Coal-fired power plant unit total coal feed quantity and air quantity prediction method based on multi-task learning and related device

Abstract

The invention provides a coal-fired power plant unit total coal feed quantity and air quantity prediction method based on multi-task learning and a related device. The method comprises the steps that unit loads and operation state parameters of all coal mills are obtained; inputting the acquired data into a pre-trained multi-task learning framework model to obtain a prediction result of the total coal feed quantity and the air quantity of the unit; the multi-task learning framework model is trained by adopting a multi-task learning technology, the first task is used for predicting the total coal feed of the unit based on shared features of model input data, and the second task is used for predicting fan current data based on the shared features and the total coal feed of the unit; and the third task is used for predicting the fan volume based on the shared characteristics and the fan current data. According to the method, the sharing features are integrated through a multi-task learning framework, accurate prediction of the total coal feeding amount and the air amount of the coal-fired power plant unit is achieved, the efficiency and stability of a combustion control system are improved, and therefore the effects of energy conservation and emission reduction are achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of combustion control of coal-fired power plants, and specifically relates to a method and related devices for predicting the total coal feed rate and air volume of coal-fired power plant units based on multi-task learning. Background Art

[0002] With the development of industrial automation and intelligence, the combustion control system of coal-fired power plants plays a vital role in ensuring the safe, economical and environmentally friendly operation of power plants. Traditional combustion control systems rely on fixed control strategies and are difficult to cope with complex and changing operating conditions, resulting in low combustion efficiency, high pollutant emissions and unstable equipment operation. In the face of very complex boiler combustion processes, the use of intermediate variables can provide more accurate predictions.

[0003] At present, the main problems existing in the combustion control system of coal-fired power plants are: on the one hand, single-task machine learning models (such as LSTM, SVM) can only predict a single combustion parameter, which not only causes the information island effect, but also limits the accuracy of the prediction and the generalization ability of the model; on the other hand, although the physical model-based method can provide a detailed description of the combustion process, due to its highly complex modeling requirements and sensitivity to specific operating conditions, it is difficult for such models to respond quickly to changes in the actual operating environment, and the implementation cost is high.

[0004] In recent years, multi-task learning (MTL) has attracted widespread attention as a machine learning method that can simultaneously optimize multiple related tasks. MTL can effectively improve the generalization ability and prediction accuracy of the model by sharing information between tasks, and is particularly suitable for dealing with complex industrial control problems. By introducing the MTL framework to develop intelligent combustion prediction models, it is expected to integrate multiple subtasks (such as boiler load prediction, air volume control, etc.) to achieve more sophisticated and efficient combustion control. Summary of the invention

[0005] In view of this, the present invention provides a method and related device for predicting the total coal feed and air volume of a coal-fired power plant unit based on multi-task learning, aiming to reduce the complexity of modeling and calculation while improving prediction accuracy and adaptability by combining multi-task learning and LSTM model, which is more suitable for complex scenarios in industrial control.

[0006] In order to achieve the above object, the technical solution provided by the present invention is as follows:

[0007] In a first aspect, the present invention provides a method for predicting the total coal feed rate and air volume of a coal-fired power plant unit based on multi-task learning, comprising the following steps:

[0008] Obtain unit load and operation status parameters of each coal mill;

[0009] The unit load and the operating status parameters of each coal mill are used as model input data and input into the pre-trained multi-task learning framework model to obtain the prediction results of the unit's total coal feed and air volume;

[0010] Among them, the multi-task learning framework model is trained using multi-task learning technology. The multi-task learning framework model is used to perform at least the first task, the second task and the third task. The first task is used to predict the total coal feed of the unit based on the shared characteristics of the model input data, the second task is used to predict the fan current data based on the shared characteristics and the total coal feed of the unit, and the third task is used to predict the fan air volume based on the shared characteristics and the fan current data.

[0011] Furthermore, the architecture of the multi-task learning framework model includes:

[0012] The feature extraction layer is used to extract common features from the model input data as shared features;

[0013] The task-specific layer includes multiple LSTM layers, each of which corresponds to a task. Each LSTM layer is used to extract task-specific features related to the corresponding task based on the shared features and the prediction results of the associated tasks.

[0014] The output layer is used to output the prediction results of each task based on the shared features and the task-specific features of each task.

[0015] Furthermore, the feature extraction layer includes:

[0016] The input layer is used to receive various time series data including unit load and operation status parameters of each coal mill;

[0017] Channel attention mechanism layer, used to increase the feature weights related to key features in various types of time series data;

[0018] The hybrid expert model layer is used to select the most appropriate expert network to process the corresponding data according to the characteristics of various time series data;

[0019] The shared LSTM layer is used to extract common time series features from the processed time series data as shared features.

[0020] Furthermore, the process of processing data in the channel attention mechanism layer includes:

[0021] Perform global average pooling and global maximum pooling on the input feature map to obtain the average description and maximum description of the global features;

[0022] The average description and the maximum description are input into the shared fully connected layer respectively, and the weight of each channel is calculated through the activation function to finally obtain the channel attention weight;

[0023] The channel attention weights are applied to each channel of the input feature map, and the feature map is reweighted to obtain a channel enhanced feature map.

[0024] Furthermore, the multi-task learning framework model uses a task uncertainty weighted loss method to adjust the loss weight of each task. The steps of the task uncertainty weighted loss method include:

[0025] Define a loss function for each task;

[0026] Assign a learnable weight parameter to each task;

[0027] The loss functions and weight parameters of each task are combined to form a total loss function.

[0028] Furthermore, the expression of the total loss function is as follows:

[0029]

[0030] In the formula, is the total loss function, is the loss function of the i-th task, is the weight parameter of the i-th task.

[0031] In a second aspect, the present invention provides a device for predicting the total coal supply and air volume of a coal-fired power plant unit based on multi-task learning, comprising:

[0032] Data acquisition module, used to obtain unit load and operation status parameters of each coal mill;

[0033] The prediction module is used to input the unit load and the operating status parameters of each coal mill as model input data into the pre-trained multi-task learning framework model to obtain the prediction results of the total coal feed and air volume of the unit;

[0034] Among them, the multi-task learning framework model is trained using multi-task learning technology. The multi-task learning framework model is used to perform at least the first task, the second task and the third task. The first task is used to predict the total coal feed of the unit based on the shared characteristics of the model input data, the second task is used to predict the fan current data based on the shared characteristics and the total coal feed of the unit, and the third task is used to predict the fan air volume based on the shared characteristics and the fan current data.

[0035] In a third aspect, the present invention provides a computer device, the device comprising a processor and a memory:

[0036] The memory is used to store the computer program and send the instructions of the computer program to the processor;

[0037] The processor executes a method for predicting the total coal feed rate and air volume of a coal-fired power plant unit based on multi-task learning according to the instructions of the computer program as described in the first aspect.

[0038] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, a method for predicting the total coal feed and air volume of a coal-fired power plant unit based on multi-task learning as in the first aspect is implemented.

[0039] In a fifth aspect, the present invention provides a computer program product, comprising a computer program, which, when executed by a processor, implements a method for predicting the total coal feed and air volume of a coal-fired power plant unit based on multi-task learning according to the first aspect.

[0040] In summary, the present invention provides a method and related device for predicting the total coal supply and air volume of a coal-fired power plant unit based on multi-task learning, including obtaining the unit load and the operation status parameters of each coal mill; using the unit load and the operation status parameters of each coal mill as model input data, inputting them into a pre-trained multi-task learning framework model, and obtaining the prediction results of the total coal supply and air volume of the unit; wherein the multi-task learning framework model is trained using multi-task learning technology, and the multi-task learning framework model is at least used to perform the first task, the second task and the third task, the first task is used to predict the total coal supply of the unit based on the shared features of the model input data, the second task is used to predict the fan current data based on the shared features and the total coal supply of the unit, and the third task is used to predict the fan air volume based on the shared features and the fan current data. The present invention integrates shared features through a multi-task learning framework, realizes the accurate prediction of the total coal supply and air volume of coal-fired power plant units, improves the efficiency and stability of the combustion control system, and thus achieves the effect of energy saving and emission reduction. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0042] Figure 1 A flowchart of a method for predicting the total coal supply and air volume of a coal-fired power plant unit based on multi-task learning provided by an embodiment of the present invention;

[0043] Figure 2 A block diagram of an implementation method for predicting the total coal supply and air volume of a coal-fired power plant unit based on multi-task learning provided in an embodiment of the present invention;

[0044] Figure 3A structural diagram of the LSTM multi-task learning model provided in an embodiment of the present invention;

[0045] Figure 4 A block diagram of a device for predicting the total coal supply and air volume of a coal-fired power plant unit based on multi-task learning provided by an embodiment of the present invention;

[0046] Figure 5 A block diagram of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0047] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] The prior art is first further introduced below.

[0049] (1) Single-task machine learning model

[0050] Use a single-task machine learning model (such as long short-term memory network LSTM, support vector machine SVM, etc.) to predict a combustion parameter. For example, use the LSTM model to predict the total coal feed or the air volume of a single fan. The model input includes historical operating data (such as boiler load, coal mill status, etc.), and the output is the target prediction parameter. Through data preprocessing, feature engineering, model training and verification, the prediction model is established and optimized.

[0051] Disadvantages: Information isolation. Single-task models can only learn and predict a single task and cannot utilize relevant information between different tasks. Limited prediction accuracy. Due to the failure to share information between tasks, the generalization ability and prediction accuracy of the model are low. Waste of resources. The training and maintenance of multiple single-task models require more computing resources and manpower.

[0052] (2) Combustion prediction based on physical models

[0053] Construct a detailed physical model of the boiler combustion process, and perform combustion prediction and optimization through accurate description of the combustion mechanism (including chemical reaction, heat transfer, fluid mechanics, etc.). Commonly used methods include numerical simulation, computational fluid dynamics (CFD), etc., using computer simulation technology to simulate each link of the combustion process in the boiler. Based on the simulation results, predict key parameters such as the combustion efficiency, pollutant emissions and equipment operating status of the coal-fired power plant, thereby optimizing the combustion process.

[0054] Disadvantages: Modeling is complex. The combustion process involves a large number of complex physical and chemical reactions, and it is very difficult to establish an accurate physical model. The physical model has poor adaptability to changes in operating conditions and requires frequent adjustment of model parameters. The accurate physical model has a large amount of calculation and poor real-time performance, making it difficult to apply in actual operation.

[0055] Based on this, in one embodiment of the present invention, Figure 1 As shown, a method for predicting the total coal supply and air volume of a coal-fired power plant unit based on multi-task learning is provided, comprising the following steps:

[0056] S1: Obtain the unit load and operation status parameters of each coal mill.

[0057] It should be noted that this step obtains key parameters that reflect the current operating status of the power plant, including but not limited to unit load (reflecting the power generation demand of the power plant) and the operating status parameters of each coal mill (such as whether the coal mill is started, operating efficiency, etc.). These data are crucial for predicting the total coal feed and air volume.

[0058] Among them, unit load is an important indicator reflecting the operating status of the power plant, which directly affects the combustion process and load changes of the boiler. Therefore, unit load is a key variable for predicting the total coal supply and air volume, which can help the model reflect the actual operating conditions more accurately.

[0059] Coal mill operation status parameters (wA, wB, wC, wD, wE, wF): The operation status of the coal mill directly affects the change of coal feed rate and is an important factor affecting the combustion process. By monitoring the operation status of the coal mill, the impact of different coal mill combinations on the total coal feed rate and air volume can be captured, and the prediction accuracy of the model can be improved.

[0060] S2: The unit load and the operating status parameters of each coal mill are used as model input data and input into the pre-trained multi-task learning framework model to obtain the prediction results of the total coal feed and air volume of the unit;

[0061] Among them, the multi-task learning framework model is trained using multi-task learning technology. The multi-task learning framework model is used to perform at least the first task, the second task and the third task. The first task is used to predict the total coal feed of the unit based on the shared characteristics of the model input data, the second task is used to predict the fan current data based on the shared characteristics and the total coal feed of the unit, and the third task is used to predict the fan air volume based on the shared characteristics and the fan current data.

[0062] It should be noted that in this step, the unit load and coal mill operation status parameters obtained in step S1 are input into a pre-trained multi-task learning framework model. This model is a deep learning model that integrates multiple tasks, and improves the accuracy and generalization ability of prediction through shared feature extraction and mutual collaboration between other tasks.

[0063] The first task in the multi-task learning framework model predicts the total coal supply of the unit based on the shared features of the model input data. This task uses the common features in the input data to predict the total coal supply required for the entire unit. The second task predicts the fan current data based on the shared features and the total coal supply of the unit. Based on the first task, the predicted total coal supply is combined to further predict the related fan current data. The third task predicts the fan air volume based on the shared features and the fan current data. Finally, the fan air volume is predicted using the results of the first two tasks, namely the shared features and the fan current data.

[0064] The purpose of Task 1 is to predict the total coal supply of the unit. By processing the shared features and further refining them, the total coal supply of the unit is predicted. With the shared features as input, the total coal supply of the unit is output and passed to Task 2.

[0065] The purpose of Task 2 is to predict the fan current by processing the shared features and the total coal feed data passed by Task 1. The shared features and the total coal feed (predicted by Task 1) are used as inputs to output the fan current and pass it to Task 3.

[0066] The purpose of Task 3 is to predict the air volume of the six fans by processing the shared features and the fan current data. The input data is the shared features and the fan current (predicted by Task 2), and the output is the predicted fan air volume.

[0067] In the above scheme, the total coal supply of the unit predicted by Task 1 is used as the input of Task 2, reflecting the total fuel supply situation and is the basic data for predicting the fan current. There is a close correlation between the total coal supply and the fan current, and the change of the fan current can be predicted more accurately through the total coal supply. The six major fan currents predicted by Task 2 are used as the input of Task 3, which directly affects the change of air volume. There is a significant linear relationship between the fan current and the air volume. The air volume can be predicted more accurately through the current, thereby achieving optimal control of the combustion process.

[0068] The implementation process of this method is as follows Figure 2As shown. Its implementation process is that through the multi-task learning framework, the model can simultaneously learn multiple related tasks, such as coal feed prediction and air volume control. Combined with the LSTM (Long Short-Term Memory Network) model, by sharing information between tasks and using historical data and real-time data to build an accurate prediction model, efficient feature sharing is achieved, and the generalization ability and prediction accuracy of the model are improved. Selecting intermediate (or auxiliary) variables for prediction, such as the six major fan currents, not only helps to capture the inherent correlation between tasks and improve the prediction accuracy of each task, but also gradually builds a hierarchical model structure of complex systems, which is convenient for step-by-step optimization and control. This hierarchical prediction method enables each subtask to cooperate with each other and improve the response speed and control accuracy of the overall system. Therefore, this technical solution predicts the total coal feed of the unit, the six major fan currents and the six major fan air volumes by learning three tasks, respectively, to achieve comprehensive optimization of the combustion process of coal-fired power plants, thereby achieving the goal of energy conservation and emission reduction and ensuring the efficient and safe operation of the system.

[0069] This embodiment provides a method for predicting the total coal feed and air volume of a coal-fired power plant unit based on multi-task learning. Through the multi-task learning framework, the model can share features between different tasks, which improves the generalization ability and prediction accuracy of the model, and can improve the prediction accuracy of the combustion prediction model, especially under a large number of complex combustion reactions, and more accurately reflect the actual combustion state. By learning three tasks (total coal feed, fan current, and fan air volume), the model can enhance its adaptability to the dynamic changes of various parameters during boiler operation, respond to changes in fuel characteristics and environmental conditions in a timely manner, maintain the optimal combustion state, improve combustion efficiency, and reduce pollutant emissions.

[0070] In one embodiment of the present invention, the multi-task learning model architecture includes a feature extraction layer, a task-specific layer, and an output layer.

[0071] Multi-Task Learning (MTL) is a machine learning method that improves the generalization ability and prediction accuracy of the model by sharing information between different tasks. The framework includes a feature extraction layer, a task-specific layer, and a prediction layer, and can simultaneously optimize multiple related tasks, such as total coal feed prediction and air volume control.

[0072] (1) Feature extraction layer

[0073] The feature extraction layer uses the Mixture of Experts (MOE) model to extract common features of the input data by sharing information between tasks. This layer improves the generalization ability and prediction accuracy of the model by capturing the intrinsic correlation between tasks.

[0074] (2) Task-specific layer

[0075] The task-specific layer consists of multiple LSTM layers, one for each task, which is used to further extract task-related features. In this way, the model can focus on the unique requirements of each task, improving the accuracy of prediction and the adaptability of the model.

[0076] (3) Output layer

[0077] The output layer outputs the prediction results of each task through the fully connected layer.

[0078] In addition, embedding the attention mechanism in the model improves the model's ability to focus on key features, thereby improving the feature extraction capability and prediction accuracy under complex working conditions.

[0079] This embodiment can optimize multiple related tasks simultaneously, such as total coal feed prediction and air volume control, through a multi-task learning framework. Compared with the single-task model, this embodiment uses a shared network to extract common features of each task, thereby improving the generalization ability of the model. In practical applications, the model can more accurately predict combustion parameters under different working conditions, significantly improving the efficiency and stability of the combustion control system.

[0080] In a further embodiment of the present invention, a multi-task learning model architecture based on an LSTM network is constructed. The LSTM multi-task learning model architecture includes a feature extraction layer (input layer, shared LSTM layer, multi-task learning shared part (MOE)), a task-specific LSTM layer, and an output layer. The shared LSTM layer extracts the common features of the time series data, and then the task-specific LSTM layer optimizes different tasks. Finally, the output layer gives the prediction results of each task. The structural diagram of the model is shown in FIG. Figure 3 The specific design of each layer is as follows:

[0081] (1) Input layer: Receives time series data, including unit load, operation status parameters of each coal mill (wA, wB, wC, wD, wE, wF), six major wind turbine currents and environmental conditions, etc.

[0082] (2) Channel Attention Mechanism: In the multi-task learning framework, since some features are shared between different tasks, how to effectively extract and utilize these shared features is the key to improving model performance. Traditional feature extraction methods may not be able to fully capture the information of key features when facing complex working conditions. Therefore, the present invention adopts a channel attention mechanism. By performing global average pooling and global maximum pooling on the input feature map, and calculating the weight of each channel, the weight is finally applied to the input feature map, and the feature map is re-weighted to accurately extract key feature data. The channel attention mechanism can focus on features related to key features such as unit load, coal mill operation status parameters, total coal feed of the unit, and current of the six major fans, improve prediction accuracy, and enhance the feature extraction capability and overall performance of the model.

[0083] (3) Mixture of Experts (MOE): MOE is an effective method that significantly expands the capacity of the model while minimizing computational overhead by selectively activating different experts (i.e., different sub-models) in the model. In the MOE layer, there are multiple "expert networks" and one "gated network". The function of the gated network is to dynamically select the most appropriate expert network to process the input data based on the characteristics of the input data, thereby fully utilizing the expertise of the expert network while ensuring efficient computation.

[0084] (4) Shared LSTM layer: It consists of multiple LSTM units and is used to extract common time series features of the input data. This layer improves the effectiveness of feature extraction and enhances the generalization ability of the model by capturing long-term dependencies in the input data.

[0085] (5) Task-specific LSTM layer: For each specific task (such as total coal feed prediction, fan current, and fan air volume), an independent LSTM layer is used to further extract features related to the task. Since there is a certain correlation between multiple tasks, information transfer can use the prediction results of one task as the input of another task. In this way, the model can focus on the unique needs of each task, improving the accuracy of the prediction and the adaptability of the model.

[0086] (6) Output layer: The prediction results of each task are output through the fully connected layer. The output layer of each task independently calculates its corresponding prediction value, such as total coal feed, fan current, and air volume. By combining shared features and task-specific features, the prediction layer can more accurately reflect the actual working conditions and improve the prediction accuracy of the model.

[0087] In the multi-task learning framework of the present invention, the channel attention mechanism is embedded in the feature extraction stage and the task-specific feature processing stage. The specific implementation steps are as follows:

[0088] a. Global average pooling and global maximum pooling: for the input feature map Perform global average pooling and global maximum pooling to obtain global feature description and .

[0089] b. Weight calculation: and They are input into the shared fully connected layer respectively, and the weight of each channel is calculated through activation functions (such as ReLU and Sigmoid), and finally the channel attention weight is obtained. .

[0090] c. Feature reweighting: channel attention weight Applied to each channel of the input feature map, reweighting the feature map to obtain a channel enhanced feature map .

[0091] This embodiment introduces a channel attention mechanism, which enables the model to effectively focus on key features, accurately extract data such as unit load, coal mill operation status parameters, total coal feed of the unit and current of six major fans, and enhance the prediction ability of the model under complex working conditions, thereby achieving precise control and optimization of the combustion process of coal-fired power plants.

[0092] In order to improve the prediction accuracy and generalization ability of the multi-task learning model, a task uncertainty weighted loss is designed in one embodiment of the present invention to improve the overall performance of the model by balancing the error weights of different tasks.

[0093] Uncertainty-Weighted Loss is an adaptive method that adjusts the loss weight of each task according to its uncertainty. The specific steps are as follows:

[0094] a. Task loss function: for each task Define a loss function .

[0095] b. Task weight: Assign a learnable weight parameter to each task , which reflects the uncertainty of the task.

[0096] c. Total loss function: Combine the loss functions and weights of each task to form a total loss function , the calculation formula is preferably as follows:

[0097]

[0098] in, The larger the value, the higher the uncertainty of the task, and the model needs to pay more attention to the task.

[0099] This method can automatically adjust the loss weight of each task, dynamically adjust the model's learning process according to the uncertainty of the task, and improve the model's comprehensive predictive ability for each task.

[0100] Compared with the combustion control method based on physical model, the present invention adopts data-driven approach and simplifies the modeling process through deep learning technology and multi-task learning framework. By introducing dynamic adjustment mechanism, real-time monitoring and prediction of various parameters, the model can quickly respond to changes in working conditions and ensure the optimal state of the combustion process. Compared with traditional physical models, the present invention has significant advantages in adaptability and real-time performance.

[0101] Based on the same inventive concept, the embodiment of the present application also provides a device for predicting the total coal supply and air volume of a coal-fired power plant unit based on multi-task learning, which is used to implement the method for predicting the total coal supply and air volume of a coal-fired power plant unit based on multi-task learning. The implementation scheme for solving the problem provided by the system is similar to the implementation scheme recorded in the above method, so the specific limitations in the embodiment of the device for predicting the total coal supply and air volume of a coal-fired power plant unit based on multi-task learning provided below can be referred to the limitations of the method for predicting the total coal supply and air volume of a coal-fired power plant unit based on multi-task learning above, and will not be repeated here.

[0102] See also Figure 4 This embodiment provides a device for predicting the total coal supply and air volume of a coal-fired power plant unit based on multi-task learning, comprising:

[0103] Data acquisition module, used to obtain unit load and operation status parameters of each coal mill;

[0104] The prediction module is used to input the unit load and the operating status parameters of each coal mill as model input data into the pre-trained multi-task learning framework model to obtain the prediction results of the total coal feed and air volume of the unit;

[0105] Among them, the multi-task learning framework model is trained using multi-task learning technology. The multi-task learning framework model is used to perform at least the first task, the second task and the third task. The first task is used to predict the total coal feed of the unit based on the shared characteristics of the model input data, the second task is used to predict the fan current data based on the shared characteristics and the total coal feed of the unit, and the third task is used to predict the fan air volume based on the shared characteristics and the fan current data.

[0106] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0107] Reference Figure 5 An embodiment of the present invention further provides a computer device, comprising: a memory and a processor and a computer program stored in the memory. When the computer program is executed on the processor, a method for predicting the total coal supply and air volume of a coal-fired power plant unit based on multi-task learning as described in any one of the above methods is implemented.

[0108] The computer device may be a desktop computer, a notebook, a PDA, a cloud server or other computing device. The computer device may include, but is not limited to, a processor and a memory. Those skilled in the art will understand that Figure 5 It is only an example of a computer device and does not constitute a limitation of the computer device. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, it may also include input and output devices, network access devices, etc.

[0109] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0110] In some embodiments, the memory may be an internal storage unit of the computer device, such as a hard disk or memory of the computer device. In other embodiments, the memory may also be an external storage device of the computer device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the computer device. Further, the memory may include both an internal storage unit and an external storage device of the computer device. The memory is used to store an operating system, an application program, a boot loader (BootLoader), data, and other programs, such as the program code of the computer program. The memory may also be used to temporarily store data that has been output or is to be output.

[0111] An embodiment of the present invention also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for predicting the total coal feed and air volume of a coal-fired power plant unit based on multi-task learning as described in any of the above methods is implemented.

[0112] In this embodiment, if the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium can at least include: any entity or device that can carry the computer program code to the camera / terminal device, recording medium, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. For example, USB flash drive, mobile hard disk, disk or optical disk. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electric carrier signals and telecommunication signals.

[0113] An embodiment of the present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements a method for predicting the total coal feed and air volume of a coal-fired power plant unit based on multi-task learning as described in any one of the above methods.

[0114] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0115] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0116] In the embodiments disclosed in the present application, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are only schematic. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0117] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting the total coal supply and air volume of coal-fired power plant units based on multi-task learning, characterized in that: The steps include: Obtain unit load and operation status parameters of each coal mill; The unit load and the operating status parameters of each coal mill are used as model input data and input into the pre-trained multi-task learning framework model to obtain the prediction results of the total coal feed rate and air volume of the unit; Among them, the multi-task learning framework model is trained using multi-task learning technology, and the multi-task learning framework model is used to perform at least a first task, a second task and a third task. The first task is used to predict the total coal feed of the unit based on the shared characteristics of the model input data, the second task is used to predict the fan current data based on the shared characteristics and the total coal feed of the unit, and the third task is used to predict the fan air volume based on the shared characteristics and the fan current data.

2. The method for predicting the total coal supply and air volume of a coal-fired power plant unit based on multi-task learning according to claim 1 is characterized in that: The architecture of the multi-task learning framework model includes: A feature extraction layer, used for extracting common features from the model input data as the shared features; A task-specific layer, comprising a plurality of LSTM layers, each of which corresponds to a task, and each of which is used to extract task-specific features related to the corresponding task based on the shared features and the prediction results of the associated tasks; The output layer is used to output the prediction results of each task based on the shared features and the task-specific features of each task.

3. The method for predicting the total coal supply and air volume of a coal-fired power plant unit based on multi-task learning according to claim 2 is characterized in that: The feature extraction layer comprises: An input layer, used to receive various time series data including the unit load and operation status parameters of each coal mill; A channel attention mechanism layer is used to increase the feature weights related to key features in each type of time series data; A hybrid expert model layer is used to select the most appropriate expert network to process the corresponding data according to the characteristics of each type of time series data; The shared LSTM layer is used to extract common time series features from the processed time series data as the shared features.

4. The method for predicting the total coal supply and air volume of a coal-fired power plant unit based on multi-task learning according to claim 3 is characterized in that: The process of processing data in the channel attention mechanism layer includes: Perform global average pooling and global maximum pooling on the input feature map to obtain the average description and maximum description of the global features; The average description and the maximum description are respectively input into the shared fully connected layer, and the weight of each channel is calculated through the activation function to finally obtain the channel attention weight; The channel attention weight is applied to each channel of the input feature map, and the feature map is reweighted to obtain a channel enhanced feature map.

5. The method for predicting the total coal supply and air volume of a coal-fired power plant unit based on multi-task learning according to claim 3 is characterized in that: The multi-task learning framework model uses a task uncertainty weighted loss method to adjust the loss weight of each task. The steps of the task uncertainty weighted loss method include: Define a loss function for each task; Assign a learnable weight parameter to each task; The loss function and the weight parameters of each task are combined to form a total loss function.

6. The method for predicting the total coal supply and air volume of a coal-fired power plant unit based on multi-task learning according to claim 5 is characterized in that: The expression of the total loss function is as follows: In the formula, is the total loss function, is the loss function of the i-th task, is the weight parameter of the i-th task.

7. A device for predicting the total coal supply and air volume of a coal-fired power plant unit based on multi-task learning, characterized in that: include: Data acquisition module, used to obtain unit load and operation status parameters of each coal mill; A prediction module, used to input the unit load and the operation status parameters of each coal mill as model input data into a pre-trained multi-task learning framework model to obtain prediction results of the total coal feed rate and air volume of the unit; Among them, the multi-task learning framework model is trained using multi-task learning technology, and the multi-task learning framework model is used to perform at least a first task, a second task and a third task. The first task is used to predict the total coal feed of the unit based on the shared characteristics of the model input data, the second task is used to predict the fan current data based on the shared characteristics and the total coal feed of the unit, and the third task is used to predict the fan air volume based on the shared characteristics and the fan current data.

8. A computer device, characterized in that: The device comprises a processor and a memory: The memory is used to store a computer program and send instructions of the computer program to the processor; The processor executes a method for predicting the total coal feed and air volume of a coal-fired power plant unit based on multi-task learning as described in any one of claims 1-6 according to the instructions of the computer program.

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 a method for predicting the total coal feed and air volume of a coal-fired power plant unit based on multi-task learning as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, it implements a method for predicting the total coal feed and air volume of a coal-fired power plant unit based on multi-task learning as described in any one of claims 1-6.

Images