Combustion optimization method and device based on machine learning and electronic equipment

Abstract

According to the combustion optimization method and device based on machine learning and the electronic equipment, the real-time operation data of the to-be-optimized system is input into the trained multi-target prediction model for data prediction processing, and the multi-target prediction result is obtained; the trained multi-target prediction model is a model which is established based on a graph neural network and is used for predicting performance indexes of the to-be-optimized system; according to a multi-target preset result and the real-time operation data, through the trained combustion optimization model, oxygen amount and air door opening adjustment processing is carried out on the to-be-optimized system, and the to-be-optimized system after combustion optimization is obtained; the trained combustion optimization model is a model which is constructed based on a distributed reinforcement learning algorithm and is used for carrying out combustion optimization on the to-be-optimized system by dynamically adjusting the oxygen content and the air door opening degree of the to-be-optimized system. Compared with the prior art, the embodiment of the invention can respond to the requirement of state change in real time in the combustion optimization process, improve the optimization effect and reduce the operation complexity.

Description

Technical Field

[0001] The present disclosure relates to the field of energy technology, and in particular to a combustion optimization method and device based on machine learning, and electronic equipment. Background Art

[0002] As an important part of traditional energy generation, the combustion optimization technology of thermal power plants is of great significance in improving combustion efficiency, reducing pollutant emissions and achieving green environmental protection goals. Existing thermal power plant combustion optimization methods mostly use adjustment methods based on artificial experience or simple mathematical models. For example, by adjusting boiler operating parameters (such as oxygen content, SOFA air damper opening, etc.) and combining hot adjustment tests, the optimal operating conditions under different loads are determined.

[0003] However, existing combustion optimization methods generally have the following defects: they mostly rely on the experience of operators to adjust parameters, and the test cycle is long, making it difficult to quickly adapt to the dynamic changes in the boiler's operating status, resulting in low optimization efficiency; there is a complex nonlinear relationship between boiler operating parameters and multi-objective performance indicators (such as combustion efficiency, pollutant emissions, steam temperature deviation, etc.), which is difficult to accurately describe with simple mathematical models, resulting in poor operation optimization effects; since existing combustion optimization methods require multiple tests to obtain the best combination of operating parameters, they cannot meet the dynamic adjustment needs under real-time changes in the boiler's operating status; there may be conflicts between different objectives (for example, improving combustion efficiency may lead to increased pollutant emissions).

[0004] Therefore, how to effectively solve the defects of poor real-time performance, insufficient optimization effect and complex operation in the combustion optimization process is a problem that needs to be solved urgently. Summary of the invention

[0005] The present disclosure provides a combustion optimization method, device, electronic device and storage medium based on machine learning, the main purpose of which is to effectively solve the defects of poor real-time performance, insufficient optimization effect and complex operation in the combustion optimization process.

[0006] According to a first aspect of the present disclosure, a combustion optimization method based on machine learning is provided, which comprises:

[0007] Input the real-time operation data of the system to be optimized into the trained multi-objective prediction model for data prediction processing to obtain a multi-objective prediction result; wherein the trained multi-objective prediction model is a model established based on a graph neural network and used to predict the performance indicators of the system to be optimized;

[0008] According to the multi-objective preset results and the real-time operation data, the oxygen content and air door opening of the system to be optimized are adjusted by the trained combustion optimization model to obtain the system to be optimized after combustion optimization;

[0009] Among them, the trained combustion optimization model is constructed based on a distributed reinforcement learning algorithm, and is a model used to optimize the combustion of the system to be optimized by dynamically adjusting the oxygen content and air valve opening of the system to be optimized.

[0010] Optionally, the step of inputting the target operation data into a trained multi-objective prediction model for data prediction processing to obtain a multi-objective prediction result includes:

[0011] The target operation data is input into a node feature embedding module for high-dimensional feature embedding to obtain first feature data corresponding to the target operation data; the trained multi-target prediction model includes the node feature embedding module, a graph convolution module based on an attention mechanism, and a fully connected layer output module;

[0012] Inputting the first feature data into the graph convolution module based on the attention mechanism to perform data relationship modeling processing to obtain multi-dimensional correlation feature data;

[0013] The multi-dimensional correlation feature data is input into the fully connected layer output module for full connection processing to obtain the multi-target prediction result.

[0014] Optionally, the oxygen content and air door opening of the system to be optimized are adjusted by a trained combustion optimization model according to the multi-objective preset result and the target operation data, so as to obtain the system to be optimized after combustion optimization, comprising:

[0015] The fitness of each first optimization model to the current working condition of the system to be optimized is evaluated by a preset evaluation algorithm to obtain a target optimization model; wherein the trained combustion optimization model includes a plurality of the first optimization models, and the target optimization model is the model with the highest fitness to the current working condition of the system to be optimized among the trained combustion optimization models;

[0016] According to the multi-objective preset results and the target operating data, combustion optimization processing is performed through the target optimization model to obtain target optimization parameters;

[0017] The oxygen content and air door opening of the system to be optimized are adjusted by the target optimization parameters to obtain the system to be optimized after the combustion optimization.

[0018] Optionally, performing combustion optimization processing through the target optimization model according to the multi-objective preset results and the target operation data to obtain target optimization parameters includes:

[0019] Acquiring operating condition information and preset constraint information of the system to be optimized;

[0020] When the operating condition information and / or the constraint information determine that the system to be optimized meets environmental protection requirements, based on the first preset weight information, the combustion optimization process is performed through the target optimization model according to the multi-objective preset results and the target operating data to obtain a first optimization parameter;

[0021] Wherein, the first optimization parameter is at least used to reduce the combustion efficiency of the system to be optimized and restrict pollutant emissions;

[0022] When the operating condition information and / or the constraint information determine that the system to be optimized is under high load demand, based on the second preset weight information, the combustion optimization process is performed by the target optimization model according to the multi-objective preset result and the target operating data to obtain a second optimization parameter;

[0023] The second optimization parameter is at least used to improve the combustion efficiency of the system to be optimized and reduce the constraints on pollutant emissions, and the target optimization parameter includes at least the first optimization parameter or the second optimization parameter.

[0024] Optionally, before obtaining the operating data of the system to be optimized, the method further includes:

[0025] Acquiring the real-time operation data of the system to be optimized, and performing data preprocessing on the real-time operation data to obtain target data for training;

[0026] Establishing an initial multi-objective prediction model based on the graph neural network, performing model training processing on the initial multi-objective prediction model through the training target data, and obtaining the trained multi-objective prediction model;

[0027] Based on the trained multi-objective prediction model, a combustion optimization strategy is constructed through the distributed reinforcement learning algorithm, and model training processing is performed on multiple initial combustion optimization models according to the combustion optimization strategy to obtain the trained combustion optimization model.

[0028] Optionally, the performing data preprocessing on the real-time operation data to obtain target data for training includes:

[0029] Using an outlier detection algorithm to remove abnormal data points from the real-time operation data to obtain first training data;

[0030] Smoothing the time series data of the first training data using a sliding average method to obtain second training data;

[0031] Normalizing the second training data to obtain third training data;

[0032] The third training data is labeled according to historical operating data and target performance indicators to obtain the target training data, wherein the historical operating data is the operating data of the system to be optimized at a time before the real-time operating data, and the target performance indicator is the expected performance indicator that the system to be optimized needs to achieve.

[0033] Optionally, the constructing a combustion optimization strategy based on the trained multi-objective prediction model by using the distributed reinforcement learning algorithm includes:

[0034] Inputting the training target data into the trained multi-objective prediction model to perform data prediction processing to obtain the training multi-objective prediction result;

[0035] Determine the state space according to the multi-objective prediction result for training and the operating condition information of the system to be optimized;

[0036] The action space is determined according to a preset oxygen quantity adjustment range and a preset air door opening adjustment range;

[0037] Data construction and processing are performed according to the state space, the action space and a preset reward function to obtain the combustion optimization strategy, wherein the preset reward function is used to balance the multi-objective prediction results, and the multi-objective prediction results include at least combustion efficiency, pollutant emission concentration and reheater steam temperature deviation.

[0038] Optionally, after the system to be optimized is processed by adjusting the oxygen content and the air door opening according to the multi-objective preset results and the real-time operation data through the trained combustion optimization model to obtain the system to be optimized after combustion optimization, the method further includes:

[0039] Collecting the actual performance index of the system to be optimized in real time, and calculating the deviation value between the actual performance index and the multi-objective preset result;

[0040] The trained multi-objective prediction model is updated by an adaptive gradient descent algorithm according to the deviation value to obtain an updated multi-objective prediction model.

[0041] The trained combustion optimization model is updated based on the updated multi-objective prediction model to obtain an updated combustion optimization model.

[0042] According to a second aspect of the present disclosure, a combustion optimization device based on machine learning is provided, comprising:

[0043] A prediction unit, used to input the real-time operation data of the system to be optimized into a trained multi-objective prediction model for data prediction processing to obtain a multi-objective prediction result; wherein the trained multi-objective prediction model is a model established based on a graph neural network and used to predict the performance indicators of the system to be optimized;

[0044] An adjustment unit, configured to adjust the oxygen content and air door opening of the system to be optimized by using a trained combustion optimization model according to the multi-objective preset results and the real-time operation data, so as to obtain the system to be optimized after combustion optimization;

[0045] Among them, the trained combustion optimization model is constructed based on a distributed reinforcement learning algorithm, and is a model used to optimize the combustion of the system to be optimized by dynamically adjusting the oxygen content and air valve opening of the system to be optimized.

[0046] Optionally, the prediction unit includes:

[0047] An embedding module, used to input the target operation data into a node feature embedding module for high-dimensional feature embedding to obtain first feature data corresponding to the target operation data; the trained multi-target prediction model includes the node feature embedding module, a graph convolution module based on an attention mechanism, and a fully connected layer output module;

[0048] A modeling module, used for inputting the first feature data into the graph convolution module based on the attention mechanism to perform data relationship modeling processing to obtain multi-dimensional correlation feature data;

[0049] A processing module is used to input the multi-dimensional correlation feature data into the fully connected layer output module for full connection processing to obtain the multi-target prediction result.

[0050] Optionally, the adjusting unit includes:

[0051] An evaluation module, used to evaluate the fitness of each first optimization model to the current operating condition of the system to be optimized by a preset evaluation algorithm, and obtain a target optimization model; wherein the trained combustion optimization model includes a plurality of the first optimization models, and the target optimization model is the model with the highest fitness to the current operating condition of the system to be optimized among the trained combustion optimization models;

[0052] An optimization module, used for performing combustion optimization processing through the target optimization model according to the multi-objective preset results and the target operation data to obtain target optimization parameters;

[0053] The regulating module is used to regulate the oxygen content and air door opening of the system to be optimized by using the target optimization parameters to obtain the system to be optimized after the combustion is optimized.

[0054] Optionally, the optimization module is further used for:

[0055] Acquiring operating condition information and preset constraint information of the system to be optimized;

[0056] When the operating condition information and / or the constraint information determine that the system to be optimized meets environmental protection requirements, based on the first preset weight information, the combustion optimization process is performed through the target optimization model according to the multi-objective preset results and the target operating data to obtain a first optimization parameter;

[0057] Wherein, the first optimization parameter is at least used to reduce the combustion efficiency of the system to be optimized and restrict pollutant emissions;

[0058] When the operating condition information and / or the constraint information determine that the system to be optimized is under high load demand, based on the second preset weight information, the combustion optimization process is performed by the target optimization model according to the multi-objective preset result and the target operating data to obtain a second optimization parameter;

[0059] The second optimization parameter is at least used to improve the combustion efficiency of the system to be optimized and reduce the constraints on pollutant emissions, and the target optimization parameter includes at least the first optimization parameter or the second optimization parameter.

[0060] Optionally, the device further comprises:

[0061] An acquisition unit, used for acquiring the real-time operation data of the system to be optimized;

[0062] A processing unit, used for performing data preprocessing on the real-time operation data to obtain target data for training;

[0063] A first training unit is used to establish an initial multi-objective prediction model based on the graph neural network, and perform model training on the initial multi-objective prediction model through the training target data to obtain the trained multi-objective prediction model;

[0064] A construction unit, configured to construct a combustion optimization strategy through the distributed reinforcement learning algorithm based on the trained multi-objective prediction model;

[0065] The second training unit is used to perform model training processing on multiple initial combustion optimization models according to the combustion optimization strategy to obtain the trained combustion optimization model.

[0066] Optionally, the processing unit includes:

[0067] A removal module, used for removing abnormal data points from the real-time operation data using an abnormal value detection algorithm to obtain first training data;

[0068] a smoothing module, configured to smooth the time series data of the first training data using a sliding average method to obtain second training data;

[0069] a processing module, configured to perform normalization processing on the second training data to obtain third training data;

[0070] A labeling module is used to perform data labeling processing on the third training data according to historical operating data and target performance indicators to obtain the target training data, wherein the historical operating data is the operating data of the system to be optimized at a time before the real-time operating data, and the target performance indicator is the expected performance indicator that the system to be optimized needs to achieve.

[0071] Optionally, the construction unit includes:

[0072] A prediction module, used for inputting the training target data into the trained multi-objective prediction model to perform data prediction processing to obtain a training multi-objective prediction result;

[0073] A determination module, used to determine the state space according to the multi-objective prediction result for training and the operating condition information of the system to be optimized;

[0074] The determination module is further used to determine the action space according to a preset oxygen quantity adjustment range and a preset air door opening adjustment range;

[0075] A construction module is used to perform data construction processing according to the state space, the action space and a preset reward function to obtain the combustion optimization strategy, wherein the preset reward function is used to balance the multi-objective prediction results, and the multi-objective prediction results at least include combustion efficiency, pollutant emission concentration and reheater steam temperature deviation.

[0076] Optionally, the device further comprises:

[0077] A collection unit, used for collecting the actual performance index of the system to be optimized in real time, and calculating the deviation value between the actual performance index and the multi-objective preset result;

[0078] An updating unit is used to update the trained multi-objective prediction model through an adaptive gradient descent algorithm according to the deviation value to obtain an updated multi-objective prediction model.

[0079] The updating unit is further used to update the trained combustion optimization model based on the updated multi-objective prediction model to obtain an updated combustion optimization model.

[0080] According to a third aspect of the present disclosure, there is provided an electronic device, including:

[0081] at least one processor; and

[0082] a memory communicatively connected to the at least one processor; wherein,

[0083] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method described in the first aspect.

[0084] According to a fourth aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable the computer to execute the method described in the first aspect.

[0085] According to a fifth aspect of the present disclosure, a computer program product is provided, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements the method as described in the first aspect above.

[0086] The machine learning-based combustion optimization method, device, and electronic device provided by the present disclosure input the real-time operating data of the system to be optimized into a trained multi-objective prediction model for data prediction processing to obtain a multi-objective prediction result; wherein the trained multi-objective prediction model is a model established based on a graph neural network, and is used to predict the performance indicators of the system to be optimized; according to the multi-objective preset results and the real-time operating data, the oxygen content and air valve opening of the system to be optimized are adjusted by the trained combustion optimization model to obtain the system to be optimized after combustion optimization; wherein the trained combustion optimization model is a model constructed based on a distributed reinforcement learning algorithm, and is used to dynamically adjust the oxygen content and air valve opening of the system to be optimized to perform combustion optimization on the system to be optimized. Compared with the related art, the embodiment of the present disclosure uses a combustion optimization model trained by a distributed reinforcement learning algorithm to achieve dynamic adjustment of the oxygen content and the damper opening according to the real-time operating data of the system to be optimized, so as to meet the demand for rapid changes in the operating state to be optimized. At the same time, it can also coordinate multi-objective optimization needs and respond to the needs of state changes in real time during the combustion optimization process. Through the multi-objective prediction model trained based on the graph neural network, it can fully predict the complex relationship between the operating parameters and performance indicators of the system to be optimized, obtain the multi-objective prediction results, and improve the prediction accuracy and the optimization effect of combustion optimization. Therefore, the embodiment of the present disclosure can respond to the needs of state changes in real time during the combustion optimization process, improve the optimization effect and reduce the complexity of operation.

[0087] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0089] Figure 1 A flow chart of a combustion optimization method based on machine learning provided in an embodiment of the present disclosure;

[0090] Figure 2 A schematic flow chart of another combustion optimization method based on machine learning provided in an embodiment of the present disclosure;

[0091] Figure 3 A schematic diagram of the structure of another combustion optimization device based on machine learning provided in an embodiment of the present disclosure;

[0092] Figure 4 A structural schematic diagram of a combustion optimization device based on machine learning provided in an embodiment of the present disclosure;

[0093] Figure 5 A schematic block diagram of an electronic device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0094] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0095] The following describes the combustion optimization method, device, electronic device and storage medium based on machine learning according to an embodiment of the present disclosure with reference to the accompanying drawings.

[0096] Figure 1 A flow chart of a combustion optimization method based on machine learning provided in an embodiment of the present disclosure.

[0097] like Figure 1 As shown, the method comprises the following steps:

[0098] Step 101, input the real-time operating data of the system to be optimized into the trained multi-objective prediction model for data prediction processing to obtain a multi-objective prediction result; wherein the trained multi-objective prediction model is a model established based on a graph neural network, and is used to predict the performance indicators of the system to be optimized.

[0099] In the embodiments of the present disclosure, the systems to be optimized are various systems that require combustion optimization, such as thermal power plant boiler systems, industrial furnace systems, melting furnaces and heat treatment furnaces, and other systems that need to work through combustion. Specifically, the embodiments of the present disclosure do not limit the systems to be optimized.

[0100] For ease of understanding, the system to be optimized in the subsequent embodiments is described by taking a thermal power plant boiler system as an example.

[0101] The trained multi-objective prediction model is based on an improved graph neural network (GNN), and the model is trained by processing real-time operation data to obtain a trained multi-objective prediction model. The training process includes but is not limited to: constructing an initial multi-objective prediction model based on the improved GNN, and performing model training processing on the initial multi-objective prediction model by processing real-time operation data to obtain a trained multi-objective prediction model.

[0102] The real-time operating data is data generated in real time during the operation of the system to be optimized, such as operating parameters of the boiler system of a thermal power plant, including but not limited to steam pressure, steam temperature, burner outlet temperature, exhaust temperature, oxygen content, fuel consumption, flue gas composition and other key parameters for measuring the operating status of the system to be optimized. Specifically, the present disclosed embodiment does not limit the content of the real-time operating data.

[0103] Among them, the real-time operating data includes the operating parameters and performance indicators of the system to be optimized. The process of performing data prediction processing in a trained multi-objective prediction model is a process of fully mining the complex nonlinear relationship between the operating parameters and performance indicators of the system to be optimized based on the real-time operating data. The multi-objective prediction results are the predicted performance indicators of the system to be optimized, such as: combustion efficiency, pollutant emission concentration (for example: NOx emission concentration) and reheater steam temperature deviation.

[0104] Specifically, the structure and functions of the trained multi-objective prediction model include but are not limited to the following: the trained multi-objective prediction model includes a node feature embedding module, a graph convolution module based on an attention mechanism, and a fully connected layer output module. The node feature embedding module is used to embed high-dimensional features into the implementation operation data; the graph convolution module based on the attention mechanism is used to model the relationship between the implementation operation data, extract multi-dimensional correlation features from the data after embedding high-dimensional features, and capture the nonlinear correlation between the implementation operation data; the fully connected layer output module is used to output the multi-objective prediction results of the system to be optimized based on the multi-dimensional correlation features, such as: combustion efficiency, pollutant emission concentration, and reheater steam temperature deviation.

[0105] The improved graph neural network dynamically constructs an adjacency matrix between real-time operating data to reflect the dynamic correlation between parameters under different operating conditions. It can dynamically update the connection relationship between the operating data of the system to be optimized according to different operating conditions to improve the accuracy of multi-objective prediction.

[0106] Step 102, according to the multi-objective preset results and the real-time operating data, the oxygen content and the damper opening of the system to be optimized are adjusted by the trained combustion optimization model to obtain the system to be optimized after combustion optimization; wherein the trained combustion optimization model is constructed based on a distributed reinforcement learning algorithm, and is a model for performing combustion optimization on the system to be optimized by dynamically adjusting the oxygen content and the damper opening of the system to be optimized.

[0107] In the disclosed embodiment, the trained combustion optimization model needs to be pre-deployed into the control system of the system to be optimized, such as the control system of the boiler in a thermal power plant, etc. The deployment process is implemented through edge computing devices, and the edge devices use a lightweight model structure to ensure low latency in the optimization process. According to the real-time collection of real-time operating data of the system to be optimized, the oxygen content and the damper opening are dynamically adjusted through the trained combustion optimization model to optimize the combustion efficiency and pollutant emissions of the system to be optimized.

[0108] At the same time, the dynamic output of the trained combustion optimization model can be directly embedded in the DCS (distributed control system) of the system to be optimized to achieve seamless connection; the controller delay compensation module is introduced to ensure that the time difference between the model adjustment output and the actual response of the system to be optimized is minimized, thereby improving the real-time performance of the optimization.

[0109] In order to solve the performance degradation problem that may occur in the actual operation of the trained combustion optimization model, an adaptive learning module can be introduced to update the model according to the real-time collected data: for small-scale parameter deviations, the trained combustion optimization model can be quickly corrected through incremental learning; when the operating conditions of the system to be optimized undergo major changes (for example, changes in fuel type or equipment aging), transfer learning can be used to update the model.

[0110] It should be noted that the damper opening refers to the opening size of the regulating valve used to control the air flow entering the combustion chamber of the system to be optimized, wherein the damper opening is mainly the damper opening of the SOFA damper (Separated OverFire Air Damper). The SOFA damper is used to control the air flow above the separated furnace, with the aim of reducing pollutant emissions and improving combustion efficiency.

[0111] The machine learning-based combustion optimization method provided by the present disclosure inputs the real-time operating data of the system to be optimized into a trained multi-objective prediction model for data prediction processing to obtain a multi-objective prediction result; wherein the trained multi-objective prediction model is a model established based on a graph neural network, and is used to predict the performance indicators of the system to be optimized; according to the multi-objective preset results and the real-time operating data, the oxygen content and air damper opening of the system to be optimized are adjusted by the trained combustion optimization model to obtain the system to be optimized after combustion optimization; wherein the trained combustion optimization model is a model constructed based on a distributed reinforcement learning algorithm, and is used to dynamically adjust the oxygen content and air damper opening of the system to be optimized to perform combustion optimization on the system to be optimized. Compared with the related art, the embodiment of the present disclosure uses a combustion optimization model trained by a distributed reinforcement learning algorithm to achieve dynamic adjustment of the oxygen content and the damper opening according to the real-time operating data of the system to be optimized, so as to meet the demand for rapid changes in the operating state to be optimized. At the same time, it can also coordinate multi-objective optimization needs and respond to the needs of state changes in real time during the combustion optimization process. Through the multi-objective prediction model trained based on the graph neural network, it can fully predict the complex relationship between the operating parameters and performance indicators of the system to be optimized, obtain the multi-objective prediction results, and improve the prediction accuracy and the optimization effect of combustion optimization. Therefore, the embodiment of the present disclosure can respond to the needs of state changes in real time during the combustion optimization process, improve the optimization effect and reduce the complexity of operation.

[0112] In one implementable manner of the embodiment of the present disclosure, as a refinement of the above-mentioned step 101, when performing data prediction processing, it can be implemented in the following manner but not limited to: inputting the target operation data into a node feature embedding module for high-dimensional feature embedding to obtain first feature data corresponding to the target operation data; the trained multi-target prediction model includes the node feature embedding module, a graph convolution module based on an attention mechanism, and a fully connected layer output module; inputting the first feature data into the graph convolution module based on the attention mechanism for data relationship modeling processing to obtain multi-dimensional correlation feature data; inputting the multi-dimensional correlation feature data into the fully connected layer output module for full connection processing to obtain the multi-target prediction result.

[0113] In one implementable manner of the disclosed embodiment, as a refinement of the above-mentioned step 102, when the oxygen content and air damper opening of the system to be optimized are adjusted, it can also be implemented in the following manner but not limited to: the fitness of each first optimization model to the current operating condition of the system to be optimized is evaluated by a preset evaluation algorithm to obtain a target optimization model; wherein the trained combustion optimization model contains multiple first optimization models, and the target optimization model is the model with the highest fitness for the current operating condition of the system to be optimized among the trained combustion optimization models; combustion optimization is performed through the target optimization model according to the multi-objective preset results and the target operating data to obtain target optimization parameters; the oxygen content and air damper opening of the system to be optimized are adjusted by the target optimization parameters to obtain the system to be optimized after combustion optimization.

[0114] In the disclosed embodiment, the preset evaluation algorithm is a custom algorithm used to evaluate the fitness of each first optimization model to the current working condition of the system to be optimized. When the trained combustion optimization model is deployed in the system to be optimized, multiple independent training combustion optimization models (for example, models constructed based on different training data, loss functions or algorithm structures) will be deployed in combination with multi-model integration technology. When optimizing the system to be optimized, the following methods may be used but are not limited to: Model voting mechanism: obtaining the final optimization parameters by weighted voting on the prediction results of different models;

[0115] Model confidence assessment: Real-time assessment of the adaptability of each model to the current working conditions, and dynamic selection of the optimal model as the basis for decision-making;

[0116] Abnormal switching mechanism: When the main model is abnormal or the prediction deviation is large, it quickly switches to the backup model to ensure operational stability.

[0117] In an implementable manner of the embodiment of the present disclosure, when performing combustion optimization processing, it can also be implemented in the following manner but is not limited to: obtaining the operating condition information and preset constraint information of the system to be optimized; when the operating condition information and / or the constraint information determine that the system to be optimized is in environmental protection requirements, based on the first preset weight information, the combustion optimization processing is performed through the target optimization model according to the multi-objective preset results and the target operating data, and a first optimization parameter is obtained; wherein the first optimization parameter is at least used to reduce the combustion efficiency of the system to be optimized and constrain pollutant emissions; when the operating condition information and / or the constraint information determine that the system to be optimized is in high load demand, based on the second preset weight information, the combustion optimization processing is performed through the target optimization model according to the multi-objective preset results and the target operating data, and a second optimization parameter is obtained; wherein the second optimization parameter is at least used to improve the combustion efficiency of the system to be optimized and reduce the constraints on pollutant emissions, and the target optimization parameter includes at least the first optimization parameter or the second optimization parameter.

[0118] In the embodiment of the present disclosure, the constraint information is some pre-set external constraint conditions, such as: environmental protection index restrictions, load demand changes, etc. When performing combustion optimization processing of the system to be optimized, the weight of the optimization target can be dynamically adjusted according to the operating condition information and / or the constraint information. Specifically, the following methods can be adopted but not limited to: when environmental protection is given priority (that is, when the system to be optimized is in environmental protection demand), the weight of the combustion efficiency target can be reduced, and the weight of the NOx and CO2 emission constraints can be increased;

[0119] When the load demand is high (ie, the system to be optimized is under the condition of high load demand), the combustion efficiency weight can be increased preferentially, while the pollutant emission constraints can be appropriately relaxed.

[0120] The first preset weight information is a kind of weight information of custom settings, for example: reducing the weight of optimizing combustion efficiency and increasing the weight of pollutant emission constraints (i.e., the weight of optimizing combustion efficiency is less than the weight of pollutant emission constraints), etc. The second preset weight information is also a kind of weight information of custom settings, for example: increasing the weight of optimizing combustion efficiency and reducing the weight of pollutant emission constraints (i.e., the weight of optimizing combustion efficiency is greater than the weight of pollutant emission constraints), etc. Specifically, the present disclosure embodiment does not limit the content of the first preset weight information and the second preset weight information.

[0121] In one implementable manner of the embodiment of the present disclosure, it is necessary to pre-train the model to complete the combustion optimization processing of the system to be optimized. Therefore, in order to realize the combustion optimization processing, the following method may also be adopted but is not limited to: obtaining the real-time operation data of the system to be optimized, and performing data preprocessing on the real-time operation data to obtain training target data; establishing an initial multi-objective prediction model based on the graph neural network, and performing model training processing on the initial multi-objective prediction model through the training target data to obtain the trained multi-objective prediction model; based on the trained multi-objective prediction model, constructing a combustion optimization strategy through the distributed reinforcement learning algorithm, and performing model training processing on multiple initial combustion optimization models according to the combustion optimization strategy to obtain the trained combustion optimization model.

[0122] In the embodiments of the present disclosure, the model training and deployment process may be carried out in the following manners but is not limited to: S1. Collecting the operating parameters of the thermal power plant boiler (i.e., obtaining the real-time operating data of the system to be optimized) and performing preprocessing; S2. Establishing a multi-objective prediction model based on a graph neural network, and using the preprocessed boiler operating parameters (i.e., the target training data) to train the multi-objective prediction model to obtain a trained multi-objective prediction model; S3. Based on the trained multi-objective prediction model, a multi-agent distributed reinforcement learning algorithm is used to construct a combustion optimization strategy, and a combustion optimization model capable of dynamically adjusting the oxygen content and the SOFA air damper opening is obtained through training; S4. Deploying the trained combustion optimization model to the boiler control system, and dynamically adjusting the oxygen content and the SOFA air damper opening through the combustion optimization model according to the real-time collected boiler operating parameters (real-time operating data); S5. Comparing the prediction results of the multi-objective prediction model with the multi-objective performance indicators of the actual operation of the boiler in real time, and retraining the multi-objective prediction model and the combustion optimization model according to the deviation, so as to continuously improve the optimization performance.

[0123] Among them, when obtaining the real-time operation data, the following methods can be used but are not limited to: collecting boiler operation parameters in real time through a distributed sensor network to obtain implementation operation data, and using edge computing equipment to perform preliminary processing on the implementation operation data and then upload it to the central server.

[0124] In one implementable manner of the embodiment of the present disclosure, when preprocessing the real-time operation data, the following manner may also be adopted but is not limited to: using an outlier detection algorithm to remove abnormal data points from the real-time operation data to obtain first training data; using a sliding average method to smooth the time series data of the first training data to obtain second training data; normalizing the second training data to obtain third training data; and labeling the third training data according to historical operation data and target performance indicators to obtain the target training data, wherein the historical operation data is the operation data of the system to be optimized at a time before the real-time operation data, and the target performance indicator is the expected performance indicator that the system to be optimized needs to achieve.

[0125] In the embodiment of the present disclosure, when collecting and preprocessing real-time data, the boiler operating parameters can be collected in real time through a distributed sensor network to obtain implementation operation data, and the implementation operation data can be preliminarily processed by edge computing equipment and uploaded to the central server.

[0126] Methods for preprocessing real-time operation data include but are not limited to:

[0127] 1) Use statistically based outlier detection algorithms to eliminate abnormal data that deviate from the range;

[0128] 2) Use the sliding average method to smooth the time series data;

[0129] 3) Each parameter is normalized. The normalization formula is shown in formula (1):

[0130]

[0131] Wherein, x is the original data, namely, the second training data, and x′ is the normalized data, namely, the third training data.

[0132] 4) According to the historical operation data and target performance indicators, the labels of combustion efficiency, pollutant emission concentration and reheater steam temperature deviation are marked to obtain the labeled pre-processed boiler operation parameter data, i.e., the target training data, to provide a supervision signal for the multi-objective prediction model.

[0133] In one implementable manner of the disclosed embodiment, when constructing a combustion optimization strategy, it can also be implemented in the following manner but is not limited to: inputting the training target data into the trained multi-objective prediction model for data prediction processing to obtain a training multi-objective prediction result; determining the state space based on the training multi-objective prediction result and the operating condition information of the system to be optimized; determining the action space based on a preset oxygen adjustment range and a preset air valve opening adjustment range; performing data construction processing based on the state space, the action space and a preset reward function to obtain the constructed combustion optimization strategy, wherein the preset reward function is used to balance the multi-objective prediction results, and the multi-objective prediction results include at least combustion efficiency, pollutant emission concentration and reheater steam temperature deviation.

[0134] In the disclosed embodiment, based on the trained multi-objective prediction model, a multi-agent distributed reinforcement learning (MADRL) algorithm is used to construct a combustion optimization strategy, and a trained combustion optimization model that can dynamically adjust the oxygen content and SOFA air damper opening is obtained through training.

[0135] The combustion optimization strategy is built based on reinforcement learning. Reinforcement learning adopts a multi-agent distributed training architecture. Each agent is responsible for the optimization training of different load intervals, and finally forms a global optimization strategy through parameter sharing. Specifically, the construction process of the combustion optimization strategy includes but is not limited to:

[0136] 1) State space: including the real-time load of the system to be optimized, the combustion efficiency, NOx emission concentration and reheater steam temperature deviation output by the multi-objective prediction model, i.e., the multi-objective prediction results for training;

[0137] 2) Action space: including the adjustment range of oxygen volume and SOFA air door opening;

[0138] 3) Reward function: A reward function is set to balance the optimization objectives of combustion efficiency, NOx emission concentration and reheater steam temperature deviation, that is, to balance the multi-objective prediction results. Specifically, the preset reward function can be expressed by formula (2):

[0139] R=ω1·Δη-ω2·ΔNO x -ω3·ΔT 偏差 Formula (2)

[0140] Among them, Δη is the combustion efficiency improvement rate, ΔNO x is the NOx emission increase rate, ΔT 偏差 is the reheater steam temperature deviation, ω1, ω2 and ω3 are weight coefficients, and the weight coefficients are user-defined weight coefficients.

[0141] In one possible implementation of the disclosed embodiment, the trained combustion optimization model may experience performance degradation during actual operation, and the operating conditions of the system to be optimized may also undergo significant changes (for example, changes in fuel type or aging of equipment). In this case, in order to maintain efficient and accurate combustion optimization processing, the model needs to be updated. Specifically, regarding model updating, the following methods may also be used, but are not limited to: real-time collection of actual performance indicators of the system to be optimized, and calculation of the deviation between the actual performance indicators and the multi-objective preset results; updating the trained multi-objective prediction model through an adaptive gradient descent algorithm based on the deviation value to obtain an updated multi-objective prediction model. The trained combustion optimization model is updated based on the updated multi-objective prediction model to obtain an updated combustion optimization model.

[0142] In the disclosed embodiment, the multi-objective prediction results of the trained multi-objective prediction model are compared in real time with the multi-objective performance indicators of the actual operation of the system to be optimized, that is, the actual performance indicators, and the trained multi-objective prediction model and the trained combustion optimization model are updated according to the deviation value to continuously improve the optimization performance, specifically including but not limited to:

[0143] 1) Compare the actual operating parameters of the boiler (actual performance indicators) collected in real time with the prediction results of the multi-objective prediction model (multi-objective preset results), and calculate the deviation;

[0144] 2) Update the trained multi-objective prediction model using the adaptive gradient descent algorithm according to the deviation;

[0145] 3) Use the updated multi-objective prediction model to update the trained combustion optimization model.

[0146] In the process of model updating, a method based on transfer learning is adopted to accelerate training and reduce computing resource consumption by migrating part of the weights of the existing model.

[0147] When the operating environment or fuel type of the system to be optimized changes, the existing trained combustion optimization model may no longer be applicable. Transfer learning can quickly adapt to the new environment:

[0148] For example, when switching from coal fuel to low-sulfur coal, biomass fuel, etc., common features can be extracted from old fuel data through transfer learning to quickly adapt to new fuel characteristics;

[0149] For example, the aging or updating of boiler components (such as SOFA damper modification) will lead to changes in operating characteristics. Transfer learning can reduce the need for retraining data and accelerate the reconstruction of the optimization model.

[0150] For example, when environmental protection policies lead to stricter pollutant emission limits, transfer learning can quickly update the model's target weights and output strategies.

[0151] Based on all the above embodiments, the present disclosure also provides a flow chart of another combustion optimization method based on machine learning, such as Figure 2 As shown, it can be explained by the following contents:

[0152] S1. Collect the operating parameters of the thermal power plant boiler and perform preprocessing;

[0153] S2. Establish a multi-objective prediction model based on the graph neural network, and use the pre-processed boiler operation parameters to train the multi-objective prediction model to obtain a trained multi-objective prediction model;

[0154] S3. Based on the trained multi-objective prediction model, a multi-agent distributed reinforcement learning algorithm is used to build a combustion optimization strategy, and a combustion optimization model that can dynamically adjust the oxygen content and SOFA air damper opening is obtained through training;

[0155] S4. Deploy the trained combustion optimization model to the boiler control system, and dynamically adjust the oxygen content and SOFA air damper opening through the combustion optimization model according to the pre-processed boiler operating parameters collected in real time;

[0156] S5. Compare the prediction results of the multi-objective prediction model with the multi-objective performance indicators of the actual operation of the boiler in real time, and retrain the multi-objective prediction model and combustion optimization model according to the deviation to continuously improve the optimization performance.

[0157] Furthermore, the boiler operating parameters are collected in real time through a distributed sensor network, and the collected data is preliminarily processed by edge computing devices before being uploaded to the central server.

[0158] Furthermore, the method of preprocessing the boiler operating parameters includes: using an outlier detection algorithm to eliminate abnormal data points; using a sliding average method to smooth the time series data; normalizing each parameter; and manually labeling the combustion efficiency, NOx emission concentration, and reheater steam temperature deviation labels based on historical operating data and target performance indicators.

[0159] Furthermore, the multi-objective prediction model includes a node feature embedding module, a graph convolution module based on an attention mechanism, and a fully connected layer output module; the node feature embedding module is used to perform high-dimensional feature embedding on the preprocessed boiler operating parameters; the graph convolution module based on the attention mechanism is used to model the relationship between the boiler operating parameters and extract multi-dimensional correlation features; the fully connected layer output module is used to output multi-objective prediction results, including the boiler's combustion efficiency, NOx emission concentration, and reheater steam temperature deviation.

[0160] Furthermore, the graph neural network dynamically constructs an adjacency matrix between boiler operating parameters.

[0161] Furthermore, the combustion optimization strategy is constructed based on reinforcement learning and includes a state space, an action space and a reward function, and the reward function is set to balance the optimization objectives of combustion efficiency, NOx emission concentration and reheater steam temperature deviation.

[0162] Furthermore, the reinforcement learning adopts a multi-agent distributed training architecture, where each agent is responsible for optimization training in different load intervals, and ultimately forms a global optimization strategy through parameter sharing.

[0163] Furthermore, the deployment of the combustion optimization model is achieved through edge computing devices, and the edge devices use a lightweight model structure to ensure low latency in the optimization process.

[0164] Furthermore, after obtaining the deviation between the prediction results of the multi-objective prediction model and the multi-objective performance indicators of the actual operation of the boiler, the multi-objective prediction model is retrained using an adaptive gradient descent algorithm according to the deviation, and then the combustion optimization model is retrained using the updated multi-objective prediction model.

[0165] Furthermore, the retraining process adopts a transfer learning-based method to accelerate training and reduce computing resource consumption by migrating part of the weights of the existing model.

[0166] In summary, the embodiments of the present disclosure can achieve the following technical effects:

[0167] 1. The embodiment of the present disclosure uses a combustion optimization model trained by a distributed reinforcement learning algorithm to achieve dynamic adjustment of the oxygen content and the damper opening according to the real-time operating data of the system to be optimized, so as to meet the demand for rapid changes in the operating state to be optimized. At the same time, it can also coordinate multi-objective optimization needs and respond to the demand for state changes in real time during the combustion optimization process. Through the multi-objective prediction model trained based on the graph neural network, it can fully predict the complex relationship between the operating parameters and performance indicators of the system to be optimized, obtain the multi-objective prediction results, and improve the prediction accuracy and the optimization effect of combustion optimization. Therefore, the embodiment of the present disclosure can respond to the demand for state changes in real time during the combustion optimization process, improve the optimization effect and reduce the complexity of operation.

[0168] 2. The disclosed embodiments significantly improve the quality and consistency of input data through outlier detection, data smoothing and normalization operations, and provide stable and reliable basic data support for model training.

[0169] 3. The disclosed embodiment uses a structural design based on a graph neural network to enable the multi-objective prediction model to capture the complex spatial correlation in the boiler operating parameters, further improving the prediction accuracy of the model.

[0170] 4. The disclosed embodiment has stronger convergence and adaptability through the multi-agent distributed reinforcement learning algorithm, and can quickly iterate and optimize the combustion strategy, thereby accelerating the construction and deployment of the model.

[0171] 5. The disclosed embodiment continuously retrains the multi-objective prediction model and the combustion optimization model through a real-time feedback mechanism, thereby improving the adaptability and long-term optimization effect of the model.

[0172] Corresponding to the above-mentioned combustion optimization method based on machine learning, the present invention also proposes a combustion optimization device based on machine learning. Since the device embodiment of the present invention corresponds to the above-mentioned method embodiment, the details not disclosed in the device embodiment can be referred to the above-mentioned method embodiment, and will not be repeated in the present invention.

[0173] Figure 3 A structural schematic diagram of a combustion optimization device based on machine learning provided in an embodiment of the present disclosure, such as Figure 3 As shown, including:

[0174] The prediction unit 31 is used to input the real-time operation data of the system to be optimized into the trained multi-objective prediction model for data prediction processing to obtain a multi-objective prediction result; wherein the trained multi-objective prediction model is a model established based on a graph neural network and used to predict the performance indicators of the system to be optimized;

[0175] The adjustment unit 32 is used to adjust the oxygen content and air door opening of the system to be optimized by using the trained combustion optimization model according to the multi-objective preset results and the real-time operation data, so as to obtain the system to be optimized after combustion optimization;

[0176] Among them, the trained combustion optimization model is constructed based on a distributed reinforcement learning algorithm, and is a model used to optimize the combustion of the system to be optimized by dynamically adjusting the oxygen content and air valve opening of the system to be optimized.

[0177] The machine learning-based combustion optimization device provided by the present disclosure inputs the real-time operation data of the system to be optimized into a trained multi-objective prediction model for data prediction processing to obtain a multi-objective prediction result; wherein the trained multi-objective prediction model is a model established based on a graph neural network, and is used to predict the performance indicators of the system to be optimized; according to the multi-objective preset results and the real-time operation data, the oxygen content and air valve opening of the system to be optimized are adjusted by the trained combustion optimization model to obtain the system to be optimized after combustion optimization; wherein the trained combustion optimization model is a model constructed based on a distributed reinforcement learning algorithm, and is used to dynamically adjust the oxygen content and air valve opening of the system to be optimized to perform combustion optimization on the system to be optimized. Compared with the related art, the embodiment of the present disclosure uses a combustion optimization model trained by a distributed reinforcement learning algorithm to achieve dynamic adjustment of the oxygen content and the damper opening according to the real-time operating data of the system to be optimized, so as to meet the demand for rapid changes in the operating state to be optimized. At the same time, it can also coordinate multi-objective optimization needs and respond to the needs of state changes in real time during the combustion optimization process. Through the multi-objective prediction model trained based on the graph neural network, it can fully predict the complex relationship between the operating parameters and performance indicators of the system to be optimized, obtain the multi-objective prediction results, and improve the prediction accuracy and the optimization effect of combustion optimization. Therefore, the embodiment of the present disclosure can respond to the needs of state changes in real time during the combustion optimization process, improve the optimization effect and reduce the complexity of operation.

[0178] Furthermore, in a possible implementation of the embodiment of the present disclosure, as Figure 4 As shown, the prediction unit 31 includes:

[0179] The embedding module 311 is used to input the target operation data into the node feature embedding module for high-dimensional feature embedding to obtain the first feature data corresponding to the target operation data; the trained multi-target prediction model includes the node feature embedding module, the graph convolution module based on the attention mechanism, and the fully connected layer output module;

[0180] A modeling module 312, configured to input the first feature data into the graph convolution module based on the attention mechanism to perform data relationship modeling processing to obtain multi-dimensional associated feature data;

[0181] The processing module 313 is used to input the multi-dimensional correlation feature data into the fully connected layer output module for full connection processing to obtain the multi-target prediction result.

[0182] Furthermore, in a possible implementation of the embodiment of the present disclosure, as Figure 4 As shown, the adjustment unit 32 includes:

[0183] An evaluation module 321 is used to evaluate the fitness of each first optimization model to the current operating condition of the system to be optimized by a preset evaluation algorithm to obtain a target optimization model; wherein the trained combustion optimization model includes a plurality of the first optimization models, and the target optimization model is the model with the highest fitness to the current operating condition of the system to be optimized among the trained combustion optimization models;

[0184] An optimization module 322, configured to perform combustion optimization processing through the target optimization model according to the multi-objective preset results and the target operation data, and obtain target optimization parameters;

[0185] The adjustment module 323 is used to adjust the oxygen content and air door opening of the system to be optimized by using the target optimization parameters to obtain the system to be optimized after the combustion is optimized.

[0186] Furthermore, in a possible implementation of the embodiment of the present disclosure, the optimization module 322 is further configured to:

[0187] Acquiring operating condition information and preset constraint information of the system to be optimized;

[0188] When the operating condition information and / or the constraint information determine that the system to be optimized meets environmental protection requirements, based on the first preset weight information, the combustion optimization process is performed through the target optimization model according to the multi-objective preset results and the target operating data to obtain a first optimization parameter;

[0189] Wherein, the first optimization parameter is at least used to reduce the combustion efficiency of the system to be optimized and restrict pollutant emissions;

[0190] When the operating condition information and / or the constraint information determine that the system to be optimized is under high load demand, based on the second preset weight information, the combustion optimization process is performed by the target optimization model according to the multi-objective preset result and the target operating data to obtain a second optimization parameter;

[0191] The second optimization parameter is at least used to improve the combustion efficiency of the system to be optimized and reduce the constraints on pollutant emissions, and the target optimization parameter includes at least the first optimization parameter or the second optimization parameter.

[0192] Furthermore, in a possible implementation of the embodiment of the present disclosure, as Figure 4 As shown, the device also includes:

[0193] An acquisition unit 33 is used to acquire the real-time operation data of the system to be optimized;

[0194] The processing unit 34 is used to perform data preprocessing on the real-time operation data to obtain target data for training;

[0195] A first training unit 35 is used to establish an initial multi-objective prediction model based on the graph neural network, and perform model training on the initial multi-objective prediction model through the training target data to obtain the trained multi-objective prediction model;

[0196] A construction unit 36, configured to construct a combustion optimization strategy based on the trained multi-objective prediction model by using the distributed reinforcement learning algorithm;

[0197] The second training unit 37 is used to perform model training processing on multiple initial combustion optimization models according to the combustion optimization strategy to obtain the trained combustion optimization model.

[0198] Furthermore, in a possible implementation of the embodiment of the present disclosure, as Figure 4 As shown, the processing unit 34 includes:

[0199] A removal module 341 is used to remove abnormal data points from the real-time operation data using an abnormal value detection algorithm to obtain first training data;

[0200] A smoothing module 342 is used to smooth the time series data of the first training data using a sliding average method to obtain second training data;

[0201] A processing module 343 is used to perform normalization processing on the second training data to obtain third training data;

[0202] The labeling module 344 is used to perform data labeling processing on the third training data according to the historical operation data and the target performance indicator to obtain the target training data, wherein the historical operation data is the operation data of the system to be optimized at a time before the real-time operation data, and the target performance indicator is the expected performance indicator that the system to be optimized needs to achieve.

[0203] Furthermore, in a possible implementation of the embodiment of the present disclosure, as Figure 4 As shown, the construction unit 36 ​​includes:

[0204] Prediction module 361, used for inputting the training target data into the trained multi-objective prediction model to perform data prediction processing and obtain the training multi-objective prediction result;

[0205] A determination module 362, configured to determine the state space according to the training multi-objective prediction result and the operating condition information of the system to be optimized;

[0206] The determination module 362 is further used to determine the action space according to the preset oxygen amount adjustment range and the preset air door opening adjustment range;

[0207] The construction module 363 is used to perform data construction processing according to the state space, the action space and the preset reward function to obtain the combustion optimization strategy, wherein the preset reward function is used to balance the multi-objective prediction results, and the multi-objective prediction results at least include combustion efficiency, pollutant emission concentration and reheater steam temperature deviation.

[0208] Furthermore, in a possible implementation of the embodiment of the present disclosure, as Figure 4 As shown, the device also includes:

[0209] A collection unit 38, used for collecting the actual performance index of the system to be optimized in real time, and calculating the deviation value between the actual performance index and the multi-objective preset result;

[0210] The updating unit 39 is used to update the trained multi-objective prediction model through an adaptive gradient descent algorithm according to the deviation value to obtain an updated multi-objective prediction model.

[0211] The updating unit 39 is further used to update the trained combustion optimization model based on the updated multi-objective prediction model to obtain an updated combustion optimization model.

[0212] It should be noted that the above explanation of the method embodiment is also applicable to the device of the embodiment of the present disclosure, and the principle is the same, which is no longer limited in the embodiment of the present disclosure.

[0213] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0214] Figure 5 A schematic block diagram of an example electronic device 500 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0215] like Figure 5As shown, the device 500 includes a computing unit 501, which can perform various appropriate actions and processes according to a computer program stored in a ROM (Read-Only Memory) 502 or a computer program loaded from a storage unit 508 to a RAM (Random Access Memory) 503. In the RAM 503, various programs and data required for the operation of the device 500 can also be stored. The computing unit 501, the ROM 502, and the RAM 503 are connected to each other via a bus 504. An I / O (Input / Output) interface 505 is also connected to the bus 504.

[0216] A number of components in the device 500 are connected to the I / O interface 505, including: an input unit 506, such as a keyboard, a mouse, etc.; an output unit 507, such as various types of displays, speakers, etc.; a storage unit 508, such as a disk, an optical disk, etc.; and a communication unit 509, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 509 allows the device 500 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0217] The computing unit 501 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 501 include, but are not limited to, a CPU (Central Processing Unit), a GPU (Graphic Processing Units), various dedicated AI (Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, a DSP (Digital Signal Processor), and any appropriate processor, controller, microcontroller, etc. The computing unit 501 performs the various methods and processes described above, such as a combustion optimization method based on machine learning. For example, in some embodiments, the combustion optimization method based on machine learning may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 508. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 500 via the ROM 502 and / or the communication unit 509. When the computer program is loaded into the RAM 503 and executed by the computing unit 501, one or more steps of the method described above may be performed. Alternatively, in other embodiments, the computing unit 501 may be configured to execute the aforementioned machine learning-based combustion optimization method in any other appropriate manner (eg, by means of firmware).

[0218] Various embodiments of the systems and techniques described above herein may be implemented in digital electronic circuit systems, integrated circuit systems, FPGAs (Field Programmable Gate Arrays), ASICs (Application-Specific Integrated Circuits), ASSPs (Application Specific Standard Products), SOCs (System On Chips), CPLDs (Complex Programmable Logic Devices), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include: being implemented in one or more computer programs that may be executed and / or interpreted on a programmable system including at least one programmable processor that may be a special purpose or general purpose programmable processor that may receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0219] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0220] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any suitable combination of the foregoing. More specific examples of machine-readable storage media may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a RAM, a ROM, an EPROM (Electrically Programmable Read-Only-Memory) or a flash memory, an optical fiber, a CD-ROM (Compact Dis sc Read-Only Memory), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0221] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball), through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0222] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: LAN (Local Area Network), WAN (Wide Area Network), the Internet, and blockchain networks.

[0223] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or "VPS" for short). The server may also be a server of a distributed system, or a server combined with a blockchain.

[0224] It should be noted that artificial intelligence is a discipline that studies how computers can simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, planning, etc.), and includes both hardware-level and software-level technologies. Artificial intelligence hardware technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, and big data processing; artificial intelligence software technologies mainly include computer vision technology, speech recognition technology, natural language processing technology, as well as machine learning / deep learning, big data processing technology, knowledge graph technology, and other major directions.

[0225] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0226] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A combustion optimization method based on machine learning, characterized in that: include: Input the real-time operation data of the system to be optimized into the trained multi-objective prediction model for data prediction processing to obtain a multi-objective prediction result; wherein the trained multi-objective prediction model is a model established based on a graph neural network and used to predict the performance indicators of the system to be optimized; According to the multi-objective preset results and the real-time operation data, the oxygen content and air door opening of the system to be optimized are adjusted by the trained combustion optimization model to obtain the system to be optimized after combustion optimization; Among them, the trained combustion optimization model is constructed based on a distributed reinforcement learning algorithm, and is a model used to optimize the combustion of the system to be optimized by dynamically adjusting the oxygen content and air valve opening of the system to be optimized.

2. The method according to claim 1, characterized in that The target operation data is input into the trained multi-objective prediction model for data prediction processing to obtain the multi-objective prediction results, which includes: The target operation data is input into a node feature embedding module for high-dimensional feature embedding to obtain first feature data corresponding to the target operation data; the trained multi-target prediction model includes the node feature embedding module, a graph convolution module based on an attention mechanism, and a fully connected layer output module; Inputting the first feature data into the graph convolution module based on the attention mechanism to perform data relationship modeling processing to obtain multi-dimensional correlation feature data; The multi-dimensional correlation feature data is input into the fully connected layer output module for full connection processing to obtain the multi-target prediction result.

3. The method according to claim 1, characterized in that According to the multi-objective preset results and the target operation data, the oxygen content and air door opening of the system to be optimized are adjusted by the trained combustion optimization model to obtain the system to be optimized after combustion optimization, which includes: The fitness of each first optimization model to the current working condition of the system to be optimized is evaluated by a preset evaluation algorithm to obtain a target optimization model; wherein the trained combustion optimization model includes a plurality of the first optimization models, and the target optimization model is the model with the highest fitness to the current working condition of the system to be optimized among the trained combustion optimization models; According to the multi-objective preset results and the target operating data, combustion optimization processing is performed through the target optimization model to obtain target optimization parameters; The oxygen content and air door opening of the system to be optimized are adjusted by the target optimization parameters to obtain the system to be optimized after the combustion optimization.

4. The method according to claim 3, characterized in that The combustion optimization process is performed by the target optimization model according to the multi-objective preset results and the target operation data to obtain the target optimization parameters, including: Acquiring operating condition information and preset constraint information of the system to be optimized; When the operating condition information and / or the constraint information determine that the system to be optimized meets environmental protection requirements, based on the first preset weight information, the combustion optimization process is performed through the target optimization model according to the multi-objective preset results and the target operating data to obtain a first optimization parameter; Wherein, the first optimization parameter is at least used to reduce the combustion efficiency of the system to be optimized and restrict pollutant emissions; When the operating condition information and / or the constraint information determine that the system to be optimized is under high load demand, based on the second preset weight information, the combustion optimization process is performed by the target optimization model according to the multi-objective preset result and the target operating data to obtain a second optimization parameter; The second optimization parameter is at least used to improve the combustion efficiency of the system to be optimized and reduce the constraints on pollutant emissions, and the target optimization parameter includes at least the first optimization parameter or the second optimization parameter.

5. The method according to claim 1, characterized in that Before obtaining the operating data of the system to be optimized, the method further includes: Acquiring the real-time operation data of the system to be optimized, and performing data preprocessing on the real-time operation data to obtain target data for training; Establishing an initial multi-objective prediction model based on the graph neural network, performing model training processing on the initial multi-objective prediction model through the training target data, and obtaining the trained multi-objective prediction model; Based on the trained multi-objective prediction model, a combustion optimization strategy is constructed through the distributed reinforcement learning algorithm, and model training processing is performed on multiple initial combustion optimization models according to the combustion optimization strategy to obtain the trained combustion optimization model.

6. The method according to claim 5, characterized in that The performing data preprocessing on the real-time operation data to obtain target data for training includes: Using an outlier detection algorithm to remove abnormal data points from the real-time operation data to obtain first training data; Smoothing the time series data of the first training data using a sliding average method to obtain second training data; Normalizing the second training data to obtain third training data; The third training data is labeled according to historical operating data and target performance indicators to obtain the target training data, wherein the historical operating data is the operating data of the system to be optimized at a time before the real-time operating data, and the target performance indicator is the expected performance indicator that the system to be optimized needs to achieve.

7. The method according to claim 5, characterized in that The method of constructing a combustion optimization strategy based on the trained multi-objective prediction model through the distributed reinforcement learning algorithm includes: Inputting the training target data into the trained multi-objective prediction model to perform data prediction processing to obtain the training multi-objective prediction result; Determine the state space according to the multi-objective prediction result for training and the operating condition information of the system to be optimized; The action space is determined according to a preset oxygen quantity adjustment range and a preset air door opening adjustment range; Data construction and processing are performed according to the state space, the action space and a preset reward function to obtain the combustion optimization strategy, wherein the preset reward function is used to balance the multi-objective prediction results, and the multi-objective prediction results include at least combustion efficiency, pollutant emission concentration and reheater steam temperature deviation.

8. The method according to claim 1, characterized in that After the system to be optimized is adjusted in oxygen content and air door opening according to the multi-objective preset results and the real-time operation data by using the trained combustion optimization model to obtain the system to be optimized after combustion optimization, the method further includes: Collecting the actual performance index of the system to be optimized in real time, and calculating the deviation value between the actual performance index and the multi-objective preset result; The trained multi-objective prediction model is updated by an adaptive gradient descent algorithm according to the deviation value to obtain an updated multi-objective prediction model. The trained combustion optimization model is updated based on the updated multi-objective prediction model to obtain an updated combustion optimization model.

9. A combustion optimization device based on machine learning, characterized in that: include: A prediction unit, used to input the real-time operation data of the system to be optimized into a trained multi-objective prediction model for data prediction processing to obtain a multi-objective prediction result; wherein the trained multi-objective prediction model is a model established based on a graph neural network and used to predict the performance indicators of the system to be optimized; An adjustment unit, configured to adjust the oxygen content and air door opening of the system to be optimized by using a trained combustion optimization model according to the multi-objective preset results and the real-time operation data, so as to obtain the system to be optimized after combustion optimization; Among them, the trained combustion optimization model is constructed based on a distributed reinforcement learning algorithm, and is a model used to optimize the combustion of the system to be optimized by dynamically adjusting the oxygen content and air valve opening of the system to be optimized.

10. An electronic device, characterized in that: include: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 8.

Images