A deep learning-based intelligent prediction control method and system for boiler main steam temperature

Abstract

The application discloses a kind of based on deep learning's boiler main steam temperature intelligent prediction control method and system, it is related to thermal power plant boiler control technical field, the method includes the following steps: acquisition boiler operating condition multidimensional time series data, the multidimensional time series data is preprocessed and feature extraction, obtain time series characteristic variable;The time series characteristic variable is input into pre-trained temperature prediction model, and the main steam temperature prediction value of future multiple time steps is output;By minimizing the deviation between the main steam temperature prediction value of future multiple time steps and main steam temperature preset value, rolling optimization is solved in prediction time domain, and the optimal temperature reducing water regulating door control instruction is obtained, realizes boiler main steam temperature intelligent prediction control.The application can solve the problem that prior art cannot accurately predict the trend of main steam temperature change.

Description

Technical Field

[0001] This invention relates to the field of boiler control technology in thermal power plants, specifically to a method and system for intelligent prediction and control of boiler main steam temperature based on deep learning. Background Technology

[0002] In modern thermal power plants, precise control of the boiler's main steam temperature is one of the key technologies for ensuring the safe and economical operation of the unit. The main steam temperature directly affects the turbine's thermal efficiency and equipment safety; excessively high temperatures can lead to overheating and damage to the superheater tubes, while excessively low temperatures reduce the unit's thermal efficiency. Therefore, controlling the main steam temperature within ±5℃ of the design value is a basic requirement for power plant operation. Traditional main steam temperature control mainly relies on the desuperheating water injection system, adjusting the opening of the primary and secondary desuperheating water regulating valves to control the amount of desuperheating water injected into the superheater, thereby regulating the main steam temperature. Existing control methods mainly include: First, traditional PID control, which uses a proportional-integral-derivative controller, taking the main steam temperature deviation as input and outputting a desuperheating water regulating valve opening command. However, due to the characteristics of the boiler system, such as large delay, large inertia, nonlinearity, and time-varying nature, traditional PID control often has a lag in response and low control accuracy. Second, cascade control systems employ a cascade structure of primary and secondary controllers. The primary controller uses the main steam temperature as the controlled variable, while the secondary controller uses the desuperheating water flow rate or intermediate temperature as the controlled variable. Although this improves control quality, it still struggles to handle complex operating conditions. Third, feedforward-feedback control adds feedforward control to feedback control, using unit load, fuel quantity, etc., as feedforward signals. However, feedforward models are typically based on linearization assumptions, making it difficult to accurately describe complex nonlinear relationships.

[0003] The main problems with existing technologies include: First, the long heat transfer path from the combustion chamber to the main steam temperature measuring point leads to a delayed response in the control system, making it difficult to respond promptly to load changes and disturbances, resulting in severe control lag. Second, complex multivariate coupling. The main steam temperature is affected by multiple factors such as fuel quantity, air volume, feedwater temperature, and load changes, and there are strong coupling relationships between these variables. This complex multivariate coupling makes it difficult for traditional control methods to handle. Third, poor adaptability to operating conditions. The dynamic characteristics of the boiler vary significantly under different loads, and a controller with fixed parameters struggles to maintain good control performance across the entire operating range, resulting in insufficient adaptability to operating conditions. Finally, existing control systems mainly rely on feedback adjustment based on the current state, lacking the ability to predict future temperature trends and thus failing to achieve forward-looking control. Summary of the Invention

[0004] The purpose of this invention is to provide a deep learning-based intelligent prediction and control method and system for boiler main steam temperature, in order to solve the problem that existing technologies cannot accurately predict the trend of main steam temperature changes.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: Firstly, a deep learning-based intelligent predictive control method for boiler main steam temperature includes the following steps: Collect multidimensional time series data of boiler operating conditions, preprocess and extract features from the multidimensional time series data to obtain time series feature variables; The time-series feature variables are input into a pre-trained temperature prediction model, which outputs the predicted main steam temperature values ​​for multiple future time steps. By minimizing the deviation between the predicted main steam temperature and the preset main steam temperature at multiple future time steps, rolling optimization is performed in the prediction time domain to obtain the optimal desuperheating water regulating valve control command, thereby realizing intelligent predictive control of the boiler main steam temperature.

[0006] In some implementations, the multidimensional time series data undergoes preprocessing and feature extraction, specifically including: After cleaning and normalizing the multidimensional time series data, feature extraction is performed on the preprocessed multidimensional time series data within a preset sliding time window based on correlation analysis and mutual information methods to obtain time series feature variables that are strongly correlated with the main steam temperature.

[0007] In some implementations, the pre-trained temperature prediction model is trained through the following steps: A training sample set containing the time-series feature variables and their corresponding actual main steam temperatures is constructed. The time-series feature variables are used as inputs, and the actual main steam temperatures are used as the desired outputs. The goal is to minimize the error between the predicted and actual main steam temperatures. The temperature prediction model is iteratively trained using the training sample set until the model converges, thus obtaining a pre-trained temperature prediction model.

[0008] In some implementations, the pre-trained temperature prediction model includes: a temporal convolutional network layer, an SE attention layer, and a long short-term memory network layer connected in sequence; The temporal convolutional network layer is used to perform multi-layer dilated causal convolution on the temporal feature variables to obtain a high-dimensional feature map; The SE attention layer is used to perform global average pooling on the high-dimensional feature map and generate channel attention weights through a fully connected layer. The feature channels of the high-dimensional feature map are weighted according to the channel attention weights to obtain a weighted feature sequence. The long short-term memory network layer is used to process the weighted feature sequence through a gated recurrent structure to obtain the predicted main steam temperature values ​​for multiple future time steps.

[0009] In some implementations, rolling optimization is performed in the prediction time domain by minimizing the deviation between the predicted main steam temperature and the preset main steam temperature for the next multiple time steps. Specifically, this includes: With the objective function being to minimize the deviation between the predicted and preset main steam temperatures at multiple future time steps, and with the physical opening limit, rate of change limit, and boiler safety operation parameters of the desuperheating water regulating valve as constraints, a model predictive control framework is used to perform rolling optimization in the prediction time domain.

[0010] In some implementations, the following steps are also included: The actual measured value of the main steam temperature of the boiler is obtained. When the deviation between the actual measured value of the main steam temperature and the predicted value of the main steam temperature at the corresponding time step exceeds a preset threshold, the pre-trained temperature prediction model is triggered to be updated online.

[0011] Secondly, a deep learning-based intelligent predictive control system for boiler main steam temperature includes: The data acquisition and preprocessing module is used to acquire multidimensional time series data of boiler operating conditions, preprocess the multidimensional time series data and extract features to obtain time series feature variables. The deep learning prediction module is used to input the time-series feature variables into the pre-trained temperature prediction model and output the predicted values ​​of the main steam temperature for multiple future time steps. The intelligent control decision module is used to minimize the deviation between the predicted value of the main steam temperature and the preset value of the main steam temperature in multiple future time steps, and to perform rolling optimization in the prediction time domain to obtain the optimal desuperheating water regulating valve control command, thereby realizing intelligent prediction control of the boiler main steam temperature.

[0012] In some implementations, it also includes: The human-computer interaction and monitoring module is used to obtain the actual measured value of the main steam temperature of the boiler. When the deviation between the actual measured value of the main steam temperature and the predicted value of the main steam temperature at the corresponding time step exceeds a preset threshold, the pre-trained temperature prediction model is triggered to be updated online.

[0013] Thirdly, a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the deep learning-based intelligent prediction and control method for boiler main steam temperature.

[0014] Fourthly, a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, it implements the steps of the deep learning-based intelligent prediction and control method for boiler main steam temperature.

[0015] Compared with the prior art, the present invention has the following beneficial effects: This invention provides a deep learning-based intelligent predictive control method for boiler main steam temperature. By using a pre-trained temperature prediction model and outputting predicted main steam temperature values ​​for multiple future time steps, it solves the core problem of existing technologies being unable to accurately predict the trend of main steam temperature changes. Furthermore, by using multi-step prediction results based on historical data as the pre-input of control decisions, the control system can calculate control commands in advance based on future temperature change trajectories. This effectively overcomes the control action lag caused by the large thermal inertia and long heat transfer path of the boiler, and realizes a paradigm shift from passive feedback to active feedforward intelligent control.

[0016] Furthermore, correlation analysis and mutual information methods are used to extract features from the preprocessed multidimensional time series data. By integrating two complementary metrics, linear correlation and nonlinear statistical dependence, time-series feature variables that are strongly correlated with the main steam temperature can be comprehensively and accurately screened from complex boiler operating parameters. This provides high-quality input with higher information density and less redundancy for subsequent prediction models, thereby improving the reliability of feature representation from the source.

[0017] Furthermore, the temperature prediction model takes time-series characteristic variables as input and the actual value of the main steam temperature as the desired output for training. It can directly learn the complex dynamic mapping relationship between multi-dimensional operating conditions and target temperature from the historical operating data of the boiler. This allows the pre-trained model to internalize the operating characteristics of a specific boiler object, providing a foundation for achieving high-precision generalized prediction in practical applications.

[0018] Furthermore, the specific architecture of the temperature prediction model consists of a temporal convolutional network layer, an SE attention layer, and a long short-term memory network layer connected in sequence. The temporal convolutional network layer extracts long-term dependency features through multi-layer dilated causal convolution; the SE attention layer achieves adaptive feature weighting by generating channel attention weights; and the long short-term memory network layer finally completes the sequence prediction. This hybrid architecture, through the organic synergy of the functions of each component, is specifically optimized for long-term time series modeling, key variable focusing, and multi-step sequence generation involved in boiler main steam temperature prediction, ensuring high accuracy and stability of the prediction task from the model structure level.

[0019] Furthermore, a model predictive control framework is adopted, with the objective function being to minimize the deviation between the predicted and preset main steam temperatures at multiple future time steps. The physical opening limits, rate of change limits, and boiler safety operating parameters of the desuperheating water regulating valve are used as constraints. Rolling optimization is performed in the prediction time domain, combining the multi-step prediction sequence with the constrained rolling optimization algorithm. This ensures that the generated optimal desuperheating water regulating valve control command not only compensates for prediction deviations but also strictly meets the physical limits of the actuator and the safe operating boundaries of the unit, thereby achieving a balance between optimal control performance and engineering feasibility.

[0020] Furthermore, when the deviation between the actual measured value of the main steam temperature and the predicted value of the main steam temperature at the corresponding time step exceeds a preset threshold, the pre-trained temperature prediction model is triggered to update online. This can automatically monitor the performance degradation of the prediction model and ensure that the intelligent prediction control system maintains reliable prediction accuracy and control effect throughout its entire life cycle. Attached Figure Description

[0021] Figure 1 A detailed flowchart of a deep learning-based intelligent prediction and control method for boiler main steam temperature provided in an embodiment of the present invention; Figure 2 A structural diagram of a deep learning-based intelligent predictive control system for boiler main steam temperature is provided for an embodiment of the present invention. Figure 3 The above is a flowchart of an intelligent prediction and control method for boiler main steam temperature based on deep learning, provided for an embodiment of the present invention. Detailed Implementation

[0022] To enable those skilled in the art to better understand the present invention, the technical solution of the present invention will be further described in detail below with reference to the accompanying drawings. The content described herein is for explanation rather than limitation of the present invention.

[0023] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification and claims of this invention are intended to cover a non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such processes, methods, systems, products, or devices.

[0024] like Figure 1 and Figure 3 As shown, this embodiment provides a deep learning-based intelligent predictive control method for boiler main steam temperature, including the following steps: S1, Collect multidimensional time series data of boiler operating conditions, preprocess and extract features from the multidimensional time series data to obtain time series feature variables; Data acquisition is performed in real time through the boiler distributed control system (DCS), collecting multi-dimensional operating condition data related to the main steam temperature, i.e., multi-dimensional time series data, with an acquisition frequency set to 1Hz. The collected multi-dimensional time series data specifically includes: temperature parameters, such as superheat, separator outlet temperature, inlet / outlet steam temperature of the A-side secondary superheater, inlet header temperature of the final superheater, and main steam temperature; control parameters, such as the setpoint of the secondary superheater A and the manual command values ​​of the primary / secondary desuperheating water regulating valves of the A-side superheater; operating parameters, such as AGC load, main steam pressure, total coal quantity, primary air pressure and volume of each coal mill; and combustion parameters, such as hot secondary air volume, burnout air regulating damper opening, oxygen content in the flue gas at the air preheater outlet, and flame intensity of each burner. A data quality monitoring mechanism is established to identify and process abnormal data. Preprocessing and feature extraction include data cleaning, missing value imputation, and outlier detection, and a sliding time window is constructed to extract time series features.

[0025] After cleaning and normalizing the multidimensional time series data, features are extracted from the preprocessed data within a preset sliding time window based on correlation analysis and mutual information methods to obtain time-series feature variables strongly correlated with the main steam temperature. Specifically, data cleaning is used to remove obvious outliers and noise, and to handle data loss caused by sensor malfunctions. Normalization uses MinMaxScaler to scale all features to the [0,1] interval. The time window is constructed by establishing a sliding window with a length of 96 time points as model input. Feature selection uses Pearson correlation coefficient and mutual information methods to screen feature variables strongly correlated with the main steam temperature. Systematic data cleaning and normalization ensure data quality, time window construction captures temporal dependencies, and the combined use of linear Pearson correlation coefficient and nonlinear mutual information methods selects the most effective key feature variables for main steam temperature prediction from multi-source data. This achieves automatic optimization of input feature selection, reduces interference from noise and redundant information, provides an accurate and efficient data foundation for subsequent temperature prediction models, and directly improves the model's prediction performance.

[0026] S2, input the time-series feature variables into the pre-trained temperature prediction model, and output the predicted values ​​of the main steam temperature for multiple future time steps; The pre-trained temperature prediction model is a deep learning-based time-series prediction model. Rolling optimization is performed using a model predictive control (MPC) framework. Real-time data acquisition and processing are used to obtain model inputs, and the pre-trained temperature prediction model is used to predict temperature values ​​at multiple future time steps. Prediction confidence intervals are calculated, and prediction uncertainty is assessed.

[0027] The training method for the pre-trained temperature prediction model is as follows: A training sample set containing time-series feature variables and their corresponding actual main steam temperatures is constructed. Using the time-series feature variables as input and the actual main steam temperature as the desired output, the model aims to minimize the error between the predicted and actual main steam temperatures. The model is iteratively trained using this training sample set until convergence, resulting in the pre-trained temperature prediction model. Specifically, historical boiler operating data is collected to construct the training sample set. A sliding window technique is used to generate input-output sample pairs, and the mean squared error loss function and Adam optimizer are used to train the temperature prediction model. Cross-validation is employed to evaluate the model's generalization performance. The trained model is exported in ONNX format for real-time inference. Utilizing historical operating data, supervised learning enables the temperature prediction model to learn the complex mapping relationship between historical operating condition sequences and future main steam temperature sequences. Through data-driven training, the model deeply internalizes the specific boiler's operating characteristics and dynamic patterns, thus enabling it to output high-precision, multi-step main steam temperature predictions in practical applications.

[0028] The pre-trained temperature prediction model consists of: a temporal convolutional network layer, an SE attention layer, and a long short-term memory network layer connected in sequence; Temporal convolutional network (TCN) layers are used to perform multi-layer dilated causal convolutions on temporal feature variables to obtain high-dimensional feature maps; The squeeze-excitement (SE) attention layer is used to perform global average pooling on the high-dimensional feature map and generate channel attention weights through a fully connected layer. The feature channels of the high-dimensional feature map are weighted according to the channel attention weights to obtain the weighted feature sequence. The Long Short-Term Memory (LSTM) network layer is used to process the weighted feature sequence through a gated recurrent structure to obtain the main steam temperature prediction values ​​for multiple future time steps, such as short-term (3~6 time steps), medium-term (12~18 time steps), and long-term (24 steps) main steam temperature prediction values.

[0029] Specifically, the model employs a TCN-SE-LSTM hybrid architecture. The temporal convolutional network layer consists of stacked residual blocks, each containing two layers of dilated causal convolutions with dilation factors increasing by 2^i to capture long-term temporal dependencies. The SE attention layer weights the feature channels output by the temporal convolutional network layer based on their importance, compressing the spatial dimension through global average pooling, then learning the channel weights through two fully connected layers, and finally multiplying them with the original features to achieve weighting. The Long Short-Term Memory (LSTM) network layer receives the weighted feature sequence and uses its recurrent structure to perform multi-step predictions from sequence to sequence, outputting the predicted main steam temperature for the next 24 steps. This hybrid architecture is an effective combination of TCN, SE, and LSTM. The temporal convolutional network layer extracts deep temporal features containing long-term dependencies; the SE attention layer dynamically evaluates and strengthens the weights of key feature channels; and the LSTM network layer ultimately models the refined feature sequence to generate coherent multi-step predictions for the future. This hybrid model architecture combines the advantages of three networks and is particularly well-suited for handling the challenges of nonlinearity, multivariate coupling, and long-range dependence in boiler main steam temperature prediction, thereby significantly improving the accuracy of multi-step temperature prediction.

[0030] S3, by minimizing the deviation between the predicted value of the main steam temperature and the preset value of the main steam temperature in the future multiple time steps, rolling optimization is performed in the prediction time domain to obtain the optimal desuperheating water regulating valve control command, thereby realizing intelligent prediction control of the boiler main steam temperature.

[0031] A model predictive control framework is used for rolling optimization to generate and issue optimal control commands. Through multi-step temperature prediction and model predictive control based on deep learning, forward-looking intelligent control of the boiler main steam temperature is realized, which effectively overcomes the control lag problem caused by large system delay and large inertia, and improves control response speed and control accuracy.

[0032] Specifically, by minimizing the deviation between the predicted and preset main steam temperature values ​​for multiple future time steps, rolling optimization is performed in the prediction time domain. The objective function is to minimize the deviation between the predicted and preset main steam temperature values ​​for multiple future time steps. The physical opening limit, rate of change limit, and boiler safety operation parameters of the desuperheating water regulating valve are used as constraints. The model predictive control framework is used to perform rolling optimization in the prediction time domain.

[0033] Intelligent control decisions are based on temperature prediction results and employ a model predictive control framework. The objective function is designed to minimize the deviation between the predicted temperature and the setpoint, while considering the smoothness of control variable changes. Constraints include limits on the opening degree of the desuperheating water regulating valve, limits on the rate of change, and safety operation constraints. The optimal control sequence is solved within the prediction time domain, and a rolling optimization strategy is adopted, updating the prediction and optimization results in each control cycle to achieve rolling control. In each control cycle, the model predictive control framework uses a pre-trained temperature prediction model as the internal prediction model. Under the premise of satisfying various hard constraints, it solves an optimization problem in a finite time domain to calculate the optimal control command sequence that makes the future predicted output closest to the setpoint, and implements the first command in the sequence. This close integration of model-based multi-step prediction with constrained online optimization not only achieves forward-looking control to overcome system inertia but also ensures the smoothness, feasibility, and safety of control commands, thereby improving control accuracy while guaranteeing the safe and stable operation of the unit.

[0034] S4, obtain the actual measured value of the main steam temperature of the boiler. When the deviation between the actual measured value of the main steam temperature and the predicted value of the main steam temperature at the corresponding time step exceeds a preset threshold, trigger the online update of the pre-trained temperature prediction model.

[0035] Specifically, during closed-loop control execution, the control effect and temperature response are monitored in real time. Prediction error is calculated, and an online model correction mechanism is triggered when the error exceeds a preset threshold. Online updates can include retraining or fine-tuning model parameters using the latest operational data, and recording control process data for continuous model optimization. The model's prediction accuracy is dynamically monitored by continuously comparing the actual measured value of the main steam temperature with the model's predicted value at the same time. Once a performance degradation exceeding the allowable range is detected, the model update process is automatically initiated, enabling the model to adapt to changes in object characteristics. This endows the intelligent predictive control system with self-learning and adaptive capabilities, effectively overcoming model mismatch problems caused by factors such as equipment aging and fuel changes, thereby maintaining high-precision prediction performance and excellent control effect over a long period, improving the system's adaptability and robustness. The optimized control commands are sent to the DCS system: opening commands are sent to the desuperheating water regulating valve via the communication interface; execution effect and temperature response are monitored in real time; prediction error is calculated, and model correction is triggered when the error exceeds a threshold; control process data is recorded for continuous model optimization.

[0036] like Figure 2 As shown, this embodiment provides a deep learning-based intelligent prediction and control system for boiler main steam temperature, including: The data acquisition and preprocessing module is used to collect multidimensional time-series data on boiler operating conditions, preprocess the data, and extract features to obtain time-series characteristic variables. It is configured with a multi-channel data acquisition interface for real-time communication with the DCS system. The module incorporates a data quality assessment algorithm to automatically identify and process abnormal data. It also provides data storage and historical query functions. The module includes a communication interface supporting standard industrial communication protocols such as OPC and Modbus; data quality monitoring using statistical analysis methods to identify abnormal data points; data processing providing interpolation and filtering algorithms; and storage management supporting real-time data caching and historical data storage.

[0037] The deep learning prediction module is used to input the time-series feature variables into a pre-trained temperature prediction model and output the predicted main steam temperature values ​​for multiple future time steps. The deep learning prediction module includes a model loader: supporting dynamic loading and switching of ONNX format models; an inference engine: achieving efficient model inference based on the ONNX Runtime; prediction post-processing: including functions such as inverse normalization and confidence interval calculation; and performance monitoring: real-time evaluation of model prediction accuracy and computational performance. The deep learning prediction module loads a pre-trained TCN-SE-LSTM model, supports efficient inference in the ONNX format, implements multi-step temperature prediction, outputs prediction confidence intervals, and provides model performance monitoring and online update functions.

[0038] The intelligent control decision module is used to minimize the deviation between the predicted and preset main steam temperature values ​​at multiple future time steps. It performs rolling optimization within the prediction time domain to obtain the optimal control command for the desuperheating water regulating valve, thus achieving intelligent predictive control of the boiler's main steam temperature. The intelligent control decision module includes an MPC controller (implementing model predictive control algorithms), an optimization solver (supporting optimization algorithms such as quadratic programming), constraint management (handling equipment physical and safety constraints), and control mode management (supporting multiple modes such as automatic, manual, and maintenance). The intelligent control decision module integrates model predictive control algorithms, optimizes control strategies based on temperature prediction results, considers multi-objective optimization, balances control accuracy and energy consumption, and supports switching between multiple control modes.

[0039] The human-machine interaction and monitoring module is used to acquire the actual measured value of the boiler's current main steam temperature. When the deviation between the actual measured value and the predicted value of the main steam temperature at the corresponding time step exceeds a preset threshold, the pre-trained temperature prediction model is triggered to update online. The human-machine interaction and monitoring module includes: a real-time monitoring interface displaying current operating parameters and control status; trend analysis providing historical data query and trend analysis functions; parameter configuration supporting online adjustment of control parameters; and alarm management providing multi-level alarm and fault diagnosis functions. The human-machine interaction and monitoring module provides a real-time data display and trend analysis interface; supports online adjustment of control parameters and mode switching; and has alarm and fault diagnosis functions.

[0040] Multi-threaded execution framework: Data acquisition, model prediction, and control optimization are executed in parallel in independent threads; Log recording system: records system operating status and key operations; Fault diagnosis module: Provides system health monitoring and fault location functions; Online learning module: Supports online model updates and parameter optimization.

[0041] This intelligent predictive control system adopts a layered architecture: Data Layer: Responsible for data interaction with the DCS system, including real-time data acquisition, historical data storage, and data quality monitoring. Model Layer: The core deep learning prediction module loads a pre-trained TCN-SE-LSTM model, providing high-precision multi-step temperature prediction. Control Layer: Makes control decisions based on the prediction results, integrates model predictive control algorithms, and outputs the optimal control strategy. Application Layer: Provides a human-machine interface, supporting system monitoring, parameter adjustment, and mode switching. The system employs a multi-threaded parallel architecture, with data acquisition, model prediction, and control optimization executed in independent threads to ensure real-time performance. It also provides comprehensive logging and fault diagnosis functions to ensure reliable system operation.

[0042] This embodiment was validated through engineering application on a 600MW supercritical coal-fired unit: the main steam temperature control accuracy was improved from ±5℃ of the traditional PID control to ±2℃, a 60% increase in control accuracy. The response time to load changes was shortened from 15 minutes to 8 minutes, a 47% improvement in response speed. It maintained good control performance within a load range of 30%-100%, demonstrating strong adaptability. Through precise temperature control, the unit's thermal efficiency increased by 0.2%, saving approximately 1000 tons of standard coal annually.

[0043] This invention employs a TCN-SE-LSTM hybrid deep learning architecture, which accurately captures the complex variation patterns of main steam temperature, achieving significantly higher prediction accuracy than traditional methods. Based on multi-step prediction results, it performs forward-looking control, effectively overcoming the system's large latency characteristics and significantly improving control response speed. The deep learning model can automatically learn system characteristics under different operating conditions, requiring no manual parameter tuning and exhibiting strong adaptability. Combining predictive control and intelligent decision-making, it can control the main steam temperature within ±2℃, achieving control accuracy superior to traditional PID control. The system adopts a modular design, facilitating integration with existing DCS systems and demonstrating significant engineering application value. In summary, this invention achieves intelligent predictive control of boiler main steam temperature through deep learning technology, significantly improving control accuracy and system performance, and possessing important engineering application value.

[0044] The module division in this embodiment of the invention is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in the various embodiments of the invention can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0045] This embodiment also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space, which stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the deep learning-based intelligent predictive control method for boiler main steam temperature in the above embodiment.

[0046] This embodiment also provides a computer program product, which includes a computer program that, when executed by a processor, implements the corresponding steps of the deep learning-based intelligent prediction and control method for boiler main steam temperature described in the above embodiment.

[0047] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0048] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0049] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0050] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0051] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A deep learning-based intelligent predictive control method for boiler main steam temperature, characterized in that, Includes the following steps: Collect multidimensional time series data of boiler operating conditions, preprocess and extract features from the multidimensional time series data to obtain time series feature variables; The time-series feature variables are input into a pre-trained temperature prediction model, which outputs the predicted main steam temperature values ​​for multiple future time steps. By minimizing the deviation between the predicted main steam temperature and the preset main steam temperature at multiple future time steps, rolling optimization is performed in the prediction time domain to obtain the optimal desuperheating water regulating valve control command, thereby realizing intelligent predictive control of the boiler main steam temperature.

2. The intelligent predictive control method for boiler main steam temperature based on deep learning according to claim 1, characterized in that, The preprocessing and feature extraction of the multidimensional time series data specifically includes: After cleaning and normalizing the multidimensional time series data, feature extraction is performed on the preprocessed multidimensional time series data within a preset sliding time window based on correlation analysis and mutual information methods to obtain time series feature variables that are strongly correlated with the main steam temperature.

3. The intelligent predictive control method for boiler main steam temperature based on deep learning according to claim 1, characterized in that, The pre-trained temperature prediction model is trained through the following steps: A training sample set containing the time-series feature variables and their corresponding actual main steam temperatures is constructed. The time-series feature variables are used as inputs, and the actual main steam temperatures are used as the desired outputs. The goal is to minimize the error between the predicted and actual main steam temperatures. The temperature prediction model is iteratively trained using the training sample set until the model converges, thus obtaining a pre-trained temperature prediction model.

4. A deep learning-based intelligent prediction and control method for boiler main steam temperature according to claim 1 or 3, characterized in that, The pre-trained temperature prediction model includes: a temporal convolutional network layer, an SE attention layer, and a long short-term memory network layer connected in sequence; The temporal convolutional network layer is used to perform multi-layer dilated causal convolution on the temporal feature variables to obtain a high-dimensional feature map; The SE attention layer is used to perform global average pooling on the high-dimensional feature map and generate channel attention weights through a fully connected layer. The feature channels of the high-dimensional feature map are weighted according to the channel attention weights to obtain a weighted feature sequence. The long short-term memory network layer is used to process the weighted feature sequence through a gated recurrent structure to obtain the predicted main steam temperature values ​​for multiple future time steps.

5. The intelligent predictive control method for boiler main steam temperature based on deep learning according to claim 1, characterized in that, By minimizing the deviation between the predicted main steam temperature and the preset main steam temperature at multiple future time steps, a rolling optimization solution is performed within the prediction time domain, specifically including: With the objective function being to minimize the deviation between the predicted and preset main steam temperatures at multiple future time steps, and with the physical opening limit, rate of change limit, and boiler safety operation parameters of the desuperheating water regulating valve as constraints, a model predictive control framework is used to perform rolling optimization in the prediction time domain.

6. The intelligent predictive control method for boiler main steam temperature based on deep learning according to claim 1, characterized in that, It also includes the following steps: The actual measured value of the main steam temperature of the boiler is obtained. When the deviation between the actual measured value of the main steam temperature and the predicted value of the main steam temperature at the corresponding time step exceeds a preset threshold, the pre-trained temperature prediction model is triggered to be updated online.

7. A deep learning-based intelligent predictive control system for boiler main steam temperature, characterized in that, include: The data acquisition and preprocessing module is used to acquire multidimensional time series data of boiler operating conditions, preprocess the multidimensional time series data and extract features to obtain time series feature variables. The deep learning prediction module is used to input the time-series feature variables into the pre-trained temperature prediction model and output the predicted values ​​of the main steam temperature for multiple future time steps. The intelligent control decision module is used to minimize the deviation between the predicted value of the main steam temperature and the preset value of the main steam temperature in multiple future time steps, and to perform rolling optimization in the prediction time domain to obtain the optimal desuperheating water regulating valve control command, thereby realizing intelligent prediction control of the boiler main steam temperature.

8. The intelligent predictive control system for boiler main steam temperature based on deep learning according to claim 7, characterized in that, Also includes: The human-computer interaction and monitoring module is used to obtain the actual measured value of the main steam temperature of the boiler. When the deviation between the actual measured value of the main steam temperature and the predicted value of the main steam temperature at the corresponding time step exceeds a preset threshold, the pre-trained temperature prediction model is triggered to be updated online.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the intelligent predictive control method for boiler main steam temperature based on deep learning as described in any one of claims 1 to 6.

10. A computer program product, the computer program product comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the intelligent predictive control method for boiler main steam temperature based on deep learning as described in any one of claims 1 to 6.

Images