Method and system for predicting temperature of generator pipeline of once-through steam engine

Abstract

The invention relates to a once-through steam engine generator pipeline temperature prediction method and system, and the method comprises the steps: obtaining temperature data according to a once-through steam engine generator pipe wall temperature sensor; preprocessing the temperature data, and dividing the preprocessed temperature data into a training set and a test set; using the training set to train a preset temperature prediction model to obtain a trained temperature preset model; wherein the temperature prediction model is established by a residual cavity convolutional neural network; and predicting the trained temperature prediction model by using the test set, determining an evaluation index according to the test set and the temperature prediction set, and evaluating the once-through steam engine generator pipeline temperature prediction method according to the evaluation index. According to the method, the temperature data of the once-through steam engine generator pipeline are acquired by using the temperature sensor through the residual cavity convolutional neural network, and the optimal prediction model is obtained through data training, so that the rapid and accurate prediction of the temperature change trend of the pipeline is realized.

Description

Technical Field

[0001] The invention relates to the technical field of pipeline detection, and in particular to a method and system for predicting the temperature of a direct current steam engine generator pipeline. Background Art

[0002] In industrial production, especially for complex equipment such as once-through steam generators (OTSG), predicting the change of pipeline temperature is crucial to ensure safety and efficiency. Traditional methods, such as physics-based models or statistical models, are often unable to deal with nonlinear relationships and may not be able to quickly and accurately give the required prediction results.

[0003] At present, the prediction methods for time series data are mainly based on physical models or statistical methods, but these methods have difficulties in dealing with complex nonlinear relationships and cannot quickly and accurately predict the trend of change. Invention patent CN20221155995.6 proposes a classification-based SVR-based wind power short-term power forecasting accuracy improvement method and equipment, which uses weather data such as wind speed and wind direction to obtain three types of short-term power forecast errors, and then uses SVR training to obtain a short-term power error model; Invention patent CN202211559258.6 discloses an electric vehicle load forecasting method, which obtains the optimal prediction model by inputting the normalized electric vehicle load sample data into LSTM, and verifies the model with a test set;

[0004] Although machine learning methods such as SVR support vector regression and LSTM long short-term memory network have been applied in the field of prediction, they also have their own limitations. For example, when the amount of data is very large, these methods may become inefficient or even unable to handle. Moreover, their prediction results are often affected by parameter settings, which means that if the selected parameters are not appropriate, the prediction results may be greatly reduced. Summary of the invention

[0005] The technical problem to be solved by the present invention is that the traditional time series data prediction method has difficulties in dealing with the complex nonlinear relationship in the prediction of the pipe temperature of a direct current steam engine generator, and is inefficient under large amounts of data. The prediction results are easily affected by parameter settings. A method and system for predicting the pipe temperature of a direct current steam engine generator are provided.

[0006] The technical solution adopted by the present invention to solve the technical problem is: a method for predicting the temperature of a once-through steam generator pipeline, the method comprising the following steps:

[0007] Acquire temperature data according to the tube wall temperature sensor of the direct current steam engine generator;

[0008] Preprocessing the temperature data, and dividing the preprocessed temperature data into a training set and a test set;

[0009] Using the training set to train the preset temperature prediction model to obtain a trained temperature preset model; wherein the temperature prediction model is established by a residual void convolutional neural network;

[0010] Use the trained temperature prediction model to predict the test set to obtain a temperature prediction set;

[0011] An evaluation index is determined according to the test set and the temperature prediction set.

[0012] In one embodiment, the preprocessing of the temperature data includes:

[0013] Processing the temperature data using a normalization method to obtain normalized data;

[0014] Segmenting and sampling the normalized data to obtain sampled data;

[0015] The sampled data are randomly shuffled using a random permutation algorithm to obtain pre-processed temperature data.

[0016] In one embodiment, the segmenting and sampling the normalized data to obtain the sampled data includes:

[0017] Set the sampling length and sampling interval;

[0018] The normalized data is segmented according to the sampling length and the sampling interval to obtain sampling data.

[0019] In one embodiment, the using the training set to train the preset temperature prediction model includes:

[0020] Step 1: Divide the training set into multiple batches, each batch containing a preset number of samples;

[0021] Step 2: In each batch, the input data is passed to the residual hole convolutional neural network to calculate the actual output value;

[0022] Step 3: Using mean square error as the loss function, calculate the error between the actual output value and the expected output value;

[0023] Step 4: According to the gradient of the loss function, the weight parameters of the residual hole convolutional neural network are updated through the back propagation algorithm;

[0024] Step 5: Repeat steps 2 to 4 above until the loss function converges or reaches the preset number of training rounds.

[0025] In one embodiment, the residual atrous convolutional neural network includes a residual atrous convolutional layer and a subsequent processing layer;

[0026] The step of passing the input data to the residual atrous convolutional neural network and calculating the actual output value comprises:

[0027] Substituting the input data into a preset formula to calculate and obtain the residual hole convolution layer output data;

[0028] The residual dilated convolution output data is passed to the subsequent processing layer, and the actual output value is calculated.

[0029] In one embodiment, the preset formula includes:

[0030] Y d =H(f2(W d2 (f1(W d1 X+b d1 ))+b d2 ),X)

[0031] In the formula, X represents the input data, Y d Represents the residual dilated convolution output data, W d1 , W d2 Respectively represent the residual void convolution weights, b d1 , b d2 represents the depth convolution threshold, f1, f2 represent the activation function of the convolution operation, and H represents the residual connection.

[0032] In one embodiment, the evaluation indicators include mean square error, root mean square error and mean absolute error;

[0033] Determining the evaluation index according to the test set and the temperature prediction set includes:

[0034] Determine a mean square error value, a root mean square error value, and a mean absolute error value according to the test set and the temperature prediction set, respectively;

[0035] The performance of the trained model is evaluated based on the mean square error value, the root mean square error value and the mean absolute error value.

[0036] In one embodiment, the mean square error equation is:

[0037]

[0038] The root mean square error formula is:

[0039]

[0040] The mean absolute error formula is:

[0041]

[0042] Where y represents the real temperature data in the test set; χ represents the predicted temperature data in the test set.

[0043] In one embodiment, dividing the preprocessed temperature data into a training set and a test set includes: dividing 80% of the preprocessed temperature data into a training set and 20% into a test set.

[0044] The present invention also provides a direct current steam engine generator pipeline temperature prediction system, comprising a processor and a memory storing a computer program, wherein the processor implements the steps of any of the direct current steam engine generator pipeline temperature prediction methods described above when executing the computer program.

[0045] The implementation of the present invention has the following beneficial effects: including, obtaining temperature data according to the temperature sensor of the pipe wall of the direct current steam engine generator; preprocessing the temperature data, and dividing the preprocessed temperature data into a training set and a test set; using the training set to train the preset temperature prediction model to obtain a trained temperature preset model; wherein the temperature prediction model is established by a residual void convolutional neural network; using the test set to predict the trained temperature prediction model; determining an evaluation index according to the test set and the temperature prediction set, and evaluating the direct current steam engine generator pipeline temperature prediction method according to the evaluation index. The present invention uses a residual void convolutional neural network and a temperature sensor to collect the temperature data of the direct current steam engine generator pipeline, and obtains the optimal prediction model through these data training, thereby realizing a rapid and accurate prediction of the pipeline temperature change trend. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solution of the present invention, the accompanying drawings required for use in the description of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0047] Figure 1 It is a flow chart of the method for predicting the temperature of a pipe of a direct current steam engine generator of the present application. DETAILED DESCRIPTION

[0048] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] Once-through steam generator (OTSG) is one of the important equipment in many power plants. In this type of equipment, pipeline temperature is an important state monitoring parameter, which is crucial to the operational stability and safety of the equipment. However, in the process of collecting pipeline temperature data, sensor failure, data transmission errors, and signal interference may cause abnormal OTSG temperature data. If the OTSG pipeline temperature change state is not predicted in advance, the system may issue erroneous instructions, seriously affecting the safety of the system. Therefore, it is necessary to establish a prediction model for OTSG pipeline temperature to predict the temperature change trend and ensure equipment safety.

[0050] In view of the technical problems existing in the background technology: the traditional prediction method is insufficient in processing nonlinear relationships, large amounts of data and parameter sensitivity, resulting in limited prediction accuracy and poor model stability. The DC steam generator pipeline temperature prediction method of the present invention adopts a residual void convolutional neural network as a prediction model and combines it with a large amount of data collected by the temperature sensor for training, thereby solving the problems of the traditional method's inability to handle complex nonlinear relationships, low efficiency in processing large amounts of data, and the model's sensitivity to parameter settings.

[0051] Figure 1 It is a flow chart of the method for predicting the temperature of a pipe of a direct current steam engine generator of the present application.

[0052] like Figure 1 As shown, the steps of the once-through steam engine generator pipeline temperature prediction method include:

[0053] S1: Acquiring temperature data from a tube wall temperature sensor of a direct current steam engine generator;

[0054] S2: preprocessing the temperature data, and dividing the preprocessed temperature data into a training set and a test set;

[0055] S3: Using the training set to train the preset temperature prediction model to obtain a trained temperature preset model; wherein the temperature prediction model is established by a residual void convolutional neural network;

[0056] S4: Use the trained temperature prediction model to predict the test set to obtain a temperature prediction set;

[0057] S5: Determine the evaluation index based on the test set and the temperature prediction set.

[0058] Specifically, the temperature data preprocessing step S2 may include noise removal, missing value filling, normalization or standardization to ensure data quality and improve the efficiency of model training. In the temperature prediction model establishment step S3, the residual dilated convolutional neural network (ResNet with Dilated Convolutions) used combines the deep feature extraction capability of the residual network and the wide field of view characteristics of the dilated convolution, which can effectively capture the complex nonlinear relationships and long-term dependencies in the OTSG pipeline temperature data. This model design helps to solve the shortcomings of traditional models in processing big data and nonlinear relationships, while reducing the sensitivity to parameter selection.

[0059] For the temperature prediction model training step S4, the model parameters are iteratively adjusted through optimization algorithms (such as Adam, SGD stochastic gradient descent, etc.) to minimize the error between the predicted value and the actual value (such as MSE, MAE, etc.). During the training process, it is also necessary to monitor the overfitting of the model and intervene in time with techniques such as early stopping and L2 regularization. In step S5. Commonly used evaluation indicators include prediction accuracy (such as R 2 Score), MSE mean square error, MAE mean absolute error, etc. These indicators can intuitively reflect the closeness between the model prediction results and the actual observed values. In addition, you can also consider drawing a comparison chart between the predicted values ​​and the actual values ​​to more intuitively show the prediction effect of the model.

[0060] It should be noted that the neural network-based direct current steam generator pipeline temperature prediction method proposed in the present invention can realize rapid and accurate prediction of the OTSG pipeline temperature change trend through training with a large amount of data. This not only improves the prediction efficiency, but also provides strong support for the stable operation of the steam engine power generation system. At the same time, the method has good generalization ability, can adapt to the temperature change prediction needs under different working conditions, and provides an effective technical means for industrial production and energy management.

[0061] Furthermore, preprocessing the temperature data includes:

[0062] The temperature data is processed by a normalization method to obtain normalized data;

[0063] Segment and sample the normalized data to obtain sampling data;

[0064] The random permutation algorithm is used to randomly shuffle the sampled data to obtain the preprocessed temperature data.

[0065] Specifically, when the temperature data is processed by the normalization method, the maximum and minimum normalization formula is used. Specifically, the normalization formula is:

[0066] In the formula, x′ represents the normalized temperature data, whose value is between 0 and 1 (if you want to normalize to other ranges, you can adjust the formula accordingly), x represents the original temperature data, and x max To represent the maximum value in the temperature data set, x min represents the minimum value in the temperature data set.

[0067] Map the original temperature data to the interval [0,1]. This step not only helps to reduce the impact of data of different dimensions on model training, but also speeds up the convergence of the model and improves training efficiency. The normalized data is smoother, which is conducive to the model capturing subtle changes in the data. In the segmentation sampling stage, multiple subsequences are extracted from the normalized temperature data as samples according to the preset sampling length and sampling interval. These subsequences not only contain the time series information of the temperature data, but also increase the diversity of samples through data enhancement technology. At the same time, the adjacent data points of each subsequence are used as the labels of the subsequence for loss calculation in the supervised learning process. The random permutation algorithm is used to randomly shuffle the sampled data in order to prevent the model from learning the specific order information of the data during the training process, thereby improving the generalization ability of the model. The data distribution after random shuffling is more even, which helps the model learn more robust feature representations.

[0068] It should be noted that data preprocessing is one of the key steps in the entire prediction method. Through normalization, segmentation sampling and random shuffling, not only can the complexity of the data be reduced and the efficiency of model training be improved, but the generalization ability of the model can also be enhanced, so that it can better adapt to the temperature change prediction needs under different working conditions. In addition, these preprocessing steps also lay a solid foundation for the subsequent establishment of the temperature prediction model, ensuring the accuracy and reliability of the prediction results.

[0069] Furthermore, the normalized data is segmented and sampled, and the obtained sampled data includes:

[0070] Set the sampling length and sampling interval;

[0071] According to the sampling length and sampling interval, the normalized data is segmented to obtain the sampling data.

[0072] Furthermore, using the training set to train the preset temperature prediction model includes:

[0073] Step 1: Divide the training set into multiple batches, each batch contains a preset number of samples;

[0074] Step 2: In each batch, pass the input data to the residual dilated convolutional neural network to calculate the actual output value;

[0075] Step 3: Use mean square error as the loss function to calculate the error between the actual output value and the expected output value;

[0076] Step 4: According to the gradient of the loss function, the weight parameters of the residual hole convolutional neural network are updated through the back propagation algorithm;

[0077] Step 5: Repeat steps 2 to 4 above until the loss function converges or reaches the preset number of training rounds.

[0078] Specifically, in the training process of the temperature prediction model, the residual dilated convolutional neural network is used as the core architecture. With the advantages of residual connection and dilated convolution, the network can efficiently capture the long-distance dependencies in time series data. During training, the training set is divided into multiple batches, each of which contains a certain number of samples to improve training efficiency and reduce memory usage. In each batch, the normalized OTSG pipeline temperature data is passed to the network as input, and the actual output value is calculated by forward propagation. The mean square error is used as the loss function to measure the difference between the actual output value and the expected output value (i.e., the label of the sampled data). Subsequently, the weight parameters of the network are updated using the back propagation algorithm according to the gradient of the loss function to reduce the error and improve the prediction accuracy. This training process is repeated until the loss function converges or the preset number of training rounds is reached. During training, the performance of the model on the validation set is also monitored to prevent overfitting, and hyperparameters such as the learning rate and batch size are adjusted in a timely manner. In order to further improve the generalization ability of the model, strategies such as data enhancement, regularization, and early stopping may also be adopted. After this series of training steps, a temperature prediction model that can accurately predict the temperature of the OTSG pipeline can be finally obtained. Furthermore, the residual dilated convolutional neural network includes a residual dilated convolutional layer and a subsequent processing layer; the input data is passed to the residual dilated convolutional neural network, and the actual output value is calculated including:

[0079] Substitute the input data into the preset formula to calculate and obtain the output data of the residual hole convolution layer;

[0080] The residual dilated convolution output data is passed to the subsequent processing layer and the actual output value is calculated.

[0081] Specifically, in the process of establishing the temperature prediction model, a residual dilated convolutional neural network with a 13-layer structure is used. When the input data is passed to this network, it first passes through three residual dilated convolutional layers. In these layers, the input data is calculated according to formula (2), where formula (2) is Y d =H(f2(Wd2 (f1(W d1 X+b d1 ))+b d2 ),X), where X represents the input data, Y d Represents the output data of the residual hole convolution layer, W d1 and W d2 is the convolution weight, b d1 and b d2 is the threshold of the depthwise convolution (or called the bias term), f1 and f2 are the activation functions of the convolution operation (the ReLU activation function is used here, which can convert negative inputs to 0 and directly output positive inputs), and H represents the residual connection, which can add the output of the previous layer to the output of the current layer, thereby helping the network learn deeper features.

[0082] After being processed by the residual dilated convolutional layer, the data is passed to three batch normalization layers. In these layers, the data is normalized to improve the stability and efficiency of the training process. The calculation process of the batch normalization layer involves calculating the mean M and variance V of the input data, as well as translating and scaling the normalized data (controlled by parameters λ and ξ) to restore the feature distribution that the original network is trying to learn.

[0083] Subsequently, the data will pass through three pooling layers. The main function of the pooling layer is to downsample the output of the convolution layer to reduce the number of parameters in the subsequent layers and improve the computational efficiency of the model. In this process, the feature map obtained by each convolution kernel will be locally counted (such as maximum pooling or average pooling) to retain the most important features in the original data and reduce the size of the feature space.

[0084] Finally, the data is passed to three fully connected layers and one tiled layer. In the fully connected layer, the data is mapped to the final output. Each neuron is connected to all neurons in the previous layer and learns a set of weights and biases to convert the output of the previous layer into the input of the current layer and obtain the output of the neuron. The tiled layer is usually used to convert multi-dimensional data into one-dimensional data for subsequent processing and analysis.

[0085] During the entire calculation process, the residual atrous convolutional neural network can take advantage of the residual connection and atrous convolution to improve the network's expression and feature extraction capabilities. At the same time, through the synergy of the BN layer, pooling layer, and fully connected layer, the network can learn deeper features and output accurate temperature prediction results.

[0086] The ReLU activation function is a nonlinear function that converts negative input to 0 and directly outputs positive input. The ReLU activation function can speed up the calculation, solve the problem of gradient disappearance during the back propagation process, and improve the generalization ability of the model.

[0087] Among them, the BN layer is a common layer type in deep neural networks. It improves the stability and efficiency of the training process by standardizing the input of the neural network. When training deep neural networks, due to the complexity of the network structure and the uneven distribution of weights, the values ​​of neuron outputs often fluctuate greatly, which will have an adverse effect on the learning effect of the network. The normalization layer standardizes the input data, and can compress the data into a relatively uniform range even when the input data distribution is relatively uneven, thereby improving the robustness of the model. The calculation process of the BN layer is as follows:

[0088]

[0089] y=λx'+ξ(4)

[0090] Among them, x′ is the normalized data, y represents the output of the BN layer, that is, the normalized output value. M and V represent the mean function and variance function of the input x, and λ and ξ represent the translation and scaling of the normalized x, which can restore the feature distribution to be learned by the original network.

[0091] The pooling layer is mainly used to downsample the output of the convolution layer. By performing local statistics on the input data, the feature map obtained by each convolution kernel is compressed, thereby reducing the number of parameters in the subsequent layers and improving the computational efficiency and robustness of the model. The pooling layer is usually placed after the convolution layer to reduce the size of the feature space while retaining the most important features in the original data.

[0092] The fully connected layers are the last few layers in a deep neural network, used to map the output of the previous feature extraction layer to the final output. The structure of the fully connected layer is very simple, consisting of multiple neurons, each of which is connected to all neurons in the previous layer to form a fully connected network. During the training process, each neuron will learn a set of weights to convert the output of the previous layer into the input of the current layer, and add a bias term to finally get the output of the neuron. The fully connected layer can automatically learn features from the raw data, thereby helping the model to better complete tasks such as classification and regression.

[0093] Furthermore, the evaluation indicators include mean square error, root mean square error, and mean absolute error;

[0094] The evaluation indicators determined based on the test set and the temperature prediction set include:

[0095] Determine the mean square error value, root mean square error value and mean absolute error value based on the test set and the temperature prediction set respectively;

[0096] The performance of the trained model was evaluated based on the mean square error, root mean square error, and mean absolute error values.

[0097] Specifically, in the process of predicting the temperature of the OTSG pipeline, the test data is first input into the trained neural network model to obtain the predicted labels of the test set, and then compared and analyzed with the true labels. In order to comprehensively evaluate the performance of the model, multiple evaluation indicators such as mean square error (MSE), root mean square error (RMSE) and mean absolute error (MAE) are used. These indicators reflect the prediction accuracy and stability of the model from different angles. The values ​​of these evaluation indicators are calculated according to the test set and the temperature prediction set respectively. By comparing and analyzing these values, the prediction performance of the model can be fully and accurately understood. Generally, the smaller these error values ​​are, the higher the prediction accuracy of the model and the better the performance. Therefore, using these evaluation indicators to evaluate the performance of the trained neural network model is an indispensable step in the process of predicting the temperature of the OTSG pipeline. It provides accurate information about the performance of the model and helps to better understand and apply the model.

[0098] Furthermore, the mean square error equation is:

[0099]

[0100] The root mean square error formula is:

[0101]

[0102] The mean absolute error formula is:

[0103]

[0104] Where y represents the real temperature data in the test set; χ represents the predicted temperature data in the test set.

[0105] Further, dividing the preprocessed temperature data into a training set and a test set includes: dividing 80% of the preprocessed temperature data into a training set and 20% into a test set.

[0106] In this embodiment, a method for predicting the temperature of a once-through steam engine generator pipeline based on a neural network includes the following steps:

[0107] (1) OTSG pipeline temperature data acquisition: 18346 data points were collected through the OTSG pipe wall temperature sensor, with a sampling interval of 1s and a sampling frequency of 1.

[0108] (2) Data preprocessing. The collected temperature data is normalized to the maximum and minimum values ​​to obtain a one-dimensional time series array. The sample length is set to 1024 and the sample interval is set to 20, so that the one-dimensional time series array can be divided into 866 samples. The next temperature data of the sample before preprocessing is used as the sample label. At the same time, the first 80% of the sample data and sample labels are selected as the training set, and the remaining samples and labels are used as the test set.

[0109] (3) Temperature prediction model establishment. The three residual dilated convolution kernels have a size of 1×3, a stride of 1, a padding of 2, a dilated convolution parameter of 2, and output channels of 5, 12, and 36, respectively. The three pooling layer filters have a size of 1×3, a stride of 2, a padding of 1, and output channels of 6, 18, and 54, respectively. The size of one tiled layer is 1×6912. The input and output of the three fully connected layers are 6912×256, 256×128, and 128×1, respectively.

[0110] (4) Temperature prediction model training. The training set is randomly shuffled and sent to the OTSG pipeline temperature prediction model in batches. The prediction model is optimized using the gradient descent method, and the model parameters are updated using the mean square error loss function. The number of training times is 100 to obtain the optimal prediction model.

[0111] (5) Predicting OTSG pipeline temperature. The optimized model is tested using the test set to obtain predicted temperature data, which is compared with the true label and compared with SVR and LSTM. The proposed method is more consistent with the true label results.

[0112] The mean square error, root mean square error and mean absolute error obtained by the method of the present invention are 0.0113, 0.1061 and 0.0895 respectively, the mean square error, root mean square error and mean absolute error obtained by SVR are 0.0344, 0.1855 and 0.1831 respectively, and the mean square error, root mean square error and mean absolute error obtained by LSTM are 0.2288, 0.4783 and 0.4769 respectively. It can be seen from the prediction results that the error of the method of the present invention is the smallest and the model fitting is the highest, which verifies that the method has good robustness for OTSG pipe wall temperature prediction.

[0113] The present invention also provides a direct current steam engine generator pipeline temperature prediction system, comprising a processor and a memory storing a computer program, wherein the processor implements the steps of any of the direct current steam engine generator pipeline temperature prediction methods described above when executing the computer program.

[0114] The present application provides a method and system for predicting the temperature of a direct current steam engine generator pipeline. The method can reduce the number of convolution parameters and reduce the computational complexity by using a dilated convolution. At the same time, the residual connection can be used to accelerate the training convergence speed, thereby improving the network training speed and reducing the sensitivity of the model to changes in input data, thereby improving the robustness and generalization ability of the model.

[0115] It can be understood that the above embodiments only express the preferred implementation modes of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the patent scope of the present invention. It should be pointed out that, for ordinary technicians in this field, the above technical features can be freely combined without departing from the concept of the present invention, and several deformations and improvements can be made, which all belong to the protection scope of the present invention. Therefore, all equivalent changes and modifications made to the scope of the claims of the present invention should belong to the coverage of the claims of the present invention.

Claims

1. A method for predicting the temperature of a once-through steam engine generator pipeline, characterized in that: Methods include: Acquire temperature data according to the tube wall temperature sensor of the direct current steam engine generator; Preprocessing the temperature data, and dividing the preprocessed temperature data into a training set and a test set; Using the training set to train the preset temperature prediction model to obtain a trained temperature preset model; wherein the temperature prediction model is established by a residual void convolutional neural network; Use the test set to make predictions on the trained temperature prediction model; An evaluation index is determined according to the test set and the temperature prediction set, and the direct current steam engine generator pipeline temperature prediction method is evaluated according to the evaluation index.

2. The method for predicting the temperature of a once-through steam engine generator pipeline according to claim 1, characterized in that: The preprocessing of the temperature data comprises: Processing the temperature data using a normalization method to obtain normalized data; Segmenting and sampling the normalized data to obtain sampled data; The sampled data are randomly shuffled using a random permutation algorithm to obtain pre-processed temperature data.

3. The method for predicting the temperature of a once-through steam engine generator pipeline according to claim 2, characterized in that: The segmenting and sampling of the normalized data to obtain the sampled data includes: Set the sampling length and sampling interval; The normalized data is segmented according to the sampling length and the sampling interval to obtain sampling data.

4. The method for predicting the temperature of a once-through steam engine generator pipeline according to claim 3, characterized in that: The using the training set to train the preset temperature prediction model comprises: Step 1: Divide the training set into multiple batches, each batch containing a preset number of samples; Step 2: In each batch, the input data is passed to the residual hole convolutional neural network to calculate the actual output value; Step 3: Using mean square error as the loss function, calculate the error between the actual output value and the expected output value; Step 4: According to the gradient of the loss function, the weight parameters of the residual hole convolutional neural network are updated through the back propagation algorithm; Step 5: Repeat steps 2 to 4 above until the loss function converges or reaches the preset number of training rounds.

5. The method for predicting the temperature of a once-through steam engine generator pipeline according to claim 4, characterized in that: The residual hole convolutional neural network includes a residual hole convolutional layer and a subsequent processing layer; The step of passing the input data to the residual atrous convolutional neural network and calculating the actual output value comprises: Substituting the input data into a preset formula to calculate and obtain the residual hole convolution layer output data; The residual hole convolution output data is passed to the subsequent processing layer, and the actual output value is calculated.

6. The method for predicting the temperature of a once-through steam engine generator pipeline according to claim 5, characterized in that: The preset formula includes: Y d =H(f2(W d2 (f1(W d1 X+b d1 ))+b d2 ),X) In the formula, X represents the input data, Y d Represents the residual dilated convolution output data, W d1 , W d2 Respectively represent the residual void convolution weights, b d1 , b d2 represents the depth convolution threshold, f1, f2 represent the activation function of the convolution operation, and H represents the residual connection.

7. The method for predicting the temperature of a once-through steam engine generator pipeline according to claim 1, characterized in that: The evaluation indicators include mean square error, root mean square error and mean absolute error; Determining the evaluation index according to the test set and the temperature prediction set includes: Determine a mean square error value, a root mean square error value, and a mean absolute error value according to the test set and the temperature prediction set, respectively; The performance of the trained model is evaluated based on the mean square error value, the root mean square error value and the mean absolute error value.

8. The method for predicting the temperature of a once-through steam engine generator pipeline according to claim 7, characterized in that: The mean square error equation is: The root mean square error formula is: The mean absolute error formula is: Where y represents the real temperature data in the test set; χ represents the predicted temperature data in the test set.

9. The method for predicting the temperature of a once-through steam engine generator pipeline according to claim 1, characterized in that: The dividing the preprocessed temperature data into a training set and a test set includes: dividing 80% of the preprocessed temperature data into a training set and 20% into a test set.

10. A once-through steam engine generator pipeline temperature prediction system, characterized in that: It comprises a processor and a memory storing a computer program, wherein the processor implements the steps of the direct current steam engine generator pipeline temperature prediction method according to any one of claims 1 to 9 when executing the computer program.

Images