Electric precipitation data enhancement method and related device
Through the adversarial generation network generation, the enhanced data generation data set of the charging and dust removal model is expanded, and the problem of low prediction accuracy in existing models under variable operating conditions is solved, achieving higher prediction accuracy and full operating conditions coverage.
Patent Information
- Application Number
- CN202411807097.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-10
- Publication Date
- 2025-03-18
AI Technical Summary
When existing deep learning models are used in the electro-dust removal industry of thermal power plants, they cannot adapt to variable working conditions due to relying on small sample data sets, resulting in low prediction accuracy.
By obtaining the historical electrocution data set, it is divided into multiple training subsets, and generating the enhanced data of the subset to be enhanced based on the pre-trained adversarial generation network, supplementing it to the subset to be enhanced, thereby expanding the training data set.
The prediction accuracy of the dust concentration prediction model is improved, so that it can provide more accurate prediction results within the entire operating range.
Smart Images
Figure CN119646519A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of data processing technology, and in particular to an electrostatic precipitator data enhancement method and related devices. Background Art
[0002] At present, the vertical application of deep learning in the electrostatic precipitator industry of thermal power plants has become increasingly common. Taking the deep learning model-dust concentration prediction model as an example, the historical database of the electrostatic precipitator currently used in the model is a small sample data set, which cannot cover the full range of operating conditions. This results in the model being unable to adapt to changing operating conditions after being put into industrial production sites for electrostatic precipitators, and the prediction accuracy is low. Summary of the invention
[0003] In view of the above problems, the present application provides an electrostatic precipitator data enhancement method and related devices to improve the prediction accuracy of the dust concentration prediction model. The specific scheme is as follows:
[0004] The first aspect of the present application provides an electrostatic precipitator data enhancement method, comprising:
[0005] Get historical ESP datasets;
[0006] A training set is obtained from the historical electrostatic precipitator dataset;
[0007] Dividing the training set into multiple training subsets, determining the amount of training data in each training subset, and determining the training subset with the amount lower than a preset threshold as a subset to be enhanced;
[0008] Generate enhanced data of the subset to be enhanced based on the pre-trained adversarial generative network;
[0009] Supplement the enhanced data to the subset to be enhanced.
[0010] In a possible implementation, the training set is divided into multiple training subsets, including:
[0011] According to the boiler load range, the training set is divided into multiple subsets, and the data in different subsets correspond to different boiler load ranges;
[0012] The subsets are classified using one-hot encoding to obtain multiple training subsets corresponding to different data categories, each training subset corresponding to one data category.
[0013] In a possible implementation, the adversarial generative network includes: a recurrent neural network generator and a discriminator group; the discriminator group includes a bidirectional recurrent neural network true and false discriminator and a k-nearest neighbor algorithm classification discriminator;
[0014] The process of generating enhanced data of the subset to be enhanced by the pre-trained adversarial generative network includes:
[0015] Input the subset to be enhanced into the recurrent neural network generator;
[0016] The recurrent neural network generator uses a random vector generation function to generate a random generated vector corresponding to the subset to be enhanced, and inputs the random generated vector to the bidirectional recurrent neural network authenticity discriminator and the k-nearest neighbor algorithm classification discriminator respectively, and inputs the subset to be enhanced to the bidirectional recurrent neural network authenticity discriminator and the k-nearest neighbor algorithm classification discriminator respectively;
[0017] The bidirectional recurrent neural network true and false discriminator and the k-nearest neighbor algorithm classification discriminator are used to identify the subset to be enhanced and the randomly generated vector this time, and obtain the identification result this time, and input the identification result this time into the recurrent neural network generator;
[0018] The recurrent neural network generator generates the current random generated vector using a random vector generation function according to the current identification result, takes the current random generated vector as the current random generated vector, inputs the current random generated vector into the bidirectional recurrent neural network authenticity discriminator and the k-nearest neighbor algorithm classification discriminator respectively, and the bidirectional recurrent neural network authenticity discriminator and the k-nearest neighbor algorithm classification discriminator discriminate the subset to be enhanced and the current random generated vector to obtain the current identification result, takes the current identification result as the current identification result, and inputs the current identification result into the recurrent neural network generator;
[0019] Iteratively execute the steps of: the recurrent neural network generator generates a current random generated vector using a random vector generation function according to the current identification result, uses the current random generated vector as the current random generated vector, inputs the current random generated vector into the bidirectional recurrent neural network authenticity discriminator and the k-nearest neighbor algorithm classification discriminator respectively, obtains the current identification result, uses the current identification result as the current identification result, and inputs the current identification result into the recurrent neural network generator, until the identification result representation corresponding to the current random generated vector generated by the recurrent neural network generator cannot distinguish the current random generated vector from the subset to be enhanced, and uses the current random generated vector as the enhanced data of the subset to be enhanced.
[0020] In one possible implementation, the training process of a pre-trained GAN includes:
[0021] The parameters of the recurrent neural network generator are fixed to update the parameters of the bidirectional recurrent neural network authenticity discriminator and the k-nearest neighbor algorithm classification discriminator with a preset number of cycles;
[0022] The parameters of the bidirectional recurrent neural network authenticity discriminator and the parameters of the k-nearest neighbor algorithm classification discriminator are fixed to update the parameters of a periodic recurrent neural network generator;
[0023] The parameter update process of the recurrent neural network generator, the bidirectional recurrent neural network authenticity discriminator and the k-nearest neighbor algorithm classification discriminator is iteratively executed until the discrimination accuracy of the recurrent neural network generator is lower than the preset discrimination threshold and the discrimination accuracy of the bidirectional recurrent neural network authenticity discriminator reaches the preset discrimination threshold.
[0024] In a possible implementation, the method further includes:
[0025] The dust concentration prediction model is trained based on the training data in the data-enhanced training set, and the data-enhanced training set includes a subset to be enhanced after supplementing the enhanced data and a training subset whose number is not less than a preset threshold.
[0026] In a possible implementation, the method further includes:
[0027] The test set obtained from the historical electrostatic precipitator dataset is used to test the prediction accuracy of the dust concentration prediction model trained based on the training set in the training set after data enhancement.
[0028] The second aspect of the present application provides an electrostatic precipitator data enhancement device, comprising:
[0029] an acquisition unit, a training set acquisition unit, a determination unit, an enhanced data generation unit, and a supplementation unit; wherein:
[0030] An acquisition unit, used for obtaining a historical electrostatic precipitator data set;
[0031] A training set acquisition unit, used for acquiring a training set from a historical electrostatic precipitator data set;
[0032] A determination unit, configured to divide the training set into a plurality of training subsets, determine the amount of training data in each training subset, and determine the training subset with the amount lower than a preset threshold as a subset to be enhanced;
[0033] An enhanced data generation unit, used to generate enhanced data of a subset to be enhanced based on a pre-trained adversarial generative network;
[0034] The supplementing unit is used to supplement the enhanced data into the subset to be enhanced.
[0035] A third aspect of the present application provides a computer program product, comprising computer-readable instructions. When the computer-readable instructions are executed on an electronic device, the electronic device implements the electrostatic precipitator data enhancement method as described in any one of the above.
[0036] In a fourth aspect, the present application provides an electrostatic precipitator data enhancement device, comprising at least one processor and a memory connected to the processor, wherein:
[0037] The memory is used to store computer programs;
[0038] The processor is used to execute a computer program so that the electrostatic precipitator data enhancement device can implement the electrostatic precipitator data enhancement method as described in any one of the above.
[0039] A fifth aspect of the present application provides a computer storage medium, which carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement the electrostatic precipitator data enhancement method as described in any one of the above.
[0040] By means of the above technical scheme, the electrostatic precipitator data enhancement method and related devices provided by the present application divide the training set obtained from the historical electrostatic precipitator data into multiple training subsets, determine the number of training data in each training subset, and use the training subset with the number of training data lower than the preset threshold as the data to be enhanced, and then generate the enhanced data of the subset to be enhanced based on the pre-trained adversarial generative network, and finally, add the enhanced data to the subset to be enhanced. The present application mainly uses a pre-trained adversarial generative network to generate enhanced data for a training subset with a small number of training data, thereby obtaining training data under all working conditions of the electrostatic precipitator, and the dust concentration prediction model trained with the data-enhanced training data has high prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and the originals and elements are not necessarily drawn to scale.
[0042] Figure 1 A schematic diagram of a flow chart of the electrostatic precipitator data enhancement method provided in this application;
[0043] Figure 2 Another schematic diagram of the process of the electrostatic precipitator data enhancement method provided in this application;
[0044] Figure 3 An example diagram of the operation of the dust concentration prediction model provided for this application;
[0045] Figure 4 A schematic diagram of the structure of an electrostatic precipitator data enhancement device provided in this application;
[0046] Figure 5 A schematic diagram of the structure of the electrostatic precipitator data enhancement device provided in this application. DETAILED DESCRIPTION
[0047] The following describes the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. The terms used in the implementation method section of the present application are only used to explain the specific embodiments of the present application, and are not intended to limit the present application.
[0048] The embodiments of the present application are described below in conjunction with the accompanying drawings. Those skilled in the art will appreciate that, with the development of technology and the emergence of new scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.
[0049] The terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and need not be used to describe a specific order or sequential order. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances, which is only to describe the distinction mode adopted by the objects of the same attributes when describing in the embodiments of the present application. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, so that the process, method, system, product or equipment comprising a series of units need not be limited to those units, but may include other units that are not clearly listed or inherent to these processes, methods, products or equipment.
[0050] The deep learning model used in the known technology - the dust concentration prediction model, is mainly based on the analysis of the historical database of the original electrostatic precipitator, and is obtained after training with a small sample data set. After being put into industrial production sites, it was found that the model cannot adapt to changing working conditions and has large deviations. It can be seen that the prediction accuracy of the dust concentration prediction model trained based on a small sample data set cannot meet actual needs. A method for enhancing the electrostatic precipitator data is needed to expand the original data set within the full range of working conditions and cover all working conditions as much as possible, thereby increasing the diversity of historical data and improving the prediction accuracy of the trained dust concentration prediction model.
[0051] In order to solve the above problems, the present application provides an electrostatic precipitator data enhancement method and related devices.
[0052] Optional, see Figure 1 , a flow chart of the electrostatic precipitator data enhancement method provided in this application.
[0053] like Figure 1 As shown, the electrostatic precipitator data enhancement method includes the following steps:
[0054] Step 101: Obtain a historical electrostatic precipitator dataset.
[0055] It should be noted that historical ESP data generally include the operating parameters of the ESP, such as voltage, current, etc.; emission concentration, specifically refers to the dust concentration at the outlet of the ESP; cleaning cycle, that is, the cleaning frequency and time of the ESP; particulate matter concentration data, specifically including the particulate matter concentration at the inlet and outlet of the ESP; and so on.
[0056] The historical electrostatic precipitator data set obtained in this application includes historical electrostatic precipitator data extracted from the original historical electrostatic precipitator data set according to the time series length. Specifically, assuming that the length of the original historical electrostatic precipitator data set is m, the feature dimension is d, and the historical electrostatic precipitator data with a feature dimension of d, a time series length of t, and a total length of mt are extracted from the original historical electrostatic precipitator data set according to the time series length t.
[0057] Optionally, a historical electrostatic precipitator data set is obtained, which is extracted from the original historical electrostatic precipitator data set according to the time series length and is consistent with the feature dimension of the original historical electrostatic precipitator data.
[0058] Step 102: Obtain a training set from a historical electrostatic precipitator data set.
[0059] Then, the historical electrostatic precipitator data set needs to be divided into a training set and a test set. The training set is the basis for enhancing the electrostatic precipitator data, and the test set is mainly used for testing
[0060] Specifically, the historical electrostatic precipitator data set can be divided into a training set and a test set according to a certain ratio.
[0061] Optionally, the historical electrostatic precipitator dataset is divided according to a preset ratio to obtain a training set and a test set.
[0062] Exemplarily, the preset ratio may be 8:2.
[0063] Step 103: divide the training set into multiple training subsets, determine the amount of training data in each training subset, and determine the training subset with the amount lower than a preset threshold as the subset to be enhanced.
[0064] First, the training set is divided into multiple subsets according to the boiler load range interval. Specifically, the training set is divided into multiple subsets according to the boiler load range as the range and the preset interval division granularity.
[0065] It should be noted that the training set is divided according to the boiler load range because the boiler load size is an important parameter that can reflect the operating conditions of the electrostatic precipitator. That is, dividing the training set according to the boiler load range can reflect the distribution characteristics of the training set within the entire operating range.
[0066] For example, assuming that the boiler load range is 0-300MW and 50 is the boiler load range interval, the six subsets obtained are 0-50|50-100|100-150|150-200|200-250|250-300|.
[0067] Then, each subset is classified by one-hot encoding, that is, the data category in each subset is encoded to obtain multiple training subsets corresponding to different data categories, each training subset corresponding to a data category. Exemplarily, the generated training subsets are C1, C2, C3, ..., Ck, a total of k.
[0068] Specifically, each training subset corresponds to a data category, and the number of training data in the training subset is the number of training data under the data category.
[0069] The amount of training data in each training subset is determined, and the training subsets whose amount of training data is lower than a preset threshold are selected as the subsets to be enhanced. That is, the training subsets with less amount of training data are selected as the subsets to be enhanced.
[0070] It should be noted that, by determining a training subset with a small amount of training data as the subset to be enhanced, and after the enhanced data is added to the subset to be enhanced, the number of training data in the subsets of different boiler load intervals can be kept balanced, so that the training data used to train the dust concentration prediction model can be a time series data set of the full working conditions of the electrostatic precipitator.
[0071] Optionally, the training plan is first divided into multiple subsets according to the boiler load range, and then each subset is divided into training subsets corresponding to different data categories using a one-hot encoding method, and a training subset with a smaller amount of training data is selected from each training subset as the subset to be enhanced.
[0072] Step 104: Generate enhanced data of the subset to be enhanced based on the pre-trained generative adversarial network.
[0073] It should be noted that the pre-trained adversarial generative network provided in this application includes a recurrent neural network (RNN, Recurrent Neural Networks) generator and a discriminator group, and the discriminator group includes a bidirectional recurrent neural network authenticity discriminator (bidirectional RNN discriminator) and a k-nearest neighbor algorithm classification discriminator (K-NearestNeighbors, KNN).
[0074] Optionally, the process of using a pre-trained generative adversarial network to generate enhanced data of the to-be-enhanced subset may be:
[0075] The subset to be enhanced is input into the RNN generator, which uses a random vector generation function to generate a random generated vector corresponding to the subset to be enhanced. The random generated vector is input into the bidirectional RNN discriminator and the KNN classification discriminator respectively. The subset to be enhanced is also input into the two discriminators together with the random generated vector. Among them, the random generated vector can be understood as
[0076] Then, the bidirectional RNN discriminator and the KNN classification discriminator will identify the authenticity of the randomly generated vector and the training data in the subset to be enhanced, and obtain the identification result. Then, the identification result is input into the RNN generator.
[0077] After receiving the identification result, the RNN generator will use a random vector to generate the current generation vector according to the identification result, and use the current random generation vector as the current random generation vector. Then, the current random generation vector is input into the bidirectional RNN discriminator and the KNN classification discriminator. The RNN discriminator and the KNN classification discriminator identify the subset to be enhanced and the current random generation vector to obtain the current identification result. The current identification result is used as the current identification result, and the current identification result is input into the RNN generator.
[0078] Iteratively execute the steps of: the RNN generator generates a current random generated vector using a random vector generation function according to the current identification result, uses the current random generated vector as the current random generated vector, inputs the current random generated vector into the bidirectional RNN discriminator and the KNN classification discriminator respectively, obtains the current identification result, uses the current identification result as the current identification result, and inputs the current identification result into the RNN generator, until the identification result representation corresponding to the current random generated vector generated by the RNN generator cannot distinguish the current random generated vector from the subset to be enhanced, and uses the current random generated vector as the enhanced data of the subset to be enhanced.
[0079] It should be noted that the role of the pre-trained adversarial generative network provided in the present application is to select a target vector that can deceive the discriminator group from the iteratively generated randomly generated vectors, that is, the randomly generated vector is equivalent to pseudo data, the training data in the subset to be enhanced is real data or can also be called reference data, and the target generated by the RNN generator is pseudo data that is infinitely close to the real data. When the discriminator group cannot distinguish the authenticity of the randomly generated vector, the adversarial generative network ends its work, and the finally generated randomly generated vector is used as the enhanced data of the subset to be enhanced.
[0080] The above describes the process of using a pre-trained adversarial generative network to generate enhanced data for the subset to be enhanced. Next, the training process of the pre-trained adversarial generative network is described:
[0081] In the adversarial generative network, the RNN generator can be called the generative model G, and the discriminator group can be called the discriminative model D. The two models are completely independent and constitute a zero-sum game. Therefore, the two models are generally trained in an alternating optimization manner. The D model is trained for a preset number of times first, and then the G network is trained and updated once.
[0082] Specifically, firstly, the parameters of the RNN generator are frozen, and the preset number of periodic bidirectional RNN discriminators and KNN classification discriminators are updated, and then the bidirectional RNN discriminators and KNN classification discriminators are fixed, and the parameters of the RNN generator of one period are updated;
[0083] The parameter update process of the RNN generator, the bidirectional RNN discriminator and the KNN classification discriminator is iteratively executed until the discrimination accuracy of the RNN generator is lower than the preset discrimination threshold and the discrimination accuracy of the discriminator group reaches the preset discrimination threshold, that is, the discrimination accuracy of the RNN is minimized and the discrimination accuracy of the discriminator group is maximized.
[0084] Step 105: Supplement the enhanced data into the subset to be enhanced.
[0085] Finally, the enhanced data is added to the subset to be enhanced, and the subset to be enhanced supplemented with the enhanced data and the training subset whose number is not less than the preset threshold constitute a new training set. This application subsequently uses the new training set to train the dust concentration prediction model.
[0086] In summary, the electrostatic precipitator data enhancement method provided by the present application divides the training set obtained from the historical electrostatic precipitator data into multiple training subsets, determines the number of training data in each training subset, takes the training subset with the number of training data lower than the preset threshold as the data to be enhanced, and then generates the enhanced data of the subset to be enhanced based on the pre-trained adversarial generative network, and finally, adds the enhanced data to the subset to be enhanced. The present application mainly uses a pre-trained adversarial generative network to generate enhanced data for a training subset with a small number of training data, thereby obtaining training data under all working conditions of the electrostatic precipitator, and the dust concentration prediction model trained with the data-enhanced training data has high prediction accuracy.
[0087] Optional, see Figure 2 , another flow chart of the electrostatic precipitator data enhancement method provided in the present application.
[0088] like Figure 2 As shown, the electrostatic precipitator data enhancement method also includes the following steps:
[0089] Step 201 : training a dust concentration prediction model based on training data in a data-enhanced training set, wherein the data-enhanced training set includes a subset to be enhanced after supplementing the enhanced data and a training subset whose number is not less than a preset threshold.
[0090] For example, see Figure 3 , an example diagram of the operation of the dust concentration prediction model provided in this application.
[0091] like Figure 3 As shown, the input is read from the training set, which includes training data arranged according to the length of the time series; let the training set be x1:T, and input x1:T into the RNN regressor network, where the training set is a training set with a feature dimension of d and a time series length of T, and the reading order is: x1, x2, x3, ..., x T-1 、x T The output of the RNN regressor is an intermediate latent vector ht.
[0092] ci is the subset to be enhanced, z1:T is the randomly generated vector generated based on the random vector generation function, (ci, z1:T) is input into the RNN generator, and the intermediate hidden vector is output t, the intermediate latent vector generates the dust concentration prediction value through the fully connected network output , the loss function corresponding to this process is , specifically through training The calculation formula for enhancing the causal dependency characteristics of the synthesized data of the generator is as follows:
[0093]
[0094] Where n is the number of samples in each batch, is the predicted value of dust concentration, The dust concentration value generated by the generator.
[0095] To be enhanced subset ci, intermediate latent vector ht and intermediate latent vector t Input the D network where the RNN authenticity discriminator is located, perform authenticity identification, and output score score1, which is the loss function L D , the loss function L D The calculation formula is as follows:
[0096]
[0097] Where n is the number of samples in each batch, D(x) is the output of the training set through the bidirectional RNN authenticity discriminator network, and D(x) is the output of the randomly generated data set through the bidirectional RNN authenticity discriminator.
[0098] To be enhanced subset ci, intermediate latent vector ht and intermediate latent vector t is input into the C network where the KNN classification discriminator is located, and classification is performed to output the score score2, which is the loss function LC. The calculation formula of LC is as follows:
[0099]
[0100] Where n is the number of samples in each batch, C(x) is the output of the training set through the KNN classification discriminator network, and C(z) is the output of the randomly generated data set through the KNN classification discriminator network.
[0101] Sum score1 and score2, output the comprehensive identification result, and find the maximum value. The calculation formula is as follows:
[0102] L U =L C +L D
[0103] Then, freeze the parameters of the RNN generator, train multiple cycle bidirectional RNN discriminators and KNN classification discriminators, and then jointly train a cycle generator and regressor, iterate the above training process for the RNN generator and the bidirectional RNN discriminator and KNN classification discriminator until the generator reaches the optimal state, and the loss function L of the RNN generator is G To find the minimum value, the calculation formula is as follows:
[0104]
[0105] in, and is the balance factor, You can take 0.04, It can be taken as 0.96.
[0106] Optionally, a LSTM (Long Short-Term Memory) dust concentration prediction model is trained based on the training data in the training set after data enhancement, and the prediction accuracy of the dust concentration prediction model is tested using a test set obtained from a historical electrostatic precipitator data set. Finally, the trained LSTM dust concentration prediction model is put into the industrial site to predict the dust concentration in real time.
[0107] Specifically, the LSTM used in the LSTM dust concentration prediction model is a three-layer network structure. The output of the first layer is 16 dimensions, and the activation function is PRELU; the output of the second layer is 32 dimensions, and the activation function is PRELU; the output of the third layer is 64 dimensions, and the activation function is PRELU; finally, it is connected to the fully connected layer to output the dust concentration prediction value;
[0108] The training process of the LSTM dust concentration prediction model can specifically set the same number of batches and the same training cycle. The dust concentration prediction model is trained using the original time series training set and the full-condition time series data set respectively. Then, the test sets are used to test them respectively, and the prediction accuracy is compared. The test results are excellent. Comparative analysis shows that after training with the full-condition time series data set, the indicators of model prediction accuracy have improved, that is, the mean square error has decreased by 5%, and the degree of fit has increased by 11.9%.
[0109] In summary, the electrostatic precipitator data enhancement method provided in the present application expands the original training set within the full operating condition range, combines and generates training data of the full operating condition time series, improves the diversity of historical electrostatic precipitator data, and thus solves the problem that the small sample of the original historical database of the electrostatic precipitator cannot cover all operating conditions.
[0110] The above introduces an electrostatic precipitator data enhancement method provided by an embodiment of the present application. The following will introduce a device for executing the above electrostatic precipitator data enhancement method.
[0111] See also Figure 4 , Figure 4 This is a schematic diagram of the structure of an electrostatic precipitator data enhancement device provided in this application. Figure 4 As shown, the device comprises:
[0112] An acquisition unit 10, a training set acquisition unit 20, a determination unit 30, an enhanced data generation unit 40 and a supplementation unit 50; wherein:
[0113] An acquisition unit 10 is used to obtain a historical electrostatic precipitator data set;
[0114] A training set acquisition unit 20, used to obtain a training set from a historical electrostatic precipitator data set;
[0115] A determination unit 30 is used to divide the training set into multiple training subsets, determine the amount of training data in each training subset, and determine the training subset with the amount lower than a preset threshold as a subset to be enhanced;
[0116] An enhanced data generating unit 40, configured to generate enhanced data of a subset to be enhanced based on a pre-trained generative adversarial network;
[0117] The supplementing unit 50 is used to supplement the enhanced data into the subset to be enhanced.
[0118] In one implementation, the determination unit 30 is specifically configured to:
[0119] According to the boiler load range, the training set is divided into multiple subsets, and the data in different subsets correspond to different boiler load ranges;
[0120] The subsets are classified using one-hot encoding to obtain multiple training subsets corresponding to different data categories, each training subset corresponding to one data category.
[0121] In one embodiment, the adversarial generation network in the enhanced data generation unit 40 includes: a recurrent neural network generator and a discriminator group; the discriminator group includes a bidirectional recurrent neural network true and false discriminator and a k-nearest neighbor algorithm classification discriminator; the process of generating enhanced data of the subset to be enhanced by the pre-trained adversarial generation network in the enhanced data generation unit 40 includes:
[0122] Input the subset to be enhanced into the recurrent neural network generator;
[0123] The recurrent neural network generator uses a random vector generation function to generate a random generated vector corresponding to the subset to be enhanced, and inputs the random generated vector to the bidirectional recurrent neural network authenticity discriminator and the k-nearest neighbor algorithm classification discriminator respectively, and inputs the subset to be enhanced to the bidirectional recurrent neural network authenticity discriminator and the k-nearest neighbor algorithm classification discriminator respectively;
[0124] The bidirectional recurrent neural network true and false discriminator and the k-nearest neighbor algorithm classification discriminator are used to identify the subset to be enhanced and the randomly generated vector this time, and obtain the identification result this time, and input the identification result this time into the recurrent neural network generator;
[0125] The recurrent neural network generator generates the current random generated vector using a random vector generation function according to the current identification result, takes the current random generated vector as the current random generated vector, inputs the current random generated vector into the bidirectional recurrent neural network authenticity discriminator and the k-nearest neighbor algorithm classification discriminator respectively, and the bidirectional recurrent neural network authenticity discriminator and the k-nearest neighbor algorithm classification discriminator discriminate the subset to be enhanced and the current random generated vector to obtain the current identification result, takes the current identification result as the current identification result, and inputs the current identification result into the recurrent neural network generator;
[0126] Iteratively execute the steps of: the recurrent neural network generator generates a current random generated vector using a random vector generation function according to the current identification result, uses the current random generated vector as the current random generated vector, inputs the current random generated vector into the bidirectional recurrent neural network authenticity discriminator and the k-nearest neighbor algorithm classification discriminator respectively, obtains the current identification result, uses the current identification result as the current identification result, and inputs the current identification result into the recurrent neural network generator, until the identification result representation corresponding to the current random generated vector generated by the recurrent neural network generator cannot distinguish the current random generated vector from the subset to be enhanced, and uses the current random generated vector as the enhanced data of the subset to be enhanced.
[0127] In one embodiment, the training process of the pre-trained generative adversarial network in the enhanced data generation unit 40 includes:
[0128] The parameters of the recurrent neural network generator are fixed to update the parameters of the bidirectional recurrent neural network authenticity discriminator and the k-nearest neighbor algorithm classification discriminator with a preset number of cycles;
[0129] The parameters of the bidirectional recurrent neural network authenticity discriminator and the parameters of the k-nearest neighbor algorithm classification discriminator are fixed to update the parameters of a periodic recurrent neural network generator;
[0130] The parameter update process of the recurrent neural network generator, the bidirectional recurrent neural network authenticity discriminator and the k-nearest neighbor algorithm classification discriminator is iteratively executed until the discrimination accuracy of the recurrent neural network generator is lower than the preset discrimination threshold and the discrimination accuracy of the bidirectional recurrent neural network authenticity discriminator reaches the preset discrimination threshold.
[0131] In one embodiment, the electrostatic precipitator data enhancement device further includes a model training unit;
[0132] The model training unit is used to train the dust concentration prediction model based on the training data in the data-enhanced training set, and the data-enhanced training set includes the subset to be enhanced after the supplementary enhanced data and the training subset whose number is not less than a preset threshold.
[0133] In one embodiment, the electrostatic precipitator data enhancement device further includes a testing unit;
[0134] The testing unit is specifically used to use a test set obtained from a historical electrostatic precipitator data set to test the prediction accuracy of a dust concentration prediction model trained based on a training set in a training set after data enhancement.
[0135] The present application also provides an electrostatic dust removal data enhancement device. Figure 5 As shown, it shows a schematic diagram of the structure of the electrostatic precipitator data enhancement device suitable for implementing the electrostatic precipitator data enhancement device provided in the present application. The electrostatic precipitator data enhancement device in the embodiment of the present application may include but is not limited to fixed terminals such as mobile phones, laptop computers, PDAs (personal digital assistants), PADs (tablet computers), desktop computers, etc. Figure 5 The electrostatic precipitator data enhancement device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.
[0136] like Figure 5As shown, the electrostatic precipitator data enhancement device may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage device 608 to a random access memory (RAM) 603. When the electrostatic precipitator data enhancement device is powered on, various programs and data required for the operation of the electrostatic precipitator data enhancement device are also stored in the RAM 603. The processing device 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.
[0137] Typically, the following devices may be connected to the I / O interface 605: input devices 606 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 607 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 608 including, for example, a memory card, a hard disk, etc.; and communication devices 609. The communication device 609 may allow the electrostatic precipitator data enhancement device to communicate with other devices wirelessly or by wire to exchange data. Although Figure 5 The electrostatic precipitator data enhancement device with various devices is shown, but it should be understood that it is not required to implement or have all the devices shown. More or fewer devices may be implemented or have instead.
[0138] An embodiment of the present application also provides a computer program product, including computer-readable instructions. When the computer-readable instructions are executed on an electronic device, the electronic device implements any one of the electrostatic precipitator data enhancement methods provided in the embodiments of the present application.
[0139] A computer-readable storage medium is also provided in an embodiment of the present application. The storage medium carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement any one of the electrostatic precipitator data enhancement methods provided in the embodiment of the present application.
[0140] It should also be noted that the device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed over multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the drawings of the device embodiments provided by the present application, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines.
[0141] Through the description of the above implementation mode, the technicians in the field can clearly understand that the present application can be implemented by means of software plus necessary general hardware, and of course, it can also be implemented by special hardware including special integrated circuits, special CPUs, special memories, special components, etc. In general, all functions completed by computer programs can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be various, such as analog circuits, digital circuits or special circuits. However, for the present application, software program implementation is a better implementation mode in more cases. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a readable storage medium, such as a computer floppy disk, a U disk, a mobile hard disk, a ROM, a RAM, a disk or an optical disk, etc., including a number of instructions to enable a computer device (which can be a personal computer, a training device, or a network device, etc.) to execute the methods described in each embodiment of the present application.
[0142] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments may be implemented in the form of a computer program product.
[0143] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from a website site, a computer, a training device, or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, training device, or data center. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device, a data center, etc. that includes one or more available media integrations. The available medium may be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)), etc.
Claims
1. A method for enhancing electrostatic precipitator data, characterized in that: include: Get historical ESP datasets; A training set is obtained from the historical electrostatic precipitator dataset; Dividing the training set into a plurality of training subsets, determining the amount of training data in each of the training subsets, and determining the training subsets whose amount is lower than a preset threshold as subsets to be enhanced; Generate enhanced data of the to-be-enhanced subset based on a pre-trained generative adversarial network; The enhanced data is added to the to-be-enhanced subset.
2. The electrostatic precipitator data enhancement method according to claim 1, characterized in that: The step of dividing the training set into a plurality of training subsets comprises: According to the boiler load range, the training set is divided into a plurality of subsets, and the boiler load ranges corresponding to the data in different subsets are different; Each subset is classified by using a one-hot encoding method to obtain a plurality of training subsets corresponding to different data categories, each of which corresponds to one data category.
3. The electrostatic precipitator data enhancement method according to claim 1, characterized in that: The adversarial generation network includes: a recurrent neural network generator and a discriminator group; the discriminator group includes a bidirectional recurrent neural network true and false discriminator and a k-nearest neighbor algorithm classification discriminator; The process of generating enhanced data of the to-be-enhanced subset by the pre-trained generative adversarial network includes: Inputting the subset to be enhanced into the recurrent neural network generator; The recurrent neural network generator uses a random vector generation function to generate a current random generation vector corresponding to the subset to be enhanced, and inputs the current random generation vector into the bidirectional recurrent neural network authenticity discriminator and the k-nearest neighbor algorithm classification discriminator respectively, and inputs the subset to be enhanced into the bidirectional recurrent neural network authenticity discriminator and the k-nearest neighbor algorithm classification discriminator respectively; The bidirectional recurrent neural network true and false discriminator and the k-nearest neighbor algorithm classification discriminator are used to identify the subset to be enhanced and the randomly generated vector this time, obtain the current identification result, and input the current identification result into the recurrent neural network generator; The recurrent neural network generator generates a current random generated vector using the random vector generation function according to the current identification result, uses the current random generated vector as the current random generated vector, and inputs the current random generated vector into the bidirectional recurrent neural network authenticity discriminator and the k-nearest neighbor algorithm classification discriminator respectively, and the bidirectional recurrent neural network authenticity discriminator and the k-nearest neighbor algorithm classification discriminator discriminate the subset to be enhanced and the current random generated vector to obtain a current identification result, uses the current identification result as the current identification result, and inputs the current identification result into the recurrent neural network generator; Iteratively execute the steps of: the recurrent neural network generator generates the current random generated vector according to the current identification result by using the random vector generation function, uses the current random generated vector as the current random generated vector, inputs the current random generated vector into the bidirectional recurrent neural network authenticity discriminator and the k-nearest neighbor algorithm classification discriminator respectively, obtains the current identification result, uses the current identification result as the current identification result, and inputs the current identification result into the recurrent neural network generator, until the identification result representation corresponding to the current random generated vector generated by the recurrent neural network generator cannot distinguish between the current random generated vector and the subset to be enhanced, and uses the current random generated vector as the enhanced data of the subset to be enhanced.
4. The electrostatic precipitator data enhancement method according to claim 3, characterized in that: The training process of the pre-trained adversarial generative network includes: Fixing the parameters of the recurrent neural network generator to update the parameters of the bidirectional recurrent neural network authenticity discriminator and the k-nearest neighbor algorithm classification discriminator for a preset number of cycles; Fixing the parameters of the bidirectional recurrent neural network authenticity discriminator and the parameters of the k-nearest neighbor algorithm classification discriminator to update the parameters of the recurrent neural network generator for one cycle; Iteratively execute the parameter update process of the recurrent neural network generator, the bidirectional recurrent neural network authenticity discriminator and the k-nearest neighbor algorithm classification discriminator until the discrimination accuracy of the recurrent neural network generator is lower than a preset discrimination threshold and the discrimination accuracy of the bidirectional recurrent neural network authenticity discriminator reaches a preset discrimination threshold.
5. The electrostatic precipitator data enhancement method according to claim 1, characterized in that: Also includes: The dust concentration prediction model is trained based on the training data in the data-enhanced training set, wherein the data-enhanced training set includes the subset to be enhanced after the enhanced data is supplemented and the training subset whose number is not less than a preset threshold.
6. The electrostatic precipitator data enhancement method according to claim 5, characterized in that: Also includes: The prediction accuracy of the dust concentration prediction model trained by the training set in the training set after data enhancement is tested using a test set obtained from the historical electrostatic precipitator data set.
7. An electrostatic precipitator data enhancement device, characterized in that: include: an acquisition unit, a training set acquisition unit, a determination unit, an enhanced data generation unit, and a supplementation unit; wherein: The acquisition unit is used to obtain a historical electrostatic precipitator data set; The training set acquisition unit is used to obtain a training set from a historical electrostatic precipitator data set; The determining unit is used to divide the training set into multiple training subsets, determine the amount of training data in each of the training subsets, and determine the training subset whose amount is lower than a preset threshold as a subset to be enhanced; The enhanced data generating unit is used to generate enhanced data of the to-be-enhanced subset based on a pre-trained generative adversarial network; The supplementing unit is used to supplement the enhanced data into the subset to be enhanced.
8. An electrostatic precipitator data enhancement device, characterized in that: The method comprises at least one processor and a memory connected to the processor, wherein: The memory is used to store computer programs; The processor is used to execute the computer program so that the electrostatic precipitator data enhancement device can implement the electrostatic precipitator data enhancement method as described in any one of claims 1 to 6.
9. A computer program product, characterized in that It comprises computer-readable instructions, and when the computer-readable instructions are executed on an electronic device, the electronic device implements the electrostatic precipitator data enhancement method as claimed in any one of claims 1 to 6.
10. A computer storage medium, characterized in that: The storage medium carries one or more computer programs, and when the one or more computer programs are executed by an electronic device, the electronic device can implement the electrostatic precipitator data enhancement method as described in any one of claims 1 to 6.