Main steam flow soft measurement method and system based on TCN-TA model and medium
By using the TCN-TA model to perform soft measurement of main steam flow, the problem of low measurement accuracy of main steam flow in thermal power units is solved, achieving efficient and accurate flow prediction, and reducing energy loss and computational complexity.
Patent Information
- Application Number
- CN202510869427.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-26
- Publication Date
- 2025-11-28
AI Technical Summary
Existing technologies are insufficient for accurately monitoring the main steam flow of thermal power units, especially in ultra-supercritical units where there are problems of low measurement accuracy and energy loss. In particular, under the background of deep peak shaving, the measurement results have large deviations under the operating conditions of unit load variation and low load.
A soft measurement method for main steam flow based on TCN-TA is adopted. Data is collected from the power plant monitoring system, missing and outlier values are filled and normalized, and dimensionality is reduced by kernel principal component analysis (KPCA). A TCN-TA model is constructed, and feature information is extracted by causal convolution, extended convolution and residual blocks. An attention module is introduced for prediction.
It achieves accurate prediction of main steam flow, avoids gradient vanishing and gradient exploding problems, improves model performance and generalization ability, reduces computational complexity, and improves the accuracy and stability of prediction results.
Smart Images
Figure CN121031260A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application belongs to the technical field of thermal parameters monitoring of thermal power generating units, and particularly relates to a main steam flow soft measurement method, system and medium based on TCN-TA. BACKGROUND
[0002] As one of the core parameters of the thermal system of a power plant, the main steam flow has an important influence on the operation and performance of the unit. Accurate prediction of the main steam flow can achieve the goals of operation monitoring, energy efficiency evaluation, safety control and operation optimization of the unit, thereby improving the operation efficiency and reliability of the unit. At present, the measurement technology of the main steam flow of the unit can be mainly divided into direct measurement and indirect measurement according to the measurement principle. The direct measurement mainly adopts the differential pressure method, which is realized by installing a throttling device such as a flow orifice plate or a nozzle flowmeter. Although this method has intuitive and simple measurement results and is convenient to operate, in the actual operation process of the unit, the steam passing through the throttling device inevitably causes a certain pressure loss, which leads to an increase in system energy loss. In addition, in recent years, a large number of ultra-supercritical units have been put into operation, and the existing standard throttling device has low precision, which cannot even meet the measurement requirements of high flow of large-capacity units, and at the same time brings a non-negligible throttling loss. Therefore, the indirect measurement method is usually used to replace the throttling device for measuring the main steam flow of modern large units.
[0003] Based on the Friulli formula, the indirect measurement usually uses the steam turbine stage group front and rear pressure and temperature correction and other thermal parameters to substitute and convert to obtain the value of the main steam flow. However, when the flow path of the steam turbine is changed due to scaling or corrosion, the measurement result of the main steam flow will have a large deviation. Especially under the background of deep peak regulation, the unit needs to be in variable load and low load operation condition for a long time, and the relationship between the regulating stage rear pressure and the main steam flow is not ideal linear. Although researchers have made many modifications to the Friulli formula, the model calculation result still has a large deviation, and the use range is limited. SUMMARY
[0004] The purpose of the present application is to provide a main steam flow soft measurement method, system and medium based on TCN-TA to solve the problem that the main steam flow of the unit is difficult to accurately monitor at present.
[0005] To achieve the above-mentioned purpose, the present application adopts the following technical solutions:
[0006] A main steam flow soft measurement method based on TCN-TA, comprising the following steps:
[0007] Step 1, collecting the original data of the operation parameters related to the main steam flow from the power plant monitoring system;
[0008] Step 2, fill in the missing values and outliers in the original data obtained in step 1, and then normalize the data;
[0009] Step 3, dimensionality reduction processing is performed on the data processed in step 2 by kernel principal component analysis (KPCA);
[0010] Step 4, constructing a TCN-TA model, including an input layer, a TCN layer, a TA layer, and an output layer connected in sequence, wherein the TCN layer extracts feature information of the main steam flow, and the TA layer focuses on key time point features for prediction;
[0011] Training the TCN-TA model based on the data processed in step 3;
[0012] Step 5, inputting the to-be-tested data into the TCN-TA model trained in step 4, and outputting the main steam flow measurement value of the thermal power generating unit.
[0013] Further, in step 2, the missing values and outliers in the original data are replaced with linear interpolation results according to the time label.
[0014] Further, in step 3, the dimensionality reduction process of kernel principal component analysis (KPCA) includes: using a kernel function to obtain the random high-order correlation between input variables, and mapping the original data set to a high-dimensional feature space, so that the originally linearly inseparable data becomes linearly separable in the new feature space.
[0015] Further, in step 4, the TCN layer is composed of causal convolution, extended convolution, and residual block.
[0016] Further, in step 4, the TA layer automatically focuses on the most important part of the input sequence for the current task, while ignoring other irrelevant or secondary information.
[0017] Further, in step 4, the TCN-TA model training includes a hyperparameter optimization process: using a grid search technique to further optimize the hyperparameters of the TCN-TA model, while calculating the evaluation index under each parameter.
[0018] Further, the hyperparameter optimization process includes:
[0019] (1) determining the optimal value as 5 in the convolution kernel size candidate set [3, 5, 7, 9];
[0020] (2) determining the optimal value as 2 in the attention head number candidate set [1, 2, 4, 8];
[0021] (3) selecting 256 as the optimal batch size in the batch size candidate set [16, 32, 64, 128, 256, 512];
[0022] (4) Adopt Adam optimizer, determine the optimal learning rate is 0.01;
[0023] (5) Determine the optimal value as 2 in the expansion coefficient candidate set [1, 2, 4];
[0024] (6) Determine the optimal value as 64 in the channel number candidate set [32, 64, 128].
[0025] A TCN-TA-based main steam flow soft measurement system, comprising:
[0026] A data acquisition module acquires power plant operation parameters in real time;
[0027] A preprocessing module performs linear interpolation and KPCA dimension reduction;
[0028] A model operation module deploys a TCN-TA model and performs prediction calculation;
[0029] A visual terminal displays real-time measurement values and historical trend curves.
[0030] A non-transitory computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the steps of the TCN-TA-based main steam flow measurement method.
[0031] Beneficial effects: The application combines the efficient parallel computing capability of TCN and the key information focusing feature of TA, constructs a TCN-TA model to predict the main steam flow. Compared with existing prediction models, the model avoids the gradient disappearance and gradient explosion problems generated during time series prediction, and the introduced TA module can optimize and integrate the TCN output results, further making the prediction results more accurate. At the same time, the TA module structure is simple, which can reduce the calculation complexity and improve the performance and generalization ability of the model under the premise of ensuring the prediction effect of the main steam flow, and is convenient for model training. BRIEF DESCRIPTION OF DRAWINGS
[0032] Figure 1 A flowchart of a main steam flow measurement method based on a TCN-TA model;
[0033] Figure 2 A unit load curve diagram in the embodiment;
[0034] Figure 3 A kernel principal component analysis result diagram in the embodiment;
[0035] Figure 4 A TA module structure diagram in the embodiment;
[0036] Figure 5is a schematic diagram of the TCN-TA model in the embodiment;
[0037] Figure 6 is a TCN-based main steam flow prediction result graph in the embodiment;
[0038] Figure 7 is a TCA-TA-based main steam flow prediction result graph in the embodiment. DETAILED DESCRIPTION
[0039] The application will be further explained in conjunction with the accompanying drawings and embodiments.
[0040] EMBODIMENT
[0041] As shown in the figure, the TCN-TA-based main steam flow soft measurement method of the embodiment includes the following steps: Figure 1
[0042] (1) Collect the original data of the operation parameters related to the main steam flow from the power plant monitoring system, fill in the missing values and abnormal values in the obtained original data, and then normalize the data, and then perform dimensionality reduction processing on the data through kernel principal component analysis (KPCA).
[0043] Modern large generator units are generally reheat units. After pretreatment, the feed water is sent into the boiler through the feed water pump to exchange heat with the fuel to generate high-temperature and high-pressure steam. After entering the superheater through the steam pipeline, the steam is heated again to form superheated steam that meets the specific pressure and temperature, and drives the steam turbine to rotate and do work, and then drives the generator to convert mechanical energy into electrical energy. After the energy is released, part of the steam is extracted for reheating circulation or heating, and the remaining steam (waste steam) is condensed into water through the condenser and is sent back to the feed water pump to form a cycle.
[0044] Through the above analysis of the steam circulation process of the unit, the generation and consumption of the main steam of the thermal power unit involves complex processes and the influence of multiple factors, including boiler fuel and combustion efficiency, boiler parameters, water circulation system, power generation load, auxiliary equipment demand, steam extraction, etc. Combined with the suggestions of operation experts, 18 variables are initially selected as the target auxiliary variables for this experiment, as shown in Table 1:
[0045] Table 1 List of auxiliary variables
[0046]
[0047] Taking a 1000MW ultra-supercritical once-through reheat power generation unit as the research object, historical data from March 6, 2025, 0:00 to March 12, 2025, 22:40 of the unit were extracted from the SIS system, a total of 10000 groups of samples, the sampling time was 60s. From the Figure 2 The load curve can be seen that the set of data basically contains the target unit stable operation and variable load (including the load, load operation) of various operating conditions, the unit load range is [330MW, 1000MW].
[0048] In the face of a large number of raw data collected, due to the complex production site environment of thermal power generating units and the influence of various factors such as signal noise, some data are missing or abnormal. In this case, first of all, according to the time label, the linear interpolation result is used to replace the missing values and abnormal values in the data, so as to ensure the integrity and accuracy of the data. In addition, in order to avoid the problem of gradient explosion in the model output and improve the accuracy and training efficiency of the model, the original input data is normalized, and the calculation formula is as follows:
[0049]
[0050] Among them, x i,min , x i,max respectively represent the minimum value and maximum value of the i th variable, The normalized value is x i (t) represents the original data.
[0051] Since there are both linear and nonlinear relationships among the 18 candidate variables shown in Table 1, in order to prevent the curse of dimensionality, the KPCA method is used to further realize the dimensionality reduction and optimization of the feature data. The main principle is as follows:
[0052] Principal Component Analysis (PCA) is a data analysis method widely used for dimensionality reduction. Through linear transformation, the original data is projected into a new low-dimensional space to realize the purpose of transforming high-dimensional data set into low-dimensional representation while preserving the key information in the data set. However, due to the complexity of thermal power generating unit system, there are complex nonlinear and strong coupling relationships in the large amount of operation data generated, which may lead to PCA being unable to fully capture the nonlinear characteristics of thermal power generating unit data. In addition, there may be abnormal operation or measurement errors during the operation of the unit, which may further adversely affect the results of PCA. In contrast, Kernel Principal Component Analysis (KPCA) as a nonlinear multivariate statistical technique has better applicability in processing unit data. KPCA uses kernel functions to obtain the random high-order correlation between input variables and maps the original data set into a high-dimensional feature space, so that the originally linearly inseparable data becomes linearly separable in the new feature space, and the algorithm principle is as follows:
[0053] Suppose that after standardization, N sample points a1, a2, …, aN If each sample point has M features, then the original dataset A can be represented as:
[0054] A = {a1, a2, ..., a} N}, a i ∈R M , i = 1, 2, ..., N;
[0055] In the formula, M is the data dimension, which is also the number of features, and R... M represents the input space, and i represents the index of the sample point.
[0056] To facilitate the mapping of the original input space to the high-dimensional feature space F via nonlinearity, a nonlinear mapping is defined.
[0057]
[0058] Where Q represents the dimension of the high-dimensional feature space data sample, Q>M, R Q This represents a space of dimension Q.
[0059] Further through a i Mapped to This represents the result of mapping the i-th sample point to a high-dimensional space, realizing the transformation of the original dataset from a low-dimensional feature space to a high-dimensional feature space. Therefore, the covariance matrix C of the mapped data in the high-dimensional feature space F is... F As shown below:
[0060]
[0061] Where the superscript T denotes the transpose of the matrix, The covariance matrix C represents the result of mapping the i-th sample point to a high-dimensional space. F Let V be a square matrix of order Q, and assume its eigenvalues are λ. Then the eigenvectors V satisfy:
[0062] λV=C F V;
[0063] Furthermore, when λ≠0, the eigenvector V can be linearly represented as:
[0064]
[0065] Where, α i This represents the coefficients of the linear combination.
[0066] Define the kernel function matrix K, and let Where i and j represent the sample point indices, and K ijThe high-dimensional inner product of the sample pair is represented, and by combining the above three formulas, the following can be obtained:
[0067] Nλα=Kα, α=[α1,α2,…,α N ] T ;
[0068] wherein, α represents a vector composed of N linear combination coefficients, and the superscript T represents the transpose of a matrix.
[0069] At this time, the problem of solving the eigenvector is converted into the problem of solving the corresponding α value. KPCA obtains the eigenvalue and eigenvector through the kernel function matrix, and selects the eigenvector with the highest cumulative contribution rate according to the eigenvalue as the new principal component to form a new feature space, so as to realize the dimension reduction of data.
[0070] The kernel principal component analysis result of the candidate variable is shown in Figure 3 . Since the contribution rates of the 13th to 18th principal components are relatively low and not obvious, only the specific contribution rates of the first 12 principal components are shown in Figure 3 . According to Figure 3 , it can be known that the contribution rates of the first four principal components represented by the principal components are the largest, and the variance contribution rate of the first kernel principal component KPCA1 is as high as 40%. It can be seen from the cumulative variance contribution rate curve that the cumulative contribution rate of the first eight principal components is 98%, which basically covers the main information of the original data. The contribution rate of the ninth and subsequent principal components is close to horizontal, and the cumulative variance contribution rate curve also tends to be flat, indicating that the contribution rate of the ninth kernel principal component and subsequent principal components is very low, and the amount of original information contained is very small. In summary, the first eight principal component variables are selected as the final input of the model.
[0071] (2) Construct a TCN-TA model, including an input layer, a TCN layer, a TA layer, and an output layer connected in sequence, wherein: the TCN layer extracts the feature information of the main steam flow, and then the attention module focuses on the key time point features for prediction. The TCN-TA model is trained based on the data processed in step (1).
[0072] TCN is a time series-based convolutional neural network architecture composed of causal convolution, dilated convolution and residual block. In the prediction of main steam flow, TCN layer can extract the feature information of main steam flow, which can effectively solve the time series problem. In causal convolution, the causal relationship between layers is strictly followed, and the value of each network layer at time t depends on the value of the previous network layer at time t and before time t, which fully utilizes the relevant historical information. To solve the problem of too many layers of causal convolution, the time convolution network introduces dilated convolution to increase the receptive field by expanding the field of view, so as to obtain more historical information with fewer network layers. Residual fast can effectively solve the problem of gradient disappearance and gradient explosion in deep network model, so that the model converges more quickly. The operation formula of one-dimensional dilated causal convolution is:
[0073]
[0074] In the formula, k is the size of the convolution kernel, f is the filter, x is the input sequence, and s-di represents only the convolution operation on the historical input.
[0075] TA (Temporal Attention) is a model architecture that can automatically focus on important information at different time steps when processing sequence data. The core idea is to assign a weight to each time step in the sequence, which represents the model's attention to that time step. By learning these weights, the model can automatically focus on the most important parts of the input sequence and ignore irrelevant or secondary information. For the input model feature X, first use average pooling and maximum pooling operations to aggregate the time channel features. To effectively integrate these time features, two learnable parameters α and β are introduced to balance global (maximum pooling) and local (average pooling) information. Its structure is shown in Figure 4 . Specifically, Feature extraction M is performed using the following formula:
[0076] M = α·f avg (X) + β·f max (X)
[0077] where T is the number of time steps, C is the number of channels, f avg is the average pooling, and f max is the maximum pooling.
[0078] After extraction, the average value of M in the time dimension is calculated. Then, a linear layer and a Sigmoid function are used to obtain the weights of different time steps Finally, the input is time-enhanced.
[0079] M mean = mean(M)
[0080] T w =Sigmoid(Linear(M mean ))
[0081] X t =X+X·T w
[0082] where M mean is the result of calculating the average value of M in the time dimension, X t is the output feature vector after time enhancement, and T w represents the weight of different time steps.
[0083] Due to the characteristics of long-term operation without shutdown of thermal power units, the historical data generated by the thermal power units usually have long time series, large data volume and complex relationships. Therefore, the TCN-TA model is constructed to predict the main steam flow by combining the efficient parallel computing capability of TCN and the characteristics of TA focusing on key information. Compared with the existing prediction model, the model avoids the problems of gradient disappearance and gradient explosion in time series prediction, and the introduced TA module can optimize and integrate the TCN output results, further making the prediction result more accurate. At the same time, the TA module has a simple structure, which can reduce the calculation complexity and improve the performance and generalization ability of the model in combination with the TCN layer under the premise of ensuring the prediction effect of the main steam flow, and is convenient for the training of the model. The constructed TCN-TA model is as shown in Figure 5 .
[0084] During the experiment, the processed original data set was selected according to the ratio of 8:2 to select 8000 groups as the training set and 2000 groups of data as the test set, and the TCN-TA model was trained. The mean absolute error MAE, root mean square error RMSE and determination coefficient R 2 were used to evaluate the model performance standard, and the definitions are as follows:
[0085]
[0086]
[0087] In the formula, y i , respectively represent the actual value, the predicted value and the mean value of the actual value, and n represents the total number of samples.
[0088] To further improve the prediction accuracy and fitting effect of the network model, the grid search technology is used to further optimize the hyperparameters. In the candidate set [3, 5, 7, 9] of the convolution kernel size, the optimal value is determined as 5; in the candidate set [1, 2, 4, 8] of the attention head number, the optimal value is determined as 2; in the candidate set [16, 32, 64, 128, 256, 512] of the batch size, 256 is selected as the optimal batch size; the Adam optimizer is used, and the optimal learning rate is determined as 0.01; in the candidate set [1, 2, 4] of the expansion coefficient, the optimal value is determined as 2; in the candidate set [32, 64, 128] of the channel number, the optimal value is determined as 64.
[0089] (3) input the to-be-measured data into the TCN-TA model, and output the main steam flow measurement value of the thermal power generating unit.
[0090] The average value and standard deviation of the prediction error of the TCN-TA and the TCN network are shown in Table 2.
[0091] Table 2 Prediction performance of TCN-TA and TCN on the test set
[0092]
[0093] The RMSE of the TCN-TA model is 31.658 t / h, the MAE is 23.562 t / h, the R 2 is 0.991, and all performance indicators are better than those of the TCN network.
[0094] Based on the above hyperparameter optimization results, the test set data is used to predict the main steam flow, and the TCN-TA model constructed by the application can accurately predict the change trend. The model prediction result can determine the turning point of the actual value change of the main steam flow, and remains stable in the time period when the flow is stable, and has good fitting degree with the actual value. As Figure 6 shown. Figure 7 The prediction result of the main steam flow based on the TCN is shown in the figure.
[0095] The embodiment proposes a main steam flow soft measurement method based on the TCN-TA model to solve the problem that the main steam flow of the steam turbine is difficult to accurately monitor, and uses the operation data of a certain 1000 MW ultra-supercritical once-reheating unit for simulation verification. The KPCA algorithm is used to select and reduce the dimension of the original candidate variables, and the first 8 principal component variables are selected as the final input of the model, and the cumulative contribution rate is 98%, which basically covers the main information of the original data. The TCN-TA model after hyperparameter optimization has good prediction effect, compared with the prediction result of the non-dimensionality data, the use of KPCA dimensionality reduction can effectively reduce the influence on the prediction accuracy without losing the main information of the original data, and reduce the model training time.
[0096] The application further provides a TCN-TA-based main steam flow soft measurement system, comprising:
[0097] A data acquisition module acquires power plant operation parameters in real time;
[0098] A preprocessing module performs linear interpolation and -KPCA dimension reduction;
[0099] A model operation module deploys a TCN-TA model and performs prediction calculation;
[0100] A visual terminal displays real-time measurement values and historical trend curves.
[0101] The application further provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the TCN-TA-based main steam flow soft measurement method of the application.
[0102] The above merely describes the preferred embodiments of the application, and it should be noted that, for those skilled in the art, some improvements and refinements can be made without departing from the principles of the application, and these improvements and refinements should also be considered as falling within the protection scope of the application.
Claims
1. A soft measurement method for main steam flow rate based on TCN-TA, characterized in that, Includes the following steps: Step 1: Collect raw data of operating parameters related to main steam flow from the power plant monitoring system; Step 2: Impute missing and outlier values in the original data obtained in Step 1, and then normalize the data. Step 3: Perform dimensionality reduction on the data processed in Step 2 using kernel principal component analysis; Step 4: Construct the TCN-TA model, which includes an input layer, a TCN layer, a TA layer, and an output layer connected in sequence. The TCN layer extracts the feature information of the main steam flow, and the TA layer focuses on the features of key time points for prediction. The TCN-TA model is trained based on the data processed in step 3; Step 5: Input the data to be tested into the TCN-TA model trained in Step 4, and output the measured value of the main steam flow of the thermal power unit.
2. The method according to claim 1, characterized in that, In step 2, the missing and outlier values in the original data are replaced with linear interpolation results according to the time label.
3. The method according to claim 1, characterized in that, In step 3, the dimensionality reduction process of kernel principal component analysis includes: using kernel functions to obtain random higher-order correlations between input variables, and mapping the original dataset to a high-dimensional feature space, thereby making the originally linearly inseparable data linearly separable in the new feature space.
4. The method according to claim 1, characterized in that, In step 4, the TCN layer consists of causal convolution, extended convolution, and residual blocks.
5. The method according to claim 1, characterized in that, In step 4, the TA layer automatically focuses on the part of the input sequence that is most important to the current task, while ignoring other irrelevant or secondary information.
6. The method according to claim 1, characterized in that, In step 4, the TCN-TA model training includes a hyperparameter tuning process: grid search technology is used to further optimize the hyperparameters of the TCN-TA model to avoid random errors in the prediction results, and evaluation indicators are calculated for each parameter.
7. The method according to claim 6, characterized in that, The hyperparameter tuning process includes: (1) Determine the optimal value as 5 from the candidate set of convolution kernel size [3,5,7,9]; (2) The optimal value is determined to be 2 from the candidate set of attention heads [1,2,4,8]; (3) Select 256 as the optimal batch size from the candidate set of batch size [16,32,64,128,256,512]. (4) Using the Adam optimizer, the optimal learning rate was determined to be 0.01; (5) Determine the optimal value as 2 from the candidate set of expansion coefficients [1,2,4]; (6) The optimal value is determined to be 64 from the candidate set of channel number [32, 64, 128].
8. A main steam flow measurement system for implementing the method of any one of claims 1 to 7, characterized in that, include: The data acquisition module obtains power plant operating parameters in real time; The preprocessing module performs linear interpolation and KPCA dimensionality reduction. The model computation module deploys the TCN-TA model and performs prediction calculations; The visualization terminal displays real-time measurements and historical trend curves.
9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of a main steam flow measurement method based on TCN-TA as described in any one of claims 1 to 7.