are evaluated through a comparative experiment by taking-N, TN and TP as examples, so that the prediction effect is greatly improved.
Drawings
FIG. 1 is a flow chart of the inventive soft measurement method for industrial wastewater treatment with feature structure optimization and deep learning;
FIG. 2-1 is a diagram of the prediction of effluent COD by a trained model in an embodiment in accordance with the invention cr Evaluation schematic of the results;
FIGS. 2-2 are graphs of the NH of water predicted by a trained model in an embodiment in accordance with the invention 3 -evaluation schematic of N results;
FIGS. 2-3 are schematic diagrams illustrating the evaluation of the results of the trained model to predict water-out TN in accordance with an embodiment of the present invention;
fig. 2-4 are schematic diagrams illustrating the evaluation of the water TP outcome predicted by the trained model in accordance with the exemplary embodiment of the present invention.
Detailed Description
The following describes the implementation steps of the present invention in further detail with reference to fig. 1.
A soft measurement method for industrial sewage treatment based on feature structure optimization and deep learning specifically comprises the following steps:
before the model is constructed, pearson correlation analysis (PCC) is carried out on data collected by an industrial sewage treatment plant, and the strength of the PCC indicates whether input information is overlapped; correlation coefficient ρ of signals x (t) and y (t) of PCC xy Is defined as:
in the formula, σ xy Is the covariance, σ, of the signals x (t) and y (t) x ，σ y Standard deviation of the signals x (t) and y (t), respectively; finally, normalizing the calculated correlation coefficient, wherein rho is xy ∈[-1，1]，ρ xy The value is more close to 1, the correlation between the two signals is stronger, a threshold value is set, and when rho xy If the input signal is larger than the set threshold value, the input signal is determined as an overlapped signal.
Step 2, performing empirical mode decomposition on the characteristics, specifically as follows:
because data in the industrial sewage treatment plant has high nonlinearity and non-stationarity and is mixed with uncertain noise, the method adopts a complementary integrated empirical mode decomposition (CEEMD) method to process original data, and combines a characteristic screening algorithm to remove high-frequency noise interference and false components and screen out components capable of representing effective information. The complementary integrated empirical mode decomposition (CEEMD) not only keeps the good adaptivity of the empirical mode decomposition, but also combines the advantage that the integrated empirical mode decomposition effectively inhibits the mode aliasing phenomenon, and effectively overcomes the residual problem of Gaussian white noise. Therefore, the integral model has better noise reduction performance, and the algorithm is as follows:
adding n positive and negative white noise omega to an original signal s (t) i (t)：
And (2) decomposing the signal after the noise is added again according to the EMD decomposition process to obtain two groups of corresponding intrinsic mode functions IMFs and residual errors r i (t), averaging:
in the formula, C ij + (t)、C ij - (t) denotes the jth IMF component, r i + (t)、r i - (t) represents the i-th calculated residual.
Data acquisition in industrial sewage treatment plants is often in a complex environment, and the acquired data is often mixed with a large amount of noise. The CEEMD decomposition process is to decompose the original signal at different time scales to obtain IMFs components with frequencies from high to low, and the process is an adaptive filtering process similar to quadrature. The IMFs components include main feature information, and some IMFs represent noise components in the original signal, and these spurious components may become an obstacle to subsequent processing. The invention will select the effective IMF component by setting the IMF screening threshold in the next step.
And 3, performing feature screening on the decomposed IMFs, specifically comprising the following steps:
the basic idea of the Relief F algorithm is to give a weight to each feature of a sample, iteratively update the weight, sort corresponding features according to the weight, and select a feature subset according to the order, so that good features gather similar samples and disperse heterogeneous samples. It can solve multi-class problems and regression problems, and can handle noisy, incomplete features, and multi-class attribute datasets. The idea of the algorithm is that k nearest neighbor samples are selected from the same type of samples and different types of samples, and the average value of the samples is calculated to obtain each feature weight, so that the correlation between each feature and the type in each sample instance is obtained; then, sorting the features according to the weight values, judging whether the features are effective or invalid by setting a threshold, selecting n features with the maximum weight values, and removing other features to select the features;
the method comprises the following steps:
d is set as a feature set, m is the repeated sampling times of the samples, delta is a feature statistic index, k is the number of nearest neighbor samples, and T is a statistic index for outputting each feature;
(1) Set 0 all feature weights
(2)For i＝1 to m do
(3) R is sample data arbitrarily extracted from the feature set D; h j For k nearest neighbors H found from the same class sample set of R j Finding k nearest neighbors M in a feature set inconsistent from each class j (C) Wherein j =1, 2.. K;
(4)for i＝1 to N do
the statistic indicators for a feature are:
wherein p (C) is the proportion of the class; p (Class (R)) is the proportion of the Class of a certain sample selected at random;
(6) Ordering W
After the feature weight is calculated, the larger the weight is, the stronger the distinguishing capability of the feature on the sample is, and a new feature subset can be selected by setting a threshold.
Step 4, designing a CNN-GRU-AM network model, which comprises the following specific steps:
the CNN-GRU-AM hybrid model has the following modules: the Convolutional Neural Network (CNN) is a multi-layer feedforward artificial neural network, has the characteristics of pooling operation, local connection, weight sharing and the like, and is widely used in different fields. CNN has two significant advantages: sparse connection and weight sharing can extract deeper features from data and reduce the complexity of a network model. The CNN structure mainly comprises a convolution layer, a pooling layer, a full-connection layer, an output layer and the like. The structure can reduce the number of the weights, so that a network model is simple, and meanwhile, the time sequence data can be directly used as the input of the network, thereby effectively reducing the complexity of feature extraction and data reconstruction. The CNN model uses gradient descent method training parameters, and the trained model can learn features in the time series data.
Gated round-robin unit (GRU) neural networks are a variant of LSTM optimization refinement that reduces training parameters while ensuring prediction accuracy. The GRU comprises an update gate and a reset gate, has fewer structural parameters compared with three gates of LSTM, has higher convergence rate and has good data characteristic learning capability. The data propagation process in the GRU unit is described as follows: first, the evaluation state h is transmitted by the previous one t-1 And input x of the current node t To obtain a reset gate R t And a refresh door Z t The state of (2):
r t ＝σ(x t W xr +h t-1 W hr +b r )
z t ＝σ(x t W xr +h t-1 W hr +b z )
then, at present
The current time state memorized in the set is represented as:
the GRU model then updates the state by:
finally, the forward propagated output is:
y t ＝σ(W o ·h t )
wherein h is t Is the output of the GRU, W is the weight vector, b is the offset vector of the GRU;
by last transmitted state H t-1 And input x of the current node t σ is a sigmoid function by which data is transformed into a value in the range of 0-1 to serve as a gate signal; wherein H t-1 Contains past information, which indicates a Hadamard product;
the implementation steps of the attention mechanism AM are shown in the following formula;
(1) Calculating a given h j Value and target state S t-1 Similarity, i.e. state h at each time t j The weight of (c):
e tj ＝a(S t-1 ，h j )
(2) Normalizing the weight coefficients:
(3) For state h j Carrying out weighted average:
where T is the total number of time steps in the input sequence.
11532 pieces of real data of the industrial sewage treatment plant collected at the industrial sewage treatment plant are divided into a training set of 70% and a test set of 30%. The window length of each sample point of the prediction model herein is set to 10, the structure of each module includes a GRU layer having 32 nodes, and a DNN layer having 16 nodes, and the epochs size is 100. This section will evaluate the proposed model by three experiments.
Specifically, the invention will be described by Mean Absolute Error (MAE), mean Absolute Percent Error (MAPE), R 2 The prediction model was evaluated with sclore and Mean Absolute Percent Error (MAPE) and compared to the comparative experimental model.
Wherein N is the length of the time series,
indicates the predicted value, y t The actual value is represented by the value of,
is the average of the true values. The indexes are selected to estimate the proposed model, wherein the basic method for estimating the predicted value and the true value is MAE and RMSE, and the smaller the numerical value of the predicted value and the true value is after calculation, the smaller the error value representing the prediction is; the MAPE not only considers the error of the predicted value and the true value, but also considers the proportion of the error and the true value; r 2 The range of the _scoreis between 0 and 1, and the closer to 1, the closer to the real value the predicted value of the representation model is; as shown in fig. 2-1, 2-2, 2-3, 2-4.
Claims (5)
1. The industrial sewage treatment soft measurement method based on feature structure optimization and deep learning is characterized by comprising the following steps of:
step 1, removing overlapping characteristics aiming at data collected by an industrial sewage treatment plant;
step 2, performing empirical mode decomposition on the data in the industrial sewage treatment plant with the overlapping features removed;
step 3, performing characteristic screening on the decomposed IMFs;
step 4, designing a CNN-GRU-AM hybrid network model;
and 5, training and verifying the proposed model network, and outputting a prediction index.
2. The industrial sewage treatment soft measurement method based on feature structure optimization and deep learning according to claim 1, wherein the elimination of overlapping features in step 1 is specifically as follows:
before a model is constructed, pearson correlation analysis is carried out on data collected by an industrial sewage treatment plant, and the strength of the correlation indicates whether input information is overlapped; correlation coefficient ρ of signals x (t) and y (t) of PCC xy Is defined as:
in the formula, σ xy Is the covariance, σ, of the signals x (t) and y (t) x ,σ y Standard deviations of the signals x (t) and y (t), respectively; finally, normalizing the calculated correlation coefficient, rho xy ∈[-1,1]，ρ xy The value is more close to 1, the correlation between the two signals is stronger, a threshold value is set, and when rho xy If the input signal is larger than the set threshold value, the input signal is determined as an overlapping signal.
3. The industrial sewage treatment soft measurement method based on feature structure optimization and deep learning according to claim 1, wherein the empirical mode decomposition is performed on the features in the step 2, and specifically the following steps are performed:
processing original data by adopting a complementary integrated empirical mode decomposition method, and eliminating high-frequency noise interference and false components by combining a characteristic screening algorithm to screen out components capable of representing effective information; the method comprises the following specific steps:
adding n positive and negative white noise omega to an original signal s (t) i (t):
And (2) decomposing the signal added with the noise again according to an EMD decomposition process to obtain two groups of corresponding intrinsic mode functions IMFs and residual errors r i (t), averaging:
in the formula, C ij + (t)、C ij - (t) denotes the jth IMF component, r i + (t)、r i - (t) represents the i-th calculated residual.
4. The method for soft measurement of industrial sewage treatment based on feature structure optimization and deep learning according to claim 3, wherein the step 3 of feature screening the decomposed IMFs specifically comprises the following steps:
and (3) performing feature screening on the decomposed IMFs through a Relief (iterative regression) algorithm, specifically: selecting k nearest neighbor samples from the same type of samples and different types of samples, and averaging the k nearest neighbor samples to obtain each feature weight, thereby obtaining the correlation between each feature and each class in each sample instance; then, sorting the features according to the weight values, judging whether the features are effective or invalid by setting a threshold, selecting n features with the maximum weight values, and removing other features to select the features;
the method comprises the following steps:
d is set as a feature set, m is the repeated sampling times of the samples, delta is a feature statistic index, k is the number of nearest neighbor samples, and T is a statistic index for outputting each feature;
(1) Set 0 all feature weights
(2)For i＝1 to m do
(3) R is sample data arbitrarily extracted from the feature set D; h j For k nearest neighbors H found from the same class sample set of R j Finding k nearest neighbors M in feature sets inconsistent from each class j (C) Wherein j =1,2, \8230k;
(4)for i＝1 to N do
the statistics for a feature are labeled as:
wherein p (C) is the proportion of the class; p (Class (R)) is the proportion of the Class of a certain sample selected at random;
(6) Ordering W
After the feature weight is calculated, the larger the weight is, the stronger the distinguishing capability of the feature on the sample is, and a new feature subset is selected by setting a threshold.
5. The industrial sewage treatment soft measurement method based on feature structure optimization and deep learning of claim 4, wherein the method comprises the following steps: the CNN-GRU-AM network model designed in the step 4 is specifically as follows:
the convolutional neural network CNN structure comprises a convolutional layer, a pooling layer, a full-link layer and an output layer; the CNN model trains parameters by using a gradient descent method, and the trained model can learn characteristics in time series data;
the gated loop unit GRU neural network comprises an updating gate and a resetting gate; the data propagation process in the GRU unit is described as follows: first, the evaluation state h is transmitted by the last one t-1 And input x of the current node t To obtain a reset gate R t And a refresh door Z t The state of (2):
r t ＝σ(x t W xr +h t-1 W hr +b r )
z t ＝σ(x t W xr +h t-1 W hr +b z )
then, at present
The current time state memorized in the set is expressed as:
the GRU model then updates the state by:
finally, the output of the forward propagation is:
y t ＝σ(W o ·h t )
wherein h is t Is the output of the GRU, W is the weight vector, b is the offset vector of the GRU;
by last transmitted state H t-1 And input x of the current node t σ is a sigmoid function by which data is transformed into a value in the range of 0-1 to serve as a gate signal; wherein H t-1 Contains past information, which indicates a Hadamard product;
the implementation steps of the attention mechanism AM are shown in the following formula;
(1) Calculating a given h j Value and target state S t-1 Degree of similarity, i.e. state h at each time t j The weight of (c):
e tj ＝a(S t-1 ，h j )
(2) Normalizing the weight coefficients:
(3) For state h j Carrying out weighted average:
where T is the total number of time steps in the input sequence.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211565490.0A CN115758125A (en) | 2022-12-07 | 2022-12-07 | Industrial sewage treatment soft measurement method based on feature structure optimization and deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211565490.0A CN115758125A (en) | 2022-12-07 | 2022-12-07 | Industrial sewage treatment soft measurement method based on feature structure optimization and deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115758125A true CN115758125A (en) | 2023-03-07 |
Family
ID=85344169
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211565490.0A Pending CN115758125A (en) | 2022-12-07 | 2022-12-07 | Industrial sewage treatment soft measurement method based on feature structure optimization and deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115758125A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116360366A (en) * | 2023-03-24 | 2023-06-30 | 淮阴工学院 | Sewage treatment process optimization control method |
| CN117371873A (en) * | 2023-12-01 | 2024-01-09 | 四川省生态环境科学研究院 | Sewage assessment method for environmental protection engineering based on big data |
| CN117874583A (en) * | 2024-03-13 | 2024-04-12 | 广州华科环保工程有限公司 | Odor treatment method and system generated by sewage treatment |
| CN117923657A (en) * | 2024-03-22 | 2024-04-26 | 湖南孚瑞锑格机械设备有限公司 | Wastewater treatment method and system based on anaerobic ammonia oxidation reactor |
-
2022
- 2022-12-07 CN CN202211565490.0A patent/CN115758125A/en active Pending
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116360366A (en) * | 2023-03-24 | 2023-06-30 | 淮阴工学院 | Sewage treatment process optimization control method |
| CN116360366B (en) * | 2023-03-24 | 2023-12-01 | 淮阴工学院 | Sewage treatment process optimization control method |
| CN117371873A (en) * | 2023-12-01 | 2024-01-09 | 四川省生态环境科学研究院 | Sewage assessment method for environmental protection engineering based on big data |
| CN117371873B (en) * | 2023-12-01 | 2024-03-26 | 四川省生态环境科学研究院 | Sewage assessment method for environmental protection engineering based on big data |
| CN117874583A (en) * | 2024-03-13 | 2024-04-12 | 广州华科环保工程有限公司 | Odor treatment method and system generated by sewage treatment |
| CN117874583B (en) * | 2024-03-13 | 2024-06-07 | 广州华科环保工程有限公司 | Odor treatment method and system generated by sewage treatment |
| CN117923657A (en) * | 2024-03-22 | 2024-04-26 | 湖南孚瑞锑格机械设备有限公司 | Wastewater treatment method and system based on anaerobic ammonia oxidation reactor |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115758125A (en) | Industrial sewage treatment soft measurement method based on feature structure optimization and deep learning | |
| CN107358021B (en) | DO prediction model establishment method based on BP neural network optimization | |
| CN111354423B (en) | Method for predicting ammonia nitrogen concentration of effluent of self-organizing recursive fuzzy neural network based on multivariate time series analysis | |
| CN102854296B (en) | Sewage-disposal soft measurement method on basis of integrated neural network | |
| Sivapragasam et al. | Prediction of algal blooms using genetic programming | |
| CN107292446B (en) | Hybrid wind speed prediction method based on component relevance wavelet decomposition | |
| CN109344971B (en) | Effluent ammonia nitrogen concentration prediction method based on adaptive recursive fuzzy neural network | |
| CN110672804B (en) | Intelligent detection method for typical organic pollutants in urban river water body | |
| CN107992447A (en) | A kind of feature selecting decomposition method applied to river level prediction data | |
| CN103793604A (en) | Sewage treatment soft measuring method based on RVM | |
| CN106600001A (en) | Glass furnace temperature forecast method based on learning machine related to Gaussian mixture distribution | |
| CN113539386A (en) | CLMVO-ELM-based dissolved oxygen concentration prediction method, device, equipment and storage medium | |
| Qiao et al. | Design of modeling error PDF based fuzzy neural network for effluent ammonia nitrogen prediction | |
| CN115481784A (en) | Traffic flow prediction method based on improved combination model and application | |
| CN115147645A (en) | Membrane module membrane pollution detection method based on multi-feature information fusion | |
| Yang et al. | A multi-factor forecasting model for carbon emissions based on decomposition and swarm intelligence optimization | |
| Cao et al. | Adopting improved Adam optimizer to train dendritic neuron model for water quality prediction | |
| Du Nguyen et al. | The use of machine learning algorithms for evaluating water quality index: a survey and perspective | |
| CN113762078A (en) | Lake TN prediction method based on VMD-CSSA-LSTM-MLR combined model | |
| Sulandari et al. | Forecasting trend data using a hybrid simple moving average-weighted fuzzy time series model | |
| CN115775042B (en) | Crop heavy metal enrichment risk prediction method and system based on Bayesian theory | |
| CN117252285A (en) | Multi-index sewage water quality prediction method based on parallel CNN-GRU network | |
| CN116205135A (en) | SO based on data decomposition and neural network 2 Emission prediction method and system | |
| CN112862190B (en) | Water quality time sequence prediction method based on improved Seq2Seq frame | |
| Xu et al. | Prediction of the Wastewater's pH Based on Deep Learning Incorporating Sliding Windows. |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |