Wall temperature prediction and control method suitable for quick load change working condition

Abstract

The present application relates to a kind of wall temperature prediction and control method suitable for quick load change condition.The method includes obtaining wall temperature and relevant operating data in the DCS system of coal-fired power generating unit, relevant operating data is used as input variable, correlation analysis is carried out between input variable and wall temperature, input variable meeting preset correlation threshold is screened out, and wall temperature prediction input data set is constructed;Variational modal decomposition is carried out to the wall temperature sequence to be predicted, and three sub-sequences of high frequency, medium frequency and low frequency are obtained;Wall temperature prediction input data set and three sub-sequences are respectively spliced with features, three different deep learning models are used to predict low-frequency, medium-frequency and high-frequency sub-sequences respectively, and the wall temperature prediction sequence is obtained by merging and processing each frequency band prediction result;Over-temperature risk determination is carried out to the wall temperature prediction sequence, and the flow of desuperheating water is adjusted according to the determination result to implement advance control.The method significantly improves the wall temperature prediction accuracy and effectively eliminates the over-temperature control lag.

Description

Technical Field

[0001] This invention relates to the field of thermal power generation equipment operation control technology, and in particular to a method for predicting and controlling wall temperature under rapid load changes. Background Technology

[0002] Against the backdrop of the "dual carbon" goals driving the power system's transformation towards a new energy-based system, thermal power generating units have shifted from traditional power supply providers to core peak-shaving and backup power sources for new energy generation. This necessitates frequent and rapid load changes. The temperature of boiler superheaters, reheaters, and other heating surfaces is a core monitoring parameter for ensuring safe unit operation, making accurate prediction and effective control a key technical challenge under peak-shaving conditions. Current mainstream wall temperature monitoring and control methods have significant limitations. Traditional heat transfer mechanism models, using fixed parameters under steady-state conditions, cannot match the dynamic changes in flue gas flow rate, velocity, and heat exchange conditions within the furnace during rapid load changes, making reliable control decisions difficult. Furthermore, traditional "post-event adjustment" control modes have a 10-30 second response lag, easily leading to temporary wall temperature overheating. Long-term, repeated overheating accelerates pipe aging and can even cause pipe rupture. Summary of the Invention

[0003] This invention provides a wall temperature prediction and control method suitable for rapid load change conditions, in order to overcome at least one of the above-mentioned technical problems existing in the prior art.

[0004] To achieve the above objectives, the embodiments of the present invention adopt the following technical solutions: In a first aspect, the present invention provides a method for predicting and controlling wall temperature under rapidly changing load conditions, comprising: Obtain wall temperature and related operating data from the DCS system of a coal-fired power unit, use the related operating data as input variables, perform correlation analysis between the input variables and the wall temperature, filter out the input variables that meet the preset correlation threshold, and construct a wall temperature prediction input dataset. The predicted wall temperature sequence is subjected to variational mode decomposition to obtain a preset number of intrinsic mode components; based on the frequency domain characteristics and complexity characteristics of each intrinsic mode component, it is reconstructed into high-frequency subsequence, mid-frequency subsequence and low-frequency subsequence. The wall temperature prediction input dataset is concatenated with the high-frequency subsequence, the mid-frequency subsequence, and the low-frequency subsequence to construct high-frequency prediction sample sets, mid-frequency prediction sample sets, and low-frequency prediction sample sets, respectively. A first deep learning model is used to predict the low-frequency prediction sample set, a second deep learning model is used to predict the mid-frequency prediction sample set, and a third deep learning model is used to predict the high-frequency prediction sample set, thereby obtaining low-frequency prediction results, mid-frequency prediction results, and high-frequency prediction results. The low-frequency prediction results, the mid-frequency prediction results, and the high-frequency prediction results are merged to obtain the wall temperature prediction sequence; The wall temperature prediction sequence is used to determine the risk of overheating, and the desuperheating water flow rate is adjusted according to the determination result to implement proactive control.

[0005] In one possible implementation of the first aspect, the correlation analysis between the input variable and the wall temperature includes: A Pearson correlation analysis was performed between the input variable and the wall temperature. The formula for calculating the Pearson correlation coefficient is as follows: ; In the formula, x i Let y be the i-th sample value of any input variable; i This is the i-th sample value of the wall temperature; The sample mean of the input variable. is the sample mean of the wall temperature; n is the number of samples.

[0006] In one possible implementation of the first aspect, the variational mode decomposition of the wall temperature sequence to be predicted includes: Establish a constrained variational problem, its expression is: ; In the formula, Let be the partial derivative with respect to time t; ω is the Dirac function; k The center frequency of the k-th mode; denoted as each modal component obtained from the decomposition; j is the imaginary unit; * is the convolution operator; f(t) is the wall temperature sequence to be predicted; and K is the number of decomposed modes. Introducing the penalty factor α and the Lagrange multiplier λ(t), we construct the augmented Lagrange expression: ; The augmented Lagrangian expression is solved using the alternating direction method of multipliers, and the frequency domain iterative update formula for each modal component is as follows: ; ; ; In the formula, ω is the frequency domain variable; n is the current iteration number; and τ is the update step size parameter. , and f(t), λ(t), and u are respectively i (t) The frequency domain representation obtained by Fourier transform.

[0007] In one possible implementation of the first aspect, the frequency domain feature is the center frequency of the intrinsic mode component, and the reconstructing of the intrinsic mode component into a high-frequency subsequence, a mid-frequency subsequence, and a low-frequency subsequence based on the frequency domain features and complexity features of each intrinsic mode component includes...

[0008] The center frequencies of each intrinsic mode component are calculated using Fourier transform, and the formula is as follows: ; In the formula, ω k ω is the center frequency of the k-th intrinsic mode component; ω is a frequency domain variable; The frequency domain representation of the k-th intrinsic mode component after Fourier transform; The complexity characteristics of each intrinsic mode component are calculated using the following formula: ; In the formula, C k The complexity of the k-th intrinsic mode component; f m Let S be the frequency value at the m-th discrete frequency point. m Let M be the energy or amplitude spectrum value corresponding to the m-th frequency point, where M is the total number of frequency points. After linearly normalizing the center frequency and complexity of each intrinsic mode component, a comprehensive score is obtained by weighted summation. Arranged in descending order according to the comprehensive score, each intrinsic mode component is divided into three groups, namely high-frequency subsequence, mid-frequency subsequence and low-frequency subsequence.

[0009] In one possible implementation of the first aspect, the first deep learning model is a bidirectional gated recurrent unit model, used to capture long-term trend information of low-frequency subsequences; the second deep learning model is a Transformer model, used to extract periodic patterns and interaction features of mid-frequency subsequences; and the third deep learning model is a bidirectional temporal convolutional network model, used to capture short-term fluctuations and local details of high-frequency subsequences.

[0010] In one possible implementation of the first aspect, the input format of the bidirectional gated recurrent unit model, the Transformer model, and the bidirectional temporal convolutional network model is a three-dimensional tensor of shape [batch_size, sequence_length, num_features], where sequence_length is the backtracking window length and num_features is the input feature dimension; the output format is a three-dimensional tensor of shape [batch_size, output_sequence_length, output_features].

[0011] In one possible implementation of the first aspect, the step of merging the low-frequency prediction result, the mid-frequency prediction result, and the high-frequency prediction result to obtain a wall temperature prediction sequence includes: The low-frequency prediction results, the mid-frequency prediction results, and the high-frequency prediction results are added point by point and superimposed to reconstruct the wall temperature prediction sequence.

[0012] In one possible implementation of the first aspect, the over-temperature risk determination of the wall temperature prediction sequence includes: The deviation and the rate of change of the wall temperature prediction sequence from the preset over-temperature threshold are calculated. When the deviation is greater than zero and the rate of change of the deviation shows an upward trend, it is determined that there is an over-temperature risk.

[0013] In one possible implementation of the first aspect, the adjustment of the desuperheating water flow rate to implement proactive control includes: The deviation between the predicted wall temperature sequence and the preset over-temperature threshold is used as a feedforward signal and superimposed into the control logic of the DCS system. The DCS system generates a desuperheating water flow correction requirement by the main controller based on the deviation between the current wall temperature and the preset overheating threshold and the wall temperature change rate. The auxiliary controller outputs a valve opening command based on the correction requirement and the measured signal of the desuperheater outlet temperature or flow. The electric valve positioner converts the electrical signal output by the DCS system into a pneumatic signal, which then drives the actuator to adjust the opening of the desuperheating water regulating valve to control the flow rate of the desuperheating water.

[0014] One possible implementation of the first aspect also includes: The deviation between the actual wall temperature value and the predicted wall temperature sequence is used as a feedback quantity. The parameters of the first deep learning model, the second deep learning model, and the third deep learning model are incrementally fine-tuned according to a preset period to match the current dynamic characteristics of the equipment.

[0015] Compared with the prior art, the present invention has at least the following beneficial effects: The wall temperature prediction and control method for rapidly changing load conditions provided by this invention decomposes the original wall temperature sequence into three sub-sequences: high, medium, and low frequency. This achieves accurate separation of multi-scale time-series features. Three different deep learning models are used to predict low-frequency trends, medium-frequency cycles, and high-frequency disturbances, respectively. This ensures that each frequency band feature receives optimal model processing, significantly improving the accuracy and stability of wall temperature prediction under rapidly changing load conditions. Simultaneously, the prediction results are superimposed as feedforward signals into the DCS control logic, triggering proactive adjustment actions before overheating occurs. This effectively overcomes the lag in response of traditional "post-event adjustment" and realizes a shift from "passive response" to "active pre-adjustment" control mode.

[0016] Secondly, the present invention provides a wall temperature prediction and control system suitable for rapidly changing load conditions, comprising: The data acquisition and filtering module is used to acquire wall temperature and related operating data in the DCS system of coal-fired power units, use the related operating data as input variables, perform correlation analysis between the input variables and the wall temperature, filter out the input variables that meet the preset correlation threshold, and construct a wall temperature prediction input dataset. The signal decomposition and reconstruction module is used to perform variational mode decomposition on the wall temperature sequence to be predicted, and obtain a preset number of intrinsic mode components; based on the frequency domain characteristics and complexity characteristics of each intrinsic mode component, it is reconstructed into high-frequency subsequence, mid-frequency subsequence and low-frequency subsequence. The frequency division prediction module is used to concatenate the wall temperature prediction input dataset with the high-frequency subsequence, the mid-frequency subsequence, and the low-frequency subsequence to construct high-frequency prediction sample sets, mid-frequency prediction sample sets, and low-frequency prediction sample sets respectively; a first deep learning model is used to predict the low-frequency prediction sample set, a second deep learning model is used to predict the mid-frequency prediction sample set, and a third deep learning model is used to predict the high-frequency prediction sample set to obtain low-frequency prediction results, mid-frequency prediction results, and high-frequency prediction results; The result reconstruction module is used to merge the low-frequency prediction result, the mid-frequency prediction result and the high-frequency prediction result to obtain the wall temperature prediction sequence. The advanced control module is used to determine the risk of overheating in the predicted wall temperature sequence and adjust the desuperheating water flow rate according to the determination result to implement advanced control.

[0017] Thirdly, the present invention provides an electronic device comprising: at least one processor and at least one memory, wherein the memory stores computer-readable instructions; the computer-readable instructions are executed by one or more of the processors to cause the electronic device to implement the wall temperature prediction and control method for rapidly changing load conditions as described in any implementation of the first aspect.

[0018] Fourthly, the present invention provides a storage medium having a computer-executable program stored thereon, the computer-executable program being used to cause a computer to execute a wall temperature prediction and control method suitable for rapidly changing load conditions, as in any implementation of the first aspect.

[0019] Understandably, the beneficial effects achieved by the system of the second aspect, the electronic device of the third aspect, and the storage medium of the fourth aspect provided above can be referred to in light of the beneficial effects of the first aspect and any of its possible design embodiments, which will not be repeated here. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention; Figure 2 This is a flowchart of a wall temperature prediction and control method applicable to rapid load change conditions provided by an embodiment of the present invention; Figure 3 This is a schematic diagram of the results of Pearson correlation analysis; Figure 4 This is a schematic diagram of intrinsic mode component reconstruction. Figure 5 This is a schematic diagram of the wall temperature prediction sequence; Figure 6 This is a structural block diagram of a wall temperature prediction and control system suitable for rapid load changes, provided in an embodiment of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be described below with reference to the accompanying drawings. In the description of the present invention, unless otherwise stated, " / " indicates that the objects before and after are in an "or" relationship. For example, A / B can represent A or B. The "or" in the present invention is merely a description of the relationship between the related objects, indicating that three relationships can exist. For example, A or B can represent: A alone, A and B simultaneously, and B alone. A and B can be singular or plural. Furthermore, in the description of the present invention, unless otherwise stated, "multiple" refers to two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items.

[0023] Furthermore, to facilitate a clear description of the technical solutions of the embodiments of the present invention, the terms "first" and "second" are used in the embodiments of the present invention to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and the terms "first" and "second" are not necessarily different.

[0024] In this embodiment of the invention, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" or "for example" in this embodiment of the invention should not be construed as superior or more advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner for ease of understanding.

[0025] Against the backdrop of the "dual carbon" goals driving the power system's transformation towards a new energy-based system, thermal power generating units have shifted from traditional power supply providers to core peak-shaving and backup power sources for new energy generation. This necessitates frequent and rapid load changes. The temperature of boiler superheaters, reheaters, and other heating surfaces is a core monitoring parameter for ensuring safe unit operation, making accurate prediction and effective control a key technical challenge under peak-shaving conditions. Current mainstream wall temperature monitoring and control methods have significant limitations. Traditional heat transfer mechanism models, using fixed parameters under steady-state conditions, cannot match the dynamic changes in flue gas flow rate, velocity, and heat exchange conditions within the furnace during rapid load changes, making reliable control decisions difficult. Furthermore, traditional "post-event adjustment" control modes have a 10-30 second response lag, easily leading to temporary wall temperature overheating. Long-term, repeated overheating accelerates pipe aging and can even cause pipe rupture.

[0026] In view of this, on the one hand, embodiments of the present invention provide a wall temperature prediction and control method applicable to rapid load change conditions, including: acquiring wall temperature and related operating data in the DCS system of a coal-fired power unit; performing correlation analysis between input variables and wall temperature; selecting input variables that meet a preset correlation threshold; constructing a wall temperature prediction input dataset; performing variational mode decomposition on the wall temperature sequence to be predicted to obtain a preset number of intrinsic mode components; reconstructing the intrinsic mode components into high-frequency subsequences, mid-frequency subsequences, and low-frequency subsequences based on their frequency domain characteristics and complexity characteristics; and connecting the wall temperature prediction input dataset with the high-frequency subsequences and the mid-frequency subsequences, respectively. The low-frequency subsequence is concatenated with features to construct a high-frequency prediction sample set, a mid-frequency prediction sample set, and a low-frequency prediction sample set. A first deep learning model is used to predict the low-frequency prediction sample set, a second deep learning model is used to predict the mid-frequency prediction sample set, and a third deep learning model is used to predict the high-frequency prediction sample set, resulting in low-frequency, mid-frequency, and high-frequency prediction results. The low-frequency, mid-frequency, and high-frequency prediction results are then merged to obtain a wall temperature prediction sequence. The wall temperature prediction sequence is then used to determine the risk of overheating, and the desuperheating water flow rate is adjusted based on the determination result to implement proactive control.

[0027] This invention provides a wall temperature prediction and control method suitable for rapid load change conditions. By decomposing the original wall temperature sequence into three sub-sequences—high, medium, and low frequencies—it achieves accurate separation of multi-scale time-series features. Three different deep learning models are used to specifically predict low-frequency trends, medium-frequency cycles, and high-frequency disturbances, ensuring that each frequency band feature receives optimal model processing. This significantly improves the accuracy and stability of wall temperature prediction under rapid load change conditions. Simultaneously, the prediction results are superimposed as feedforward signals into the DCS control logic, triggering proactive adjustment actions before overheating occurs. This effectively overcomes the lag in response of traditional "post-event adjustment" and realizes a shift from "passive response" to "active pre-adjustment" control mode.

[0028] In some embodiments, the wall temperature prediction and control method for rapidly changing load conditions provided by the present invention can be executed by any electronic device 20 with data processing capabilities, such as a general-purpose computer, personal computer, laptop computer, switch, or tablet computer. The specific implementation of the electronic device 20 is not limited here.

[0029] Figure 1 A schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention is shown. The electronic device 20 includes a processor 210, a memory 220, and a communication interface 230.

[0030] Processor 210 may include one or more processing cores. Processor 210 connects to various parts within electronic device 20 using various interfaces and lines, and performs various functions and processes data of electronic device 20 by running or executing instructions, programs, code sets, or instruction sets stored in memory 220, and by calling data stored in memory 220. Optionally, processor 210 may be implemented using at least one of the following hardware forms: Central Processing Unit (CPU), Graphics Processing Unit (GPU), Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), and Programmable Logic Array (PLA).

[0031] The memory 220 may include random access memory (RAM) or read-only memory (ROM). Optionally, the memory 220 may include a non-transitory computer-readable storage medium. The memory 220 may be used to store instructions, programs, code, code sets, or instruction sets. The memory 220 may include a program storage area. This program storage area may store instructions for implementing an operating system, instructions for implementing at least one function, instructions for implementing the various method embodiments described above, etc.

[0032] Communication interface 230 is used to communicate with other devices, equipment or communication networks, such as data storage devices, image processing devices or Ethernet, wireless access network (RAN), wireless local area network (WLAN), etc.

[0033] In terms of physical implementation, the aforementioned devices (such as processor 210, memory 220, and communication interface 230) can each be devices within the same device (such as a laptop computer). Alternatively, at least two of these devices can be located within the same device, i.e., as different devices within the same device, similar to the deployment of devices or components in a distributed system.

[0034] It is understood that the structure illustrated in this embodiment does not constitute a specific limitation on the electronic device 20. In other embodiments of the present invention, the electronic device 20 may include more or fewer components than illustrated, or combine some components, or split some components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0035] The following description, in conjunction with the accompanying drawings, illustrates a wall temperature prediction and control method suitable for rapidly changing load conditions provided by an embodiment of the present invention.

[0036] like Figure 2 As shown, this embodiment of the invention provides a wall temperature prediction and control method suitable for rapidly changing load conditions, which may include, but is not limited to: S1: Obtain wall temperature and related operating data from the DCS system of the coal-fired power unit, use the related operating data as input variables, perform correlation analysis between the input variables and the wall temperature, filter out the input variables that meet the preset correlation threshold, and construct a wall temperature prediction input dataset.

[0037] Specifically, taking a 350MW coal-fired power unit in a power plant as an example, this embodiment of the invention acquires wall temperature and related operating data from the DCS (Distributed Control System) of the coal-fired power unit. The DCS historical data covers one week, with a sampling interval of 20 seconds. Rapid load change conditions are defined as not less than 3% Pe / min. The original DCS data has no missing or outlier values. The initially acquired data is shown in the table below: Using the relevant operational data as input variables, a correlation analysis is performed between the input variables and the wall temperature. Input variables that meet a preset correlation threshold are selected to construct a wall temperature prediction input dataset. This correlation analysis is used to quantitatively evaluate the degree of association between each operational parameter and wall temperature changes, and to eliminate redundant and interfering data, thereby improving the effectiveness and reliability of the input data for subsequent prediction models.

[0038] In one feasible implementation, the correlation analysis in this embodiment of the invention is a Pearson correlation analysis. The Pearson correlation coefficient is used to measure the degree of linear correlation between any input variable and wall temperature, and its calculation formula is as follows: ; In the formula, x i Let y be the i-th sample value of any input variable; i This is the i-th sample value of the wall temperature; The sample mean of the input variable. is the sample mean of the wall temperature; n is the number of samples. The coefficient ranges from [-1, 1], with the absolute value closer to 1 indicating a stronger linear correlation, and 0 indicating no linear correlation.

[0039] Figure 3 This is the analysis result of Pearson correlation analysis between the input variables and wall temperature of the coal-fired power unit in an embodiment of the present invention. In this embodiment, variables with an absolute Pearson correlation coefficient greater than 0.8 (i.e., a preset correlation threshold) are selected as model input features to construct a wall temperature prediction input dataset.

[0040] It should be noted that, in addition to using the Pearson correlation coefficient, embodiments of the present invention may also select other types of correlation analysis methods based on data characteristics, such as: Spearman's rank correlation coefficient: Applicable to ordinal variables or continuous data that do not follow a normal distribution, it is used to measure the monotonic relationship between two variables.

[0041] Kendall's rank correlation coefficient: Applicable to ordinal data or data containing a large number of ties, used to measure monotonic relationships.

[0042] Distance correlation coefficient: suitable for detecting linear and nonlinear relationships.

[0043] Maximum Information Coefficient (MIC): It can detect linear and non-linear dependencies between pairs of variables.

[0044] In practical applications, based on factors such as the distribution characteristics of the data, the type of variable (continuous / discrete), and the presence of outliers, any of the above correlation analysis methods or a combination of methods can be selected to screen out input variables that are strongly correlated with changes in wall temperature. No restrictions are imposed here.

[0045] S2: Perform Variational Mode Decomposition (VMD) on the wall temperature sequence to be predicted to obtain a preset number of Intrinsic Mode Functions (IMFs); based on the frequency domain characteristics and complexity characteristics of each IMF, reconstruct it into high-frequency subsequences, mid-frequency subsequences, and low-frequency subsequences.

[0046] VMD is an adaptive, fully non-recursive modal variational and signal processing method that can decompose non-stationary signals into several modal components with sparse properties.

[0047] Specifically, we establish a constrained variational problem, whose expression is: ; In the formula, Let be the partial derivative with respect to time t; ω is the Dirac function; k (k=1,2,…,K) represents the center frequency of the k-th mode; denoted as each modal component obtained from the decomposition; j is the imaginary unit; ∗ is the convolution operator; f(t) is the wall temperature sequence to be predicted; and K is the number of decomposed modes.

[0048] Introducing the penalty factor α and the Lagrange multiplier λ(t), we construct the augmented Lagrange expression: ; The augmented Lagrangian expression is solved using the alternating direction method of multipliers, and the frequency domain iterative update formula for each modal component is as follows: ; ; ; In the formula, ω is the frequency domain variable; n is the current iteration number; and τ is the update step size parameter. , and f(t), λ(t), and u are respectively i (t) The frequency domain representation obtained by Fourier transform.

[0049] In this embodiment of the invention, the penalty factor α for VMD was determined to be 2317 and the number of decomposition modes K to be 10 through manual iterative experiments. Under this parameter combination, there is no aliasing of the center frequencies of each IMF component, the decomposition residual is less than the preset threshold, and the decomposition effect is optimal.

[0050] After obtaining each IMF component, based on the frequency domain characteristics and complexity characteristics of each intrinsic mode component, it is reconstructed into high-frequency subsequence, mid-frequency subsequence and low-frequency subsequence, to realize multi-scale feature extraction of the wall temperature signal.

[0051] In one feasible implementation, the frequency domain feature in this embodiment of the invention is the center frequency of the intrinsic mode components. The center frequency is obtained by calculating the power-weighted average frequency of each intrinsic mode component using Fourier transform, and the calculation formula is as follows: ; In the formula, ω k ω is the center frequency of the k-th intrinsic mode component; ω is a frequency domain variable; This is the frequency domain representation of the k-th intrinsic mode component after Fourier transform.

[0052] It should be noted that in this embodiment of the invention, the center frequency is used as the frequency domain characteristic of each IMF component. In addition, the following frequency domain characteristic parameters can be selected as alternatives or supplements based on signal characteristics and application requirements: Peak frequency: The frequency point in the spectrum with the largest amplitude or power density, reflecting the location of the strongest frequency component in the signal.

[0053] Mean frequency: defined as the first moment of the power spectral density, that is, the weighted average of frequencies across the entire frequency band, reflecting the centroid of the frequency distribution of signal energy.

[0054] Frequency standard deviation: measures the dispersion of signal frequency components around the mean frequency. A larger bandwidth indicates that the frequency components of the signal are more dispersed; a smaller bandwidth indicates that the signal energy is more concentrated in a narrow frequency range. This parameter can reflect the concentration of energy in the frequency band of each IMF component and help determine the purity of the component.

[0055] In practical applications, one or more combinations of the above features can be selected as frequency domain features based on the physical meaning and signal characteristics of each IMF component, and these features, along with complexity features, can participate in frequency band allocation. The specific selection method can be determined through experimentation or experience, and is not limited here.

[0056] The complexity feature is calculated using the following formula: ; In the formula, C k The complexity of the k-th intrinsic mode component; f m Let S be the frequency value at the m-th discrete frequency point. m Let M be the energy or amplitude spectrum value corresponding to the m-th frequency point, where M is the total number of frequency points.

[0057] The reconstruction rule is as follows: after linearly normalizing the center frequency and complexity of each intrinsic mode component, the comprehensive score Q is obtained by weighted summation. k : ; In the formula, norm() represents the linear normalization function. Based on the comprehensive score Q k Arranged in descending order, each intrinsic mode component is divided into three groups, namely high-frequency subsequence, mid-frequency subsequence, and low-frequency subsequence. Figure 4 This is a schematic diagram of intrinsic mode component reconstruction according to an embodiment of the present invention.

[0058] It should be noted that the above weight ratios (center frequency weight 0.7, complexity weight 0.3) were determined through parameter optimization on the validation set. Using frequency features as the primary factor and complexity as an auxiliary adjustment maximizes the discriminative power of samples within each frequency band, achieving optimal classification results while balancing the objectivity and reproducibility of frequency band division.

[0059] S3: The wall temperature prediction input dataset is concatenated with the high-frequency subsequence, the mid-frequency subsequence, and the low-frequency subsequence to construct high-frequency prediction sample sets, mid-frequency prediction sample sets, and low-frequency prediction sample sets, respectively. A first deep learning model is used to predict the low-frequency prediction sample set, a second deep learning model is used to predict the mid-frequency prediction sample set, and a third deep learning model is used to predict the high-frequency prediction sample set, to obtain low-frequency prediction results, mid-frequency prediction results, and high-frequency prediction results.

[0060] In one feasible implementation, the first deep learning model in the present invention may, but is not limited to, adopt a bidirectional gated recurrent unit (BiGRU) model, the second deep learning model may, but is not limited to, adopt a Transformer model, and the third deep learning model may, but is not limited to, adopt a bidirectional temporal convolutional network (BiTCN) model.

[0061] Specifically, in this embodiment of the invention, the BiGRU model consists of two layers of GRU units, forward and backward. It effectively captures long-term dependencies through update gates and reset gates. The hidden state of the last time step is taken and mapped to the output dimension through a fully connected layer to capture the long-term trend information of low-frequency subsequences.

[0062] The Transformer model, based on a multi-head self-attention mechanism and stacked feedforward networks, supplemented by positional encoding and residual connections, can model long-range dependencies between any time steps in a sequence in parallel, and is used to extract periodic patterns and interaction features of mid-frequency subsequences. In this embodiment of the invention, the Transformer model includes a 4-layer encoder and a 4-head self-attention mechanism.

[0063] The BiTCN model consists of two independent but structurally symmetrical temporal convolutional networks, which perform causal dilated convolutions in both forward and reverse temporal directions to expand the receptive field and extract local temporal features. In this embodiment of the invention, BiTCN uses dilated convolution with a dilation coefficient sequence of 1, 2, 4, 8, 16. The outputs from both directions are concatenated and then passed through a fully connected layer to generate the final prediction result, which is used to capture short-term fluctuations and local details of high-frequency subsequences.

[0064] In one feasible implementation, the input format of the three models described above in this invention is a three-dimensional tensor of shape [batch_size, sequence_length, num_features]. Here, sequence_length is the backtracking window length (number of historical time steps), such as 60 steps; num_features consists of two parts: the first part is the wall temperature sequence to be predicted (subsequence values ​​corresponding to the frequency band), with a dimension of 1; the second part is auxiliary operating parameters strongly correlated with wall temperature after correlation analysis (such as total coal quantity, water flow rate, total air volume, etc.), with a dimension equal to the number of variables after filtering. Together, these form the input feature vector for each time step. The output format is also a three-dimensional tensor of shape [batch_size, output_sequence_length, output_features]. Here, output_sequence_length is the output window length (number of future time steps to predict), such as 12 steps, and the output feature dimension is 1, indicating that one wall temperature value is predicted for each output time step.

[0065] In this embodiment of the invention, all three models are trained using the Adam optimizer with an initial learning rate of 0.001 and a segmented decay strategy (e.g., the learning rate is decayed to 0.5 times the current value every 50 training epochs). The loss function is mean squared error (MSE), the batch size is set to 64, and the maximum number of training epochs is 200.

[0066] To prevent overfitting, this embodiment of the invention introduces an early stopping mechanism, terminating training prematurely when the validation set loss no longer decreases after 20 consecutive epochs. Furthermore, Dropout layers are added after the key layers of each model: the BiGRU model adds a Dropout layer after the GRU layer; the Transformer model adds a Dropout layer after each multi-head self-attention layer and feedforward network layer; and the BiTCN model adds a Dropout layer after each dilated convolutional layer. The Dropout rate is determined by tuning within the range of 0.1 to 0.3 based on the validation set performance.

[0067] S4: The low-frequency prediction results, the mid-frequency prediction results and the high-frequency prediction results are merged to obtain the wall temperature prediction sequence.

[0068] In one feasible implementation, embodiments of the present invention add the low-frequency prediction results, the mid-frequency prediction results, and the high-frequency prediction results point by point and then superimpose and reconstruct them to obtain a wall temperature prediction sequence, such as... Figure 5 As shown in the figure. This method can ensure that the reconstructed signal not only fully preserves the multi-scale features of the original data, but also effectively suppresses the cumulative propagation of prediction errors in each frequency band.

[0069] In specific implementation, in addition to the above-mentioned point-by-point addition superposition reconstruction method, embodiments of the present invention may also select one of the following reconstruction strategies as alternatives based on the confidence level, error distribution, or dynamic characteristics of the prediction results for each frequency band: (1) Weighted summation reconstruction: Different weight coefficients are assigned to the low-frequency prediction results, the mid-frequency prediction results, and the high-frequency prediction results, respectively. Each weight coefficient is positive and the sum is 1. The weight coefficients can be determined based on the prediction accuracy of each frequency band prediction model on the validation set, with higher accuracy frequency bands being assigned greater weights. The expression for weighted summation reconstruction is: ; In the formula, This is the final reconstructed predicted wall temperature value. For low-frequency prediction results, This is the mid-frequency prediction result. For high-frequency prediction results, ω L ω M and ω H These are the corresponding weighting coefficients.

[0070] (2) Adaptive weighted reconstruction: The weight coefficients are not fixed values, but are dynamically adjusted according to the relative confidence of the prediction results of each frequency band at the current prediction time. For example, when the wall temperature is in a steady state, the confidence of the low-frequency prediction results is higher, so their weight can be increased; when the wall temperature is in a stage of violent fluctuation, the high-frequency prediction results contain more transient information, so their weight should be increased accordingly. The adaptive weights can be calculated in real time based on the historical error performance of the prediction results of each frequency band within the recent window.

[0071] (3) Reconstruction based on confidence intervals: Estimate the confidence intervals of the prediction results for low frequency, mid frequency and high frequency respectively. When reconstructing, prioritize the prediction values ​​of frequency bands with narrower confidence intervals (i.e., lower prediction uncertainty). For frequency bands with wider confidence intervals, use their prediction values ​​as auxiliary correction terms to reduce their contribution to the final reconstruction results.

[0072] S5: Determine the risk of overheating in the predicted wall temperature sequence, and adjust the desuperheating water flow rate according to the determination result to implement proactive control.

[0073] In one feasible implementation, the overheat risk determination of the wall temperature prediction sequence in this embodiment of the invention may include, but is not limited to: The deviation and the rate of change of the wall temperature prediction sequence from the preset over-temperature threshold are calculated. When the deviation is greater than zero and the rate of change of the deviation shows an upward trend, it is determined that there is an over-temperature risk.

[0074] Specifically, in this embodiment of the invention, the preset over-temperature threshold can be set to 590℃. In the specific determination, firstly, the deviation ΔT between the predicted temperature value at each future time in the wall temperature prediction sequence and 590℃ is calculated, along with the rate of change of deviation d(ΔT) / dt between adjacent predicted values. When ΔT > 0 and d(ΔT) / dt > 0, meaning the predicted wall temperature has exceeded the safety threshold and is still trending upwards, an over-temperature risk is determined, triggering an advance control command.

[0075] In one feasible implementation, the adjustment of the desuperheating water flow rate to implement proactive control in this embodiment of the invention may include, but is not limited to: The deviation between the predicted wall temperature sequence and the preset over-temperature threshold is used as a feedforward signal and superimposed into the control logic of the DCS system. The DCS system generates a desuperheating water flow correction requirement by the main controller based on the deviation between the current wall temperature and the preset overheating threshold and the wall temperature change rate. The auxiliary controller outputs a valve opening command based on the correction requirement and the measured signal of the desuperheater outlet temperature or flow. The electric valve positioner converts the electrical signal output by the DCS system into a pneumatic signal, which then drives the actuator to adjust the opening of the desuperheating water regulating valve to control the flow rate of the desuperheating water.

[0076] This invention employs a predictive feedforward-based cascade composite control strategy to construct a dual-channel regulation architecture of "feedforward + feedback". The feedforward channel outputs regulation commands in advance based on the predicted wall temperature sequence, compensating for disturbances before they affect the controlled variable. The feedback channel performs closed-loop correction of the measured wall temperature value through a cascade loop composed of the main and secondary controllers. This composite control architecture balances the speed of feedforward control with the robustness of feedback control, making it particularly suitable for quality control of thermal objects with large time lags and inertia.

[0077] Specifically, the deviation between the predicted wall temperature sequence and the preset over-temperature threshold is used as a feedforward control quantity, which is superimposed on the output of the main controller of the original cascade control loop in the DCS system to form dynamic feedforward compensation. The feedforward control quantity is calculated in real time based on the deviation and rate of change of the predicted wall temperature sequence, anticipating the actual wall temperature change response to compensate for the large lag and inertia characteristics of the desuperheating water regulating channel. The calculation of the feedforward compensation quantity follows these principles: when the predicted deviation is positive and increasing, the feedforward compensation quantity acts positively on the setpoint of the secondary controller, increasing the opening of the desuperheating water regulating valve in advance; conversely, it decreases the opening or maintains the current opening.

[0078] In the feedback channel, the DCS system, based on the deviation between the current wall temperature and the preset over-temperature threshold and the rate of wall temperature change, generates a desuperheating water flow correction requirement from the main controller (wall temperature controller), which serves as the main control command for the cascade control system. The secondary controller (flow regulator), based on the correction requirement and the measured signals of the desuperheater outlet temperature or flow rate, quickly eliminates internal disturbances in the desuperheating water flow and outputs a precise valve opening command. The 4-20mA electrical signal output from the DCS system is converted into a pneumatic pressure signal by the electric valve positioner, which then drives a pneumatic diaphragm actuator or an electric actuator to adjust the opening of the desuperheating water regulating valve to control the desuperheating water flow rate.

[0079] Through the aforementioned composite control of "feedforward compensation + cascade feedback", the system implements advance adjustment before disturbance occurs and quickly eliminates deviation through cascade loop after disturbance occurs, thereby achieving precise and stable control of wall temperature.

[0080] Given that during long-term operation of thermal power units, factors such as coking and ash accumulation on heating surfaces, pipe aging, and changes in load characteristics can cause slow, time-varying dynamic characteristics of the system, the prediction accuracy of offline-trained prediction models will gradually decrease if they are not updated over a long period. To address this issue, this invention introduces an online incremental fine-tuning mechanism to enable the prediction model to continuously track the current dynamic characteristics of the unit, as detailed below: Online incremental fine-tuning is triggered according to a preset cycle. The DCS system automatically initiates the incremental fine-tuning process once every preset time window (e.g., every hour). The preset cycle can be dynamically adjusted according to the frequency of unit load changes; for example, the cycle can be shortened when load changes are frequent (e.g., every 30 minutes) and extended when the load is stable (e.g., every 2 hours). The cycle parameter is a system configuration item and can be modified online by operators.

[0081] At the end of each fine-tuning cycle, the DCS system collects the sequence of actual wall temperature values ​​recorded by the wall temperature measuring transmitter during that time period, as well as the sequence of predicted wall temperature values ​​output by each prediction model at the corresponding time. The deviation sequence between the actual wall temperature values ​​and the predicted values ​​is calculated, which constitutes the feedback quantity for that cycle.

[0082] Using the aforementioned deviation sequence as the loss signal, the parameters of the three prediction models are incrementally updated using Mini-batch Gradient Descent or its variants.

[0083] The specific update method can be, but is not limited to, as follows: For the three prediction models (low-frequency, mid-frequency, and high-frequency), calculate the prediction bias corresponding to their respective outputs, and apply a small adjustment to the model weight parameters along the negative gradient direction of the loss function. The update amount is controlled by the learning rate. In this embodiment, the incremental learning rate is set to 1 / 10 to 1 / 5 of the initial training learning rate (i.e., 0.0001 to 0.0002) to ensure that the introduction of new knowledge does not destroy the stable features already learned by the model. Each incremental update only performs a few gradient iterations (e.g., 1 to 3 times) to avoid overfitting to historical knowledge.

[0084] Incremental fine-tuning of the three models can be performed sequentially within each cycle, that is, updating the parameters of the BiGRU, Transformer and BiTCN models in turn on the same batch of deviation data, or it can be performed in parallel, depending on the computing power configuration of the DCS system, which is not limited here.

[0085] In practical implementation, the online incremental fine-tuning process of this invention incorporates a built-in safety constraint: when the prediction deviations for multiple consecutive periods exhibit a systematic deviation in the same direction (e.g., the deviation remains positive and gradually increases), the system triggers a diagnostic alarm, prompting operators to check the status of sensors or equipment, rather than adjusting model parameters indefinitely. This mechanism avoids irreversible damage to the model caused by erroneous feedback due to sensor failure or equipment malfunction.

[0086] The wall temperature prediction and control method for rapidly changing load conditions provided in the embodiments of the present invention has the following beneficial effects: (1) This embodiment of the invention decomposes the original single wall temperature sequence into three sub-sequences—high, medium, and low frequency—through VMD, achieving accurate separation of multi-timescale features of the wall temperature signal and overcoming the feature confusion problem caused by the traditional single model uniformly modeling all features. Based on this, this embodiment of the invention adopts a differentiated model matching strategy: low-frequency trends are handled by BiGRU, which excels at capturing long-term dependencies; medium-frequency periodic patterns are handled by Transformer, which excels at global interaction modeling; and high-frequency local details are handled by BiTCN, which excels at expanding the receptive field. This differentiated prediction architecture of "frequency band and model" ensures that each frequency band feature receives model processing that best matches its own timescale. Each model only needs to focus on the feature patterns of its own frequency band, avoiding the problem of insufficient fitting of different frequency band features by a single model due to structural limitations, and significantly improving the overall prediction accuracy.

[0087] (2) This embodiment of the invention links the overheating risk assessment based on the prediction results with the adjustment of the cooling water flow rate, so that the control action does not depend on the actual wall temperature exceeding the limit before triggering (traditional "post-event adjustment" has a response lag of 10-30 seconds), but is based on the prediction sequence to know the future wall temperature change trend in advance, and completes the generation and execution of the adjustment command before the overheating occurs. At the same time, this embodiment of the invention uses the wall temperature prediction deviation as a feedforward signal superimposed on the original cascade control loop of the DCS system to form a "feedforward + feedback" dual-channel composite control architecture—the feedforward channel outputs compensation commands in advance based on the prediction information, and implements advance adjustment before the disturbance affects the wall temperature; the feedback channel performs closed-loop correction of the measured wall temperature value through the cascade loop composed of the main and auxiliary controllers. The above-mentioned feedforward compensation mechanism effectively overcomes the inherent lag defect of traditional feedback control and realizes the transformation from "passive response" to "active pre-adjustment" control mode.

[0088] (3) In this embodiment of the invention, the anti-thermal water regulating valve receives the opening adjustment command before the actual wall temperature rises through the feedforward channel, thereby increasing the anti-thermal water flow rate in advance and intervening before overheating occurs, effectively reducing the peak overshoot of the wall temperature. Through the closed-loop correction of the cascade feedback channel of the main and auxiliary regulators, residual deviations such as internal disturbances in the anti-thermal water flow rate are quickly eliminated, accelerating the convergence process of the wall temperature returning to the safe threshold. Under rapid load change conditions, the dual effect of the above-mentioned "anti-thermal intervention + rapid convergence" ensures that the wall temperature is always controlled within a narrow fluctuation range near the safe threshold, avoiding the problem of large fluctuations in wall temperature under the traditional lag regulation mode, and effectively suppressing the fluctuation amplitude of the wall temperature.

[0089] (4) In the embodiments of the present invention, S1 to S5 constitute a complete technical chain of "data dimensionality reduction to multi-scale decomposition to frequency division prediction to advanced control". There is a close causal relationship and information transmission between each step: the key features selected in S1 are fused with the subsequences of each frequency band in S3 to ensure the input quality of the prediction model; the multi-scale features decomposed in S2 are processed by the differentiated model in S3 to ensure the fitting accuracy of each frequency band feature; the prediction results of S3 are converted into feedforward control commands in S5 to realize the organic connection between prediction and control. The combination of the above steps produces synergistic gains: high-precision prediction provides a reliable basis for advanced control, and advanced control reduces the damage of drastic temperature fluctuations to the stability of the input data of the prediction model. The two mutually enhance each other and form a positive cycle of "accurate prediction supports reliable control and reliable control ensures stable prediction".

[0090] (5) This embodiment of the invention introduces an online incremental fine-tuning mechanism, which uses the deviation between the actual wall temperature value and the predicted value as feedback, and performs small-batch gradient updates on the parameters of the three prediction models at a preset period, so that the model continuously tracks the slow time-varying characteristics of the unit's operation (such as dynamic characteristic drift caused by factors such as coking of the heating surface, ash accumulation, and pipe aging). Compared with periodic retraining, this mechanism achieves continuous optimization of model parameters with minimal computational overhead, overcomes the problem of the prediction accuracy gradually decreasing after long-term operation of offline models, and ensures the long-term stability of wall temperature prediction accuracy throughout the entire life cycle of the unit.

[0091] Based on the wall temperature prediction and control method for rapidly changing load conditions provided in the first aspect, embodiments of the present invention provide a wall temperature prediction and control system for rapidly changing load conditions, such as... Figure 6 As shown, the wall temperature prediction and control system suitable for rapid load change conditions includes: The data acquisition and filtering module 110 is used to acquire wall temperature and related operating data in the DCS system of coal-fired power units, perform correlation analysis between input variables and wall temperature, filter out the input variables that meet the preset correlation threshold, and construct a wall temperature prediction input dataset. The signal decomposition and reconstruction module 120 is used to perform variational mode decomposition on the wall temperature sequence to be predicted, and obtain a preset number of intrinsic mode components; based on the frequency domain characteristics and complexity characteristics of each intrinsic mode component, it is reconstructed into high-frequency subsequence, mid-frequency subsequence and low-frequency subsequence. The frequency division prediction module 130 is used to concatenate the wall temperature prediction input dataset with the high-frequency subsequence, the mid-frequency subsequence, and the low-frequency subsequence respectively to construct a high-frequency prediction sample set, a mid-frequency prediction sample set, and a low-frequency prediction sample set; a first deep learning model is used to predict the low-frequency prediction sample set, a second deep learning model is used to predict the mid-frequency prediction sample set, and a third deep learning model is used to predict the high-frequency prediction sample set to obtain low-frequency prediction results, mid-frequency prediction results, and high-frequency prediction results; The result reconstruction module 140 is used to merge the low-frequency prediction result, the mid-frequency prediction result and the high-frequency prediction result to obtain the wall temperature prediction sequence. The advanced control module 150 is used to determine the overheating risk of the wall temperature prediction sequence and adjust the desuperheating water flow rate according to the determination result to implement advanced control.

[0092] It should be noted that the data acquisition and filtering module 110, signal decomposition and reconstruction module 120, frequency division prediction module 130, result reconstruction module 140 and advance control module 150 in the embodiments of the present invention are all integrated into the DCS system or communicate with the DCS system.

[0093] Based on the wall temperature prediction and control method for rapidly changing load conditions provided in the first aspect, this embodiment of the invention also provides a storage medium storing a computer-executable program. The computer-executable program is used to cause a computer to execute the wall temperature prediction and control method for rapidly changing load conditions as described in any implementation of the first aspect. Explanations of the relevant content and descriptions of the beneficial effects of any of the computer-readable storage media provided above can be found in the corresponding embodiments described above, and will not be repeated here.

[0094] Those skilled in the art will understand that the program for implementing all or part of the steps of the above embodiments, which can be executed by a program instructing related hardware, can be stored in a computer-readable storage medium. The storage medium mentioned above can be a read-only memory, a random access memory, etc. The processing unit or processor mentioned above can be a central processing unit, a general-purpose processor, an application-specific integrated circuit (ASIC), a microprocessor (DSP), a field-programmable gate array (FPGA), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof.

[0095] This invention also provides a computer program product containing instructions that, when executed on a computer, cause the computer to perform any of the methods described in the above embodiments. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., SSD), etc.

[0096] It should be noted that the devices for storing computer instructions or computer programs provided in the embodiments of the present invention, such as, but not limited to, the aforementioned memory, computer-readable storage medium, and communication chip, are all non-transitory. Those skilled in the art should recognize that the functions described in the embodiments of the present invention in one or more of the above examples can be implemented using hardware, software, firmware, or any combination thereof. When implemented using software, these functions can be stored in a computer-readable storage medium or transmitted as one or more instructions or code on a computer-readable storage medium. Computer-readable storage media include computer storage media and communication media, wherein communication media include any medium that facilitates the transmission of computer programs from one place to another. Storage media can be any available medium accessible to general-purpose or special-purpose computers.

[0097] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for predicting and controlling wall temperature under rapid load changes, characterized in that, include: Obtain wall temperature and related operating data from the DCS system of a coal-fired power unit, use the related operating data as input variables, perform correlation analysis between the input variables and the wall temperature, filter out the input variables that meet the preset correlation threshold, and construct a wall temperature prediction input dataset. Variational mode decomposition is performed on the predicted wall temperature sequence to obtain a preset number of intrinsic mode components; Based on the frequency domain characteristics and complexity characteristics of each intrinsic mode component, it is reconstructed into high-frequency subsequence, mid-frequency subsequence and low-frequency subsequence; The wall temperature prediction input dataset is concatenated with the high-frequency subsequence, the mid-frequency subsequence, and the low-frequency subsequence to construct high-frequency prediction sample sets, mid-frequency prediction sample sets, and low-frequency prediction sample sets, respectively. A first deep learning model is used to predict the low-frequency prediction sample set, a second deep learning model is used to predict the mid-frequency prediction sample set, and a third deep learning model is used to predict the high-frequency prediction sample set, thereby obtaining low-frequency prediction results, mid-frequency prediction results, and high-frequency prediction results. The low-frequency prediction results, the mid-frequency prediction results, and the high-frequency prediction results are merged to obtain the wall temperature prediction sequence; The wall temperature prediction sequence is used to determine the risk of overheating, and the desuperheating water flow rate is adjusted according to the determination result to implement proactive control.

2. The wall temperature prediction and control method applicable to rapid load change conditions according to claim 1, characterized in that, The correlation analysis between the input variables and the wall temperature includes: A Pearson correlation analysis was performed between the input variable and the wall temperature. The formula for calculating the Pearson correlation coefficient is as follows: ； In the formula, x i Let y be the i-th sample value of any input variable; i This is the i-th sample value of the wall temperature; The sample mean of the input variable. is the sample mean of the wall temperature; n is the number of samples.

3. The wall temperature prediction and control method applicable to rapid load change conditions according to claim 1, characterized in that, The variational mode decomposition of the wall temperature sequence to be predicted includes: Establish a constrained variational problem, its expression is: ； In the formula, Let be the partial derivative with respect to time t; ω is the Dirac function; k The center frequency of the k-th mode; denoted as each modal component obtained from the decomposition; j is the imaginary unit; * is the convolution operator; f(t) is the wall temperature sequence to be predicted; and K is the number of decomposed modes. Introducing the penalty factor α and the Lagrange multiplier λ(t), we construct the augmented Lagrange expression: ； The augmented Lagrangian expression is solved using the alternating direction method of multipliers, and the frequency domain iterative update formula for each modal component is as follows: ； ； ； In the formula, ω is the frequency domain variable; n is the current iteration number; and τ is the update step size parameter. , and f(t), λ(t), and u are respectively i (t) The frequency domain representation obtained by Fourier transform.

4. The wall temperature prediction and control method applicable to rapid load change conditions according to claim 1, characterized in that, The frequency domain feature is the center frequency of the intrinsic mode component. The process of reconstructing the intrinsic mode component into high-frequency subsequences, mid-frequency subsequences, and low-frequency subsequences based on its frequency domain features and complexity characteristics includes: The center frequencies of each intrinsic mode component are calculated using Fourier transform, and the formula is as follows: ； In the formula, ω k ω is the center frequency of the k-th intrinsic mode component; ω is a frequency domain variable; The frequency domain representation of the k-th intrinsic mode component after Fourier transform; The complexity characteristics of each intrinsic mode component are calculated using the following formula: ； In the formula, C k The complexity of the k-th intrinsic mode component; f m Let S be the frequency value at the m-th discrete frequency point. m Let M be the energy or amplitude spectrum value corresponding to the m-th frequency point, where M is the total number of frequency points. After linearly normalizing the center frequency and complexity of each intrinsic mode component, a comprehensive score is obtained by weighted summation. Arranged in descending order according to the comprehensive score, each intrinsic mode component is divided into three groups, namely high-frequency subsequence, mid-frequency subsequence and low-frequency subsequence.

5. The wall temperature prediction and control method applicable to rapid load change conditions according to claim 1, characterized in that, The first deep learning model is a bidirectional gated recurrent unit model, used to capture long-term trend information of low-frequency subsequences; the second deep learning model is a Transformer model, used to extract periodic patterns and interaction features of mid-frequency subsequences; and the third deep learning model is a bidirectional temporal convolutional network model, used to capture short-term fluctuations and local details of high-frequency subsequences.

6. The wall temperature prediction and control method applicable to rapid load change conditions according to claim 5, characterized in that, The input format of the bidirectional gated recurrent unit model, the Transformer model, and the bidirectional temporal convolutional network model is a three-dimensional tensor of shape [batch_size, sequence_length, num_features], where sequence_length is the backtracking window length and num_features is the input feature dimension; the output format of the model is a three-dimensional tensor of shape [batch_size, output_sequence_length, output_features].

7. The wall temperature prediction and control method applicable to rapid load change conditions according to claim 1, characterized in that, The step of merging the low-frequency prediction results, the mid-frequency prediction results, and the high-frequency prediction results to obtain the wall temperature prediction sequence includes: The low-frequency prediction results, the mid-frequency prediction results, and the high-frequency prediction results are added point by point and superimposed to reconstruct the wall temperature prediction sequence.

8. The wall temperature prediction and control method applicable to rapid load change conditions according to claim 1, characterized in that, The process of determining the overheating risk of the predicted wall temperature sequence includes: The deviation and the rate of change of the wall temperature prediction sequence from the preset over-temperature threshold are calculated. When the deviation is greater than zero and the rate of change of the deviation shows an upward trend, it is determined that there is an over-temperature risk.

9. A wall temperature prediction and control method suitable for rapid load change conditions according to claim 8, characterized in that, The adjustment of the desuperheating water flow rate to implement proactive control includes: The deviation between the predicted wall temperature sequence and the preset over-temperature threshold is used as a feedforward signal and superimposed into the control logic of the DCS system. The DCS system generates a desuperheating water flow correction requirement by the main controller based on the deviation between the current wall temperature and the preset overheating threshold and the wall temperature change rate. The auxiliary controller outputs a valve opening command based on the correction requirement and the measured signal of the desuperheater outlet temperature or flow. The electric valve positioner converts the electrical signal output by the DCS system into a pneumatic signal, which then drives the actuator to adjust the opening of the desuperheating water regulating valve to control the flow rate of the desuperheating water.

10. A method for predicting and controlling wall temperature under rapidly changing load conditions according to claim 1, characterized in that, Also includes: The deviation between the actual wall temperature value and the predicted wall temperature sequence is used as a feedback quantity. The parameters of the first deep learning model, the second deep learning model, and the third deep learning model are incrementally fine-tuned according to a preset period to match the current dynamic characteristics of the equipment.

Images