CN112070156A - Gas emission concentration prediction method and system based on GRU network - Google Patents
Gas emission concentration prediction method and system based on GRU network Download PDFInfo
- Publication number
- CN112070156A CN112070156A CN202010928729.0A CN202010928729A CN112070156A CN 112070156 A CN112070156 A CN 112070156A CN 202010928729 A CN202010928729 A CN 202010928729A CN 112070156 A CN112070156 A CN 112070156A
- Authority
- CN
- China
- Prior art keywords
- value
- data set
- network
- parameter data
- training
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 82
- 238000012549 training Methods 0.000 claims abstract description 75
- 239000013598 vector Substances 0.000 claims abstract description 22
- 238000012360 testing method Methods 0.000 claims abstract description 18
- 230000002159 abnormal effect Effects 0.000 claims description 53
- 238000007689 inspection Methods 0.000 claims description 36
- 238000001914 filtration Methods 0.000 claims description 24
- 238000012545 processing Methods 0.000 claims description 18
- 238000010606 normalization Methods 0.000 claims description 15
- 238000001514 detection method Methods 0.000 claims description 11
- 238000012163 sequencing technique Methods 0.000 claims description 11
- 238000004364 calculation method Methods 0.000 claims description 10
- 238000013528 artificial neural network Methods 0.000 claims description 9
- 230000000306 recurrent effect Effects 0.000 claims description 3
- 238000011425 standardization method Methods 0.000 claims description 3
- 239000007789 gas Substances 0.000 description 31
- 230000008569 process Effects 0.000 description 8
- 230000006870 function Effects 0.000 description 7
- UGFAIRIUMAVXCW-UHFFFAOYSA-N Carbon monoxide Chemical compound [O+]#[C-] UGFAIRIUMAVXCW-UHFFFAOYSA-N 0.000 description 6
- 239000003546 flue gas Substances 0.000 description 6
- 239000000243 solution Substances 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 238000005259 measurement Methods 0.000 description 5
- 238000005457 optimization Methods 0.000 description 5
- 238000006467 substitution reaction Methods 0.000 description 4
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 3
- QGZKDVFQNNGYKY-UHFFFAOYSA-N Ammonia Chemical compound N QGZKDVFQNNGYKY-UHFFFAOYSA-N 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000003111 delayed effect Effects 0.000 description 2
- 238000012217 deletion Methods 0.000 description 2
- 230000037430 deletion Effects 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 229910021529 ammonia Inorganic materials 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 description 1
- 238000002485 combustion reaction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 239000003344 environmental pollutant Substances 0.000 description 1
- 230000014509 gene expression Effects 0.000 description 1
- 238000002347 injection Methods 0.000 description 1
- 239000007924 injection Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000001301 oxygen Substances 0.000 description 1
- 229910052760 oxygen Inorganic materials 0.000 description 1
- 231100000719 pollutant Toxicity 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Tourism & Hospitality (AREA)
- Development Economics (AREA)
- Educational Administration (AREA)
- Probability & Statistics with Applications (AREA)
- Health & Medical Sciences (AREA)
- Economics (AREA)
- General Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application discloses a gas emission concentration prediction method and a gas emission concentration prediction system of a GRU network, wherein the method comprises the steps of obtaining a first parameter data set for predicting the gas emission concentration; performing feature selection by using the first parameter data set to obtain a plurality of feature vectors for inputting to a GRU network for network training; dividing a first parameter data set into a training set and a test set according to the obtained multiple feature vectors; taking the training set as the input of the GRU network to be trained, training the GRU network, and outputting a pre-training network until an iterative convergence condition is met; and taking the test set as the input of the pre-training network, and obtaining a trained gas emission concentration prediction model by measuring the deviation between a predicted value output by the pre-training network and a given actual value when the deviation value is smaller than a preset deviation threshold value.
Description
Technical Field
The application relates to the technical field of environmental engineering detection, in particular to a gas emission concentration prediction method and system based on a GRU network.
Background
With the increasingly prominent problem of atmospheric pollution, the emission of atmospheric pollutants of thermal power generating units in China is strictly regulated. The traditional low NOx combustion control is difficult to meet the emission standard, and the SCR flue gas denitration technology is widely adopted at home and abroad by matching with a tail flue gas denitration device.
In the current implementation scheme, a PCA method is utilized to extract the characteristics of original data, and a NOx emission prediction model of a denitration system is established by an RNN neural network. However, principal components subjected to PCA dimension reduction analysis are fuzzy when explaining the meanings of the principal components, and cannot be clear and accurate as those described in original variables; meanwhile, the data size after dimension reduction is small, so that the reference significance in practical industrial application is not large.
The traditional artificial neural network RBF model has high spatial complexity, and meanwhile, the prediction result does not meet the requirement of practical application. The decision coefficient R2 of the LSSVM model is better than that of the RBF model, but the root mean square error and the average absolute error of the LSSVM model are higher than those of the RBF model, so that the stability of predicting the gas emission concentration by using the LSSVM model is insufficient.
Disclosure of Invention
The application provides a gas emission concentration prediction method and system based on a GRU network, and the advantages of the GRU network in model accuracy, generalization capability and training speed are utilized, so that the conditions of lag and inaccurate measured value in gas concentration measurement are avoided, and the prediction accuracy is effectively improved.
In view of the above, a first aspect of the present application provides a method for predicting a gas emission concentration from a GRU network, the method comprising:
acquiring a first parameter data set for predicting gas emission concentration;
performing feature selection on the first parameter data set to obtain a plurality of feature vectors for inputting to a GRU network for network training;
dividing the first parameter data set into a training set and a test set according to the obtained multiple feature vectors;
taking a training set as the input of the GRU network, training the GRU network, and outputting a pre-training network until an iterative convergence condition is met;
and taking the test set as the input of the pre-training network, calculating the deviation between a predicted value output by the pre-training network and a given actual value, and if the deviation value is smaller than a preset deviation threshold value, obtaining a gas emission concentration prediction model.
Optionally, before performing feature selection on the first parameter data set, the method further includes:
filtering abnormal operation data from the first parameter data set by adopting a singular value inspection method or a threshold inspection method, wherein the abnormal operation data filtering by adopting the singular value inspection method comprises the following steps:
performing a singular value check on items of data included in the first parameter data set;
defining an abnormal value according to a singular value inspection result;
removing abnormal values from the first parameter dataset or defining abnormal values as missing values and performing completion of the missing values;
the threshold value checking method comprises the step of selecting abnormal operation data with parameter values lower than a preset detection threshold value from the first parameter data set.
Optionally, the filtering, by using a singular value inspection method or a threshold inspection method, the abnormal operation data from the first parameter data set, and then further includes:
and adopting a Z-score data standardization method to carry out standardization processing on the second parameter data set with the filtered abnormal operation data.
Optionally, the feature selection performed by the first parameter data set specifically includes:
the characteristic selection of the normalized second parameter data set by using a mutual information characteristic selection algorithm comprises the following steps:
combining the data included in the second parameter data set in pairs in sequence to obtain a plurality of combination sequences, and calculating mutual information among elements aiming at each combination sequence;
sequencing a plurality of items of mutual information according to the value of the mutual information to obtain a mutual information sequencing sequence;
selecting top k elements from the mutual information sorting sequence, selecting corresponding data items from the second parameter data set based on the selected k elements, and using the corresponding data items as the input of the GRU network; k is 1 or more.
Optionally, the normalizing the second parameter data set by using the Z-score data normalization method specifically includes:
calculating the mean value of the overall data in the second parameter data set by using the formula (1)
Calculating to obtain a standard deviation s of the overall data in the second parameter data set by using a formula (2);
combining the mean value and standard deviation of the overall data, and obtaining a final standardized processing result Z by using a formula (3)i：
Where n denotes the total number of feature vectors in the second parameter dataset, xiAnd representing the mean value corresponding to the ith feature vector.
Optionally, the hidden layer of the GRU network builds a 4-layer recurrent neural network by using GRU cells, wherein:
the nodes contained in each layer are respectively: 128. 128, 64 and 32.
Optionally, the calculating a deviation between the predicted value output by the pre-training network and the given actual value specifically includes:
after a first deviation value, a second deviation value and a third deviation value between the predicted value and the given actual value are respectively calculated through a root mean square error calculation formula, an average absolute value error calculation formula and a decision coefficient calculation formula, the first deviation value, the second deviation value and the third deviation value are averaged or accumulated to obtain the deviation between the predicted value and the given actual value.
A second aspect of the present application provides a gas emission concentration prediction system by a GRU network, the system comprising:
a first parameter data set acquisition unit for acquiring a first parameter data set for predicting a gas emission concentration;
the characteristic selection unit is used for carrying out characteristic selection on the first parameter data set by utilizing a mutual information characteristic selection algorithm to obtain a plurality of characteristic vectors for inputting to a GRU network for network training; the characteristic selection by utilizing a mutual information characteristic selection algorithm comprises the following steps:
combining the data included in the first parameter data set in pairs in sequence to obtain a plurality of combination sequences, and calculating mutual information among elements aiming at each combination sequence;
sequencing a plurality of items of mutual information according to the value of the mutual information to obtain a mutual information sequencing sequence;
selecting top k elements from the mutual information sorting sequence, selecting corresponding data items from the first parameter data set based on the selected k elements, and using the corresponding data items as the input of the GRU network; k is greater than or equal to 1;
the data dividing unit is used for dividing the first parameter data set into a training set and a test set according to the obtained multiple feature vectors;
the GRU network training unit is used for taking a training set as the input of the GRU network, training the GRU network and outputting a pre-training network until an iteration convergence condition is met;
and the prediction result output unit is used for taking the test set as the input of the pre-training network, calculating the deviation between a predicted value output by the pre-training network and a given actual value, and obtaining a gas emission concentration prediction model if the deviation value is smaller than a preset deviation threshold value.
Optionally, the system further comprises a data filtering unit; wherein:
the data filtering unit is used for filtering abnormal operation data from the first parameter data set by adopting a singular value inspection method or a threshold inspection method; when the abnormal operation data filtering is carried out by adopting a singular value inspection method, the method comprises the following steps:
performing a singular value check on items of data included in the first parameter data set;
defining an abnormal value according to a singular value inspection result;
removing abnormal values from the first parameter dataset or defining abnormal values as missing values and performing completion of the missing values;
when the threshold value checking method is adopted, abnormal operation data with a parameter value lower than a detection threshold value are filtered from the first parameter data set by setting the detection threshold value.
Optionally, the system further comprises a normalization processing unit, wherein:
the normalization processing unit is configured to, after filtering the abnormal operation data from the first parameter data set by using the singular value inspection method or the threshold inspection method, perform normalization processing on the second parameter data set with the filtered abnormal operation data by using a Z-score data normalization method.
The application provides a gas emission concentration prediction method and a gas emission concentration prediction system of a GRU network, and the method comprises the steps of obtaining a first parameter data set for predicting the gas emission concentration; performing feature selection by using the first parameter data set to obtain a plurality of feature vectors for inputting to a GRU network for network training; dividing a first parameter data set into a training set and a test set according to the obtained multiple feature vectors; taking the training set as the input of the GRU network to be trained, training the GRU network, and outputting a pre-training network until an iterative convergence condition is met; and taking the test set as the input of the pre-training network, and obtaining a trained gas emission concentration prediction model by measuring the deviation between a predicted value output by the pre-training network and a given actual value when the deviation value is smaller than a preset deviation threshold value.
The method and the device utilize the advantages of the GRU network in model precision, generalization capability and training speed, and avoid the situation that gas concentration measurement lags; in order to improve the accuracy of the measured value, filtering abnormal operation data is performed in sequence before feature selection is performed on the acquired data set, and the filtered data set is subjected to standardized processing, so that the acquired data meet the actual field requirements, and the prediction precision is effectively improved; the GRU neural network structure with 4 layers is adopted, the space complexity of the network is reduced, and the requirement of practical application is met.
Drawings
FIG. 1 is a method flow diagram of one embodiment of a method of predicting a concentration of a gas emission by a GRU network of the present application;
FIG. 2 is a method flow diagram of a second embodiment of a method of predicting gas emission concentration from a GRU network of the present application;
fig. 3 is a schematic structural diagram of a first embodiment of a gas emission concentration prediction system of a GRU network according to the present application.
Detailed Description
In order to make the technical solutions of the present application better understood, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments obtained by a person of ordinary skill in the art from the embodiments in the present application without any creative effort belong to the protection scope of the present application.
Example 1:
referring to fig. 1, fig. 1 is a flowchart illustrating a method for predicting a gas emission concentration from a GRU network according to an embodiment of the present application, as shown in fig. 1, specifically:
s100, acquiring a first parameter data set for predicting gas emission concentration;
s200, performing feature selection by using the first parameter data set to obtain a plurality of feature vectors for inputting to a GRU network for network training;
wherein, to SCR flue gas denitration system NOXFor example, when data acquisition in an earlier stage is performed, an existing flue gas pollution source emission process (working condition) monitoring system is used for monitoring a denitration system in real time, and then data samples of parameters related to the training of a GRU network model are acquired from a power station DCS, wherein the data samples comprise parameter variables such as ammonia injection mass flow, boiler load, SCR inlet flue gas temperature, SCR inlet flue gas oxygen content, SCR inlet NOx mass concentration and SCR outlet NOx concentration.
Specifically, when feature selection is performed on the first parameter data set, a mutual information feature selection algorithm is considered, and measurement of interdependence between variables is performed based on mutual information or transfer information between different random variables.
S300, carrying out data division according to the obtained multiple characteristic vectors to construct a training set and a test set;
it should be noted that, the training set and the test set are constructed by including sample set time series data trained by the NOx emission prediction model of the NOx removal system according to the data structure required by the network model.
S400, taking the training set as the input of the GRU network, training the GRU network, and outputting a pre-training network until an iteration convergence condition is met. In the training process, when the loss function value meets the iterative convergence condition, a pre-training network is output; specifically, the network prediction adopts an iterative point-by-point prediction method.
In order to reduce the spatial complexity of the network and meet the requirements of practical application, the hidden layer of the GRU network in this embodiment adopts GRU cells, and a 4-layer recurrent neural network is built, wherein:
the nodes contained in each layer are respectively: 128. 128, 64 and 32.
It should be noted that deep learning often requires a lot of time and computer resources for training, which is also a significant reason for the development of deep learning algorithms. Although learning of the model can be accelerated by adopting distributed parallel training, the required computing resources are not reduced. Based on the fact that the learning speed and the learning effect of the machine can be fundamentally accelerated only by an optimization algorithm which needs fewer resources and enables the model to be converged faster, an Adam optimization algorithm is adopted when model optimization is conducted, the Adam optimization algorithm is an extended form of a random gradient descent algorithm, and the Adam optimization algorithm can update the weight of the neural network based on iteration of training data.
When the Loss function value is judged to meet the iterative convergence condition, the output value and the input value of each unit of the hidden layer are generally calculated first, then the output value of each unit of the output layer is calculated, and finally the Loss function Loss value is calculated, wherein:
the loss function is used for measuring the inconsistency degree between the predicted value f (x) and the true value Y of the network model, is a non-negative real value function and is generally expressed by L (Y, f (x)), and the smaller the loss function is, the better the robustness of the model is.
And when the Loss value does not meet the iteration convergence condition, adjusting the connection weight of each layer of the network. In the iteration process, by continuously updating the training parameters, when the iteration is completed, namely when the loss function value meets the iteration convergence condition, the convergent GRU pre-training network is obtained.
S500, taking the test set as the input of a pre-training network, calculating the deviation between a predicted value output by the pre-training network and a given actual value, and if the deviation value is smaller than a preset deviation threshold value, obtaining a gas emission concentration prediction model; and under the condition that the deviation value is greater than a preset deviation threshold value, returning to the GRU network iterative training process until the pre-training network meeting the preset deviation threshold value is output.
Specifically, when measuring the deviation between the predicted value and the given actual value, in this embodiment, after the first deviation value, the second deviation value, and the third deviation value between the predicted value and the given actual value are respectively calculated by the root mean square error calculation formula, the average absolute value error calculation formula, and the decision coefficient calculation formula, the first deviation value, the second deviation value, and the third deviation value are averaged or accumulated to obtain the deviation between the predicted value and the given actual value.
It should be noted that, when the actual deviation solution is performed, the actual deviation value may be determined according to any one of the first, second, or third deviation values, or the actual deviation may be calculated by averaging the first, second, and third deviation values or by using any other index that reflects the trend in the data set.
In the method for predicting the gas emission concentration by the GRU network disclosed in this embodiment 1, the advantages of the GRU network in terms of model accuracy, generalization ability and training speed are utilized to avoid the occurrence of the condition that the gas concentration measurement is delayed; the GRU neural network structure with 4 layers is adopted, the space complexity of the network is reduced, and the requirement of practical application is met.
Example 2:
in order to make the input training data and the test data more close to the actual field requirement and improve the prediction accuracy, please refer to fig. 2, which is a flowchart of a gas emission concentration prediction method of a GRU network according to a second embodiment of the present application, in which, on the basis of embodiment 1, step S200 includes, before performing feature selection on the first parameter data set:
s210, filtering abnormal operation data from the first parameter data set by adopting a singular value inspection method or a threshold value inspection method;
specifically, when the singular value inspection method is used for filtering abnormal operation data, the method comprises the following steps:
performing singular value inspection on each item of data included in the first parameter data set;
defining an abnormal value according to a singular value inspection result;
removing abnormal values from the first parameter dataset, or defining abnormal values as missing values and performing completion of the missing values; when removing the abnormal value, direct deletion by a deletion method, a substitution method (substitution of a continuous variable mean, substitution of a discrete variable with a mode and a median) or an interpolation method (regression interpolation, multiple interpolation) may be selected. When the missing values are supplemented, if the missing values are continuous variables, calculating the mean value of the continuous variables to be supplemented can be selected subsequently; if the variable is a discrete variable, a mode or a median may be selected for completion. Alternatively, a regression model may be fitted to the non-missing data and then used to predict the missing data.
Specifically, when the threshold value checking method is adopted, abnormal operation data with a parameter value lower than the detection threshold value is filtered from the first parameter data set by setting the detection threshold value.
If it is required to ensure that the parameters in the first parameter data set are uniform in magnitude and dimension, after step S210 is performed, the following steps may be performed:
s220, carrying out standardization processing on the second parameter data set with the filtered abnormal operation data by adopting a Z-score data standardization method; specifically, the normalizing the second parameter data set by the Z-score data normalization method includes:
calculating the mean value of the overall data in the second parameter data set by using the formula (1)
Calculating to obtain a standard deviation s of the overall data in the second parameter data set by using a formula (2);
combining the mean value and standard deviation of the overall data, and obtaining a final standardized processing result Z by using a formula (3)i：
Where n denotes the total number of feature vectors in the second parameter dataset, xiAnd representing the mean value corresponding to the ith feature vector.
Finally, in this embodiment, a mutual information feature selection algorithm is considered to be used, feature selection is performed based on the normalized second parameter data set or the normalized first parameter data set with the abnormal operating data filtered out, wherein data in the data sets are combined pairwise, mutual information between the two combined data is calculated, and a mutual information calculation formula is as follows:
(4) in the formula: p (X, Y) is the joint probability distribution of X and Y; p (X) and p (Y) are the edge probability distributions for X and Y, respectively. In the embodiment, mutual information is used for quantifying the correlation degree among 2 random variables, and when the mutual information is 0, the variable X and the variable Y are in an independent relationship; and when the mutual information value is larger, the higher the correlation degree between the variable X and the variable Y is.
After the mutual information of all pairwise combined data is calculated, sorting can be performed according to the value size of the obtained multiple items of mutual information, and the front k items of feature subsets are selected as the input of the model to complete feature selection.
In the method for predicting the gas emission concentration of the GRU network disclosed in this embodiment 2, in order to improve the accuracy of the measured value, before feature selection is performed on the acquired data set, abnormal operation data is sequentially filtered, and the filtered data set is subjected to standardized processing, so that the acquired data meets the actual field requirements, and the prediction precision is effectively improved.
Example 3:
for easy understanding, please refer to fig. 3, fig. 3 is a system structural diagram of a gas emission concentration prediction system of a GRU network according to an embodiment of the present application, and as shown in fig. 3, the system structural diagram specifically includes:
the first parameter data set acquisition unit 10 is for acquiring a first parameter data set for predicting a gas emission concentration;
the feature selection unit 20 is configured to perform feature selection on the first parameter data set by using a mutual information feature selection algorithm to obtain a plurality of feature vectors for inputting to the GRU network for network training; the characteristic selection by utilizing a mutual information characteristic selection algorithm comprises the following steps:
combining the data included in the first parameter data set in pairs in sequence to obtain a plurality of combination sequences, and calculating mutual information among elements aiming at each combination sequence;
sequencing a plurality of items of mutual information according to the value of the mutual information to obtain a mutual information sequencing sequence;
selecting front k items of elements from the mutual information sequencing sequence, selecting corresponding data items from the first parameter data set based on the selected k items of elements, and taking the data items as the input of the GRU network; k is 1 or more.
The data dividing unit 30 is configured to divide the first parameter data set into a training set and a test set according to the obtained multiple feature vectors;
the GRU network training unit 40 is configured to use the training set as an input of the GRU network, train the GRU network, and output a pre-training network until an iterative convergence condition is satisfied;
the prediction result output unit 50 is configured to use the test set as an input of the pre-training network, calculate a deviation between a predicted value output by the pre-training network and a given actual value, and obtain a gas emission concentration prediction model if the deviation value is smaller than a preset deviation threshold.
Example 4:
in order to be able to filter out abnormal operation data from the first parameter data set, the system further comprises a data filtering unit 60; wherein:
the data filtering unit 60 is configured to filter abnormal operation data from the first parameter data set by using a singular value inspection method or a threshold inspection method; when the abnormal operation data filtering is carried out by adopting a singular value inspection method, the method comprises the following steps:
performing singular value inspection on each item of data included in the first parameter data set;
defining an abnormal value according to a singular value inspection result;
removing abnormal values from the first parameter dataset, or defining abnormal values as missing values and performing completion of the missing values;
when the threshold value checking method is adopted, abnormal operation data with a parameter value lower than a detection threshold value are filtered from the first parameter data set by setting the detection threshold value.
It should be noted that, in order to ensure that the parameters in the first parameter data set are uniform in magnitude and dimension; the system further comprises a normalization processing unit 70, wherein:
the normalization processing unit 70 is configured to, after filtering the abnormal operation data from the first parameter data set by using a singular value inspection method or a threshold inspection method, perform normalization processing on the second parameter data set with the filtered abnormal operation data by using a Z-score data normalization method.
According to the gas emission concentration prediction method and system based on the GRU network, the advantages of the GRU network in model accuracy, generalization capability and training speed are utilized, and the situation that gas concentration measurement is delayed is avoided; in order to improve the accuracy of the measured value, filtering abnormal operation data is performed in sequence before feature selection is performed on the acquired data set, and the filtered data set is subjected to standardized processing, so that the acquired data meet the actual field requirements, and the prediction precision is effectively improved; the GRU neural network structure with 4 layers is adopted, the space complexity of the network is reduced, and the requirement of practical application is met.
It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
The terms "first," "second," "third," "fourth," and the like in the description of the present application and in the above-described drawings are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the application described herein are, for example, capable of operation in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
It should be understood that in the present application, "at least one" means one or more, "a plurality" means two or more. "and/or" for describing an association relationship of associated objects, indicating that there may be three relationships, e.g., "a and/or B" may indicate: only A, only B and both A and B are present, wherein A and B may be singular or plural. The character "/" generally indicates that the former and latter associated objects are in an "or" relationship. "at least one of the following" or similar expressions refer to any combination of these items, including any combination of single item(s) or plural items. For example, at least one (one) of a, b, or c, may represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", wherein a, b, c may be single or plural.
The above embodiments are only used for illustrating the technical solutions of the present application, and not for limiting the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions in the embodiments of the present application.
Claims (10)
1. A gas emission concentration prediction method based on a GRU network is characterized by comprising the following steps:
acquiring a first parameter data set for predicting gas emission concentration;
performing feature selection on the first parameter data set to obtain a plurality of feature vectors for inputting to a GRU network for network training;
dividing the first parameter data set into a training set and a test set according to the obtained multiple feature vectors;
taking the training set as the input of the GRU network, training the GRU network, and outputting a pre-training network until an iteration convergence condition is met;
and taking the test set as the input of the pre-training network, calculating the deviation between a predicted value output by the pre-training network and a given actual value, and if the deviation value is smaller than a preset deviation threshold value, obtaining a gas emission concentration prediction model.
2. The gas emission concentration prediction method according to claim 1, further comprising, before performing the feature selection on the first parameter data set:
filtering abnormal operation data from the first parameter data set by adopting a singular value inspection method or a threshold inspection method, wherein the abnormal operation data filtering by adopting the singular value inspection method comprises the following steps:
performing a singular value check on items of data included in the first parameter data set;
defining an abnormal value according to a singular value inspection result;
removing abnormal values from the first parameter dataset or defining abnormal values as missing values and performing completion of the missing values;
the threshold value checking method comprises the step of selecting abnormal operation data with parameter values lower than a preset detection threshold value from the first parameter data set.
3. The method of predicting a gas emission concentration of claim 2, wherein after filtering abnormal operation data from the first parameter data set using a singular value inspection method or a threshold inspection method, further comprising:
and adopting a Z-score data standardization method to carry out standardization processing on the second parameter data set with the filtered abnormal operation data.
4. The method of predicting a concentration of a gas emission according to claim 3, wherein the characteristic selection from the first parameter data set is specifically:
the characteristic selection of the normalized second parameter data set by using a mutual information characteristic selection algorithm comprises the following steps:
combining the data included in the second parameter data set in pairs in sequence to obtain a plurality of combination sequences, and calculating mutual information among elements aiming at each combination sequence;
sequencing a plurality of items of mutual information according to the value of the mutual information to obtain a mutual information sequencing sequence;
selecting top k elements from the mutual information sorting sequence, selecting corresponding data items from the second parameter data set based on the selected k elements, and using the corresponding data items as the input of the GRU network; k is 1 or more.
5. The method of predicting gas emission concentration according to claim 3, wherein the normalizing the second parameter data set using the Z-score data normalization method comprises:
calculating the mean value of the overall data in the second parameter data set by using the formula (1)
Calculating to obtain a standard deviation s of the overall data in the second parameter data set by using a formula (2);
combining the mean value and standard deviation of the overall data, and obtaining a final standardized processing result Z by using a formula (3)i：
Where n denotes the total number of feature vectors in the second parameter dataset, xiAnd representing the mean value corresponding to the ith feature vector.
6. The method according to claim 1, wherein the hidden layer of the GRU network builds a 4-layer recurrent neural network with GRU cells, wherein:
the nodes contained in each layer are respectively: 128. 128, 64 and 32.
7. The method according to claim 1, wherein the calculating of the deviation between the predicted value of the pre-trained network output and the given actual value is specifically:
after a first deviation value, a second deviation value and a third deviation value between the predicted value and the given actual value are respectively calculated through a root mean square error calculation formula, an average absolute value error calculation formula and a decision coefficient calculation formula, the first deviation value, the second deviation value and the third deviation value are averaged or accumulated to obtain the deviation between the predicted value and the given actual value.
8. A GRU network-based gas emission concentration prediction system, comprising:
a first parameter data set acquisition unit for acquiring a first parameter data set for predicting a gas emission concentration;
the characteristic selection unit is used for carrying out characteristic selection on the first parameter data set by utilizing a mutual information characteristic selection algorithm to obtain a plurality of characteristic vectors for inputting to a GRU network for network training; the characteristic selection by utilizing a mutual information characteristic selection algorithm comprises the following steps:
combining the data included in the first parameter data set in pairs in sequence to obtain a plurality of combination sequences, and calculating mutual information among elements aiming at each combination sequence;
sequencing a plurality of items of mutual information according to the value of the mutual information to obtain a mutual information sequencing sequence;
selecting top k elements from the mutual information sorting sequence, selecting corresponding data items from the first parameter data set based on the selected k elements, and using the corresponding data items as the input of the GRU network; k is greater than or equal to 1;
the data dividing unit is used for dividing the first parameter data set into a training set and a test set according to the obtained multiple feature vectors;
the GRU network training unit is used for taking a training set as the input of the GRU network, training the GRU network and outputting a pre-training network until an iteration convergence condition is met;
and the prediction result output unit is used for taking the test set as the input of the pre-training network, calculating the deviation between a predicted value output by the pre-training network and a given actual value, and obtaining a gas emission concentration prediction model if the deviation value is smaller than a preset deviation threshold value.
9. The system of claim 8, further comprising a data filtering unit; wherein:
the data filtering unit is used for filtering abnormal operation data from the first parameter data set by adopting a singular value inspection method or a threshold inspection method; when the abnormal operation data filtering is carried out by adopting a singular value inspection method, the method comprises the following steps:
performing a singular value check on items of data included in the first parameter data set;
defining an abnormal value according to a singular value inspection result;
removing abnormal values from the first parameter dataset or defining abnormal values as missing values and performing completion of the missing values;
when the threshold value checking method is adopted, abnormal operation data with a parameter value lower than a detection threshold value are filtered from the first parameter data set by setting the detection threshold value.
10. The gas emission concentration prediction system of claim 9, further comprising a normalization processing unit, wherein:
the normalization processing unit is configured to, after filtering the abnormal operation data from the first parameter data set by using the singular value inspection method or the threshold inspection method, perform normalization processing on the second parameter data set with the filtered abnormal operation data by using a Z-score data normalization method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010928729.0A CN112070156B (en) | 2020-09-07 | 2020-09-07 | Gas emission concentration prediction method and system based on GRU network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010928729.0A CN112070156B (en) | 2020-09-07 | 2020-09-07 | Gas emission concentration prediction method and system based on GRU network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112070156A true CN112070156A (en) | 2020-12-11 |
| CN112070156B CN112070156B (en) | 2021-12-21 |
Family
ID=73663840
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010928729.0A Active CN112070156B (en) | 2020-09-07 | 2020-09-07 | Gas emission concentration prediction method and system based on GRU network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112070156B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113947142A (en) * | 2021-10-14 | 2022-01-18 | 天津大学 | Method and system for predicting emission concentration of acid gas and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107578124A (en) * | 2017-08-28 | 2018-01-12 | 国网山东省电力公司电力科学研究院 | The Short-Term Load Forecasting Method of GRU neutral nets is improved based on multilayer |
| CN111047012A (en) * | 2019-12-06 | 2020-04-21 | 重庆大学 | Air quality prediction method based on deep bidirectional long-short term memory network |
| CN111489015A (en) * | 2020-03-20 | 2020-08-04 | 天津大学 | Atmosphere O based on multiple model comparison and optimization3Concentration prediction method |
-
2020
- 2020-09-07 CN CN202010928729.0A patent/CN112070156B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107578124A (en) * | 2017-08-28 | 2018-01-12 | 国网山东省电力公司电力科学研究院 | The Short-Term Load Forecasting Method of GRU neutral nets is improved based on multilayer |
| CN111047012A (en) * | 2019-12-06 | 2020-04-21 | 重庆大学 | Air quality prediction method based on deep bidirectional long-short term memory network |
| CN111489015A (en) * | 2020-03-20 | 2020-08-04 | 天津大学 | Atmosphere O based on multiple model comparison and optimization3Concentration prediction method |
Non-Patent Citations (2)
| Title |
|---|
| 冀林: "基于CNNs-GRU深度学习的PM2.5预测研究与实现", 《中国优秀硕士学位论文全文数据库 工程科技Ⅰ辑》 * |
| 杨堃: "火电机组烟气脱硝系统入口NO_x浓度动态估计研究", 《中国优秀硕士学位论文全文数据库 工程科技Ⅰ辑》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113947142A (en) * | 2021-10-14 | 2022-01-18 | 天津大学 | Method and system for predicting emission concentration of acid gas and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112070156B (en) | 2021-12-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110441065B (en) | Gas turbine on-line detection method and device based on LSTM | |
| US7593804B2 (en) | Fixed-point virtual sensor control system and method | |
| US11650968B2 (en) | Systems and methods for predictive early stopping in neural network training | |
| CN109992921B (en) | On-line soft measurement method and system for thermal efficiency of boiler of coal-fired power plant | |
| CN111814956B (en) | Multi-task learning air quality prediction method based on multi-dimensional secondary feature extraction | |
| CN111148118A (en) | Flow prediction and carrier turn-off method and system based on time sequence | |
| CN112181706B (en) | Power dispatching data anomaly detection method based on logarithmic interval isolation | |
| CN114282443B (en) | Residual service life prediction method based on MLP-LSTM supervised joint model | |
| CN113405106B (en) | Artificial intelligence control method for garbage incineration process | |
| CN112070156B (en) | Gas emission concentration prediction method and system based on GRU network | |
| CN111639304B (en) | CSTR fault positioning method based on Xgboost regression model | |
| CN114764682B (en) | Rice safety risk assessment method based on multi-machine learning algorithm fusion | |
| Bhuyan et al. | A survey of computational intelligence approaches for software reliability prediction | |
| CN116843080B (en) | Machine learning-based carbon element footprint prediction method and system for urea production | |
| Jiang et al. | Multi-objective optimal placement of sensors based on quantitative evaluation of fault diagnosability | |
| Wang | Data preprocessing for soft sensor using generative adversarial networks | |
| CN114692729A (en) | New energy station bad data identification and correction method based on deep learning | |
| CN113111588B (en) | NO of gas turbine X Emission concentration prediction method and device | |
| CN113761795A (en) | Aircraft engine fault detection method and system | |
| CN112862180A (en) | Denitration system inlet NOx concentration prediction method | |
| CN111291020A (en) | Dynamic process soft measurement modeling method based on local weighted linear dynamic system | |
| CN116595883B (en) | Real-time online system state correction method for numerical reactor | |
| Aytenfsu et al. | Controlling the interior of greenhouses using Elman recurrent neural network | |
| CN111310907A (en) | Microwave assembly fault diagnosis method, device and equipment | |
| Cai et al. | Research on digital instrument recognition technology of inspection robot |
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 | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: Room 501-503, annex building, Huaye building, No.1-3 Chuimao new street, Xihua Road, Yuexiu District, Guangzhou City, Guangdong Province 510000 Applicant after: China Southern Power Grid Power Technology Co.,Ltd. Address before: Room 501-503, annex building, Huaye building, No.1-3 Chuimao new street, Xihua Road, Yuexiu District, Guangzhou City, Guangdong Province 510000 Applicant before: GUANGDONG ELECTRIC POWER SCIENCE RESEARCH INSTITUTE ENERGY TECHNOLOGY Co.,Ltd. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |