Data prediction method and device, equipment and model training method
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HEBEI THINKING TECH CO LTD
- Filing Date
- 2021-06-25
- Publication Date
- 2026-06-02
Smart Images
Figure CN115527078B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of energy technology, and more specifically, to a data prediction method, apparatus, device, and model training method. Background Technology
[0002] Currently, fossil fuels still dominate energy consumption. However, with the large-scale grid connection of new energy power, thermal power units are increasingly participating in peak shaving and frequency regulation, causing fluctuations in the operating load of thermal power units and posing challenges to the prediction of thermal variables in thermal power units.
[0003] Taking coal-fired power plants as an example, optimizing coal-fired power units typically requires predicting nitrogen oxide (NOx) emissions and boiler outlet flue gas temperature (BOT). To meet emission standards, it's necessary to predict NOx emissions and BOT temperatures in the flue gas. Based on these predictions, the boiler combustion and desulfurization systems can be optimized to improve unit energy efficiency and reduce NOx emissions. However, in related technologies, data lag or instability often make it difficult to determine the real-time operating status of these systems, affecting the accuracy of the predicted thermal variables.
[0004] Therefore, there is an urgent need to propose a thermal variable prediction scheme applicable to variable load conditions, in order to more accurately predict NOx emissions and boiler outlet flue gas temperature under variable load conditions. Summary of the Invention
[0005] This invention provides a data prediction method, apparatus, device, and model training method for accurately predicting thermal variable data. For example, predicting pollutant emissions and boiler outlet flue gas temperature in thermal power units.
[0006] In a first aspect, embodiments of the present invention provide a data prediction method, the method comprising:
[0007] Acquire equipment operating condition data and flame image data corresponding to each operating point;
[0008] A fusion feature set is generated based on equipment operating condition data and flame image data;
[0009] The fused feature set is input into the thermal variable prediction model to obtain the target thermal variable data.
[0010] Secondly, embodiments of the present invention provide a data prediction apparatus, comprising:
[0011] The data acquisition module is used to acquire equipment operating condition data and flame image data corresponding to each operating point;
[0012] The feature fusion module is used to generate a fused feature set based on the equipment operating data and the flame image data;
[0013] The prediction module is used to input the fused feature set into the thermal variable prediction model to obtain the target thermal variable data.
[0014] Thirdly, embodiments of the present invention provide an electronic device, including a processor and a memory, wherein executable code is stored in the memory, and when the executable code is executed by the processor, the processor can at least implement the data prediction method in the first aspect.
[0015] Fourthly, embodiments of the present invention provide a non-transitory machine-readable storage medium storing executable code, which, when executed by a processor of an electronic device, enables the processor to at least implement the data prediction method of the first aspect.
[0016] Fifthly, embodiments of the present invention provide a model training method, the method comprising:
[0017] Acquire equipment operating condition data and flame image data corresponding to each operating point;
[0018] A fusion feature set is generated based on the equipment operating data and the flame image data;
[0019] The fused feature set is input into the initial thermal variable prediction model, and the initial thermal variable prediction model is iteratively trained to obtain a thermal variable prediction model for predicting target thermal variable data.
[0020] Sixthly, embodiments of the present invention provide a model training apparatus, the apparatus comprising:
[0021] The data acquisition module is used to acquire equipment operating condition data and flame image data corresponding to each operating point;
[0022] The feature fusion module is used to generate a fused feature set based on the equipment operating data and the flame image data;
[0023] The training module is used to input the fused feature set into the initial thermal variable prediction model, and to iteratively train the initial thermal variable prediction model to obtain a thermal variable prediction model for predicting target thermal variable data.
[0024] In a seventh aspect, embodiments of the present invention provide an electronic device, including a processor and a memory, wherein executable code is stored in the memory, and when the executable code is executed by the processor, the processor can at least implement the model training method in the fifth aspect.
[0025] Eighthly, embodiments of the present invention provide a non-transitory machine-readable storage medium storing executable code, which, when executed by a processor of an electronic device, enables the processor to at least implement the model training method of the fifth aspect.
[0026] Ninthly, embodiments of the present invention provide a data prediction method, the method comprising:
[0027] Acquire equipment operating condition data and flame image data corresponding to each operating point;
[0028] A fusion feature set is generated based on the equipment operating data and the flame image data;
[0029] The fused feature set is input into the initial thermal variable prediction model, and the initial thermal variable prediction model is iteratively trained to obtain the thermal variable prediction model.
[0030] The aforementioned thermal variable prediction model is used to predict the target thermal variable data.
[0031] In a tenth aspect, embodiments of the present invention provide an electronic device, including a processor and a memory, wherein the memory stores executable code that, when executed by the processor, enables the processor to at least implement the data prediction method of the ninth aspect.
[0032] In one aspect, embodiments of the present invention provide a non-transitory machine-readable storage medium storing executable code, which, when executed by a processor of an electronic device, enables the processor to at least implement the data prediction method of the ninth aspect.
[0033] In a twelfth aspect, embodiments of the present invention provide a measuring device, such as a two-phase flow mass flow measuring device, which is used to detect the two-phase flow mass flow in a primary air duct. The data detected by the device (such as the mass flow data of the two-phase flow and the pulverized coal in the air conveying duct) is applied in the modeling process of a thermal variable prediction model to improve the modeling accuracy of the thermal variable prediction model.
[0034] In the technical solution provided by this invention, equipment operating condition data and flame image data corresponding to each operating point are first acquired. Equipment operating condition data reflects the changes in multiple monitoring indicators during equipment operation, while flame image data further reflects the actual operating conditions of the equipment at each operating point. Therefore, by generating a fusion feature set based on equipment operating condition data and flame image data, the latency problem of equipment operating condition data can be improved, the instability of data acquisition in variable load environments can be mitigated, and the robustness and stability of the model can be enhanced. This allows the thermal variable prediction model to more accurately predict target thermal variable data based on the fusion feature set, improving the prediction accuracy of thermal variable data (such as NOx emissions and boiler outlet flue gas temperature) under variable load conditions, optimizing the unit energy utilization rate under peak shaving and frequency regulation of thermal power units, adjusting boiler outlet flue gas temperature, and reducing pollutant emissions. Attached Figure Description
[0035] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent from the following detailed description taken in conjunction with the accompanying drawings. Several embodiments of the invention are illustrated in the drawings by way of example and not limitation, wherein:
[0036] Figure 1 A flowchart illustrating a data prediction method provided in an embodiment of the present invention;
[0037] Figure 2 A schematic diagram illustrating the principle of a thermal variable prediction model provided in an embodiment of the present invention;
[0038] Figure 3 This is a schematic diagram of the structure of a BiLSTM model provided in an embodiment of the present invention;
[0039] Figure 4 This is a schematic diagram of the structure of a data prediction device provided in an embodiment of the present invention;
[0040] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention;
[0041] Figure 6 A schematic flowchart of a model training method provided in an embodiment of the present invention;
[0042] Figure 7 This is a schematic diagram of the structure of a model training device provided in an embodiment of the present invention;
[0043] Figure 8 This is a schematic diagram of the structure of another electronic device provided in an embodiment of the present invention;
[0044] Figure 9 A flowchart illustrating another data prediction method provided in an embodiment of the present invention;
[0045] Figure 10 This is a schematic diagram of another data prediction device provided in an embodiment of the present invention;
[0046] Figure 11 This is a schematic diagram of another electronic device provided in an embodiment of the present invention.
[0047] In the accompanying drawings, the same or corresponding reference numerals indicate the same or corresponding parts. Detailed Implementation
[0048] The principles and spirit of the invention will now be described with reference to several exemplary embodiments. It should be understood that these embodiments are provided merely to enable those skilled in the art to better understand and implement the invention, and are not intended to limit the scope of the invention in any way. Rather, these embodiments are provided to make the invention more thorough and complete, and to fully convey the scope of the invention to those skilled in the art.
[0049] Those skilled in the art will understand that embodiments of the present invention can be implemented as a system, apparatus, device, method, or computer program product. Therefore, the present invention can be specifically implemented in the following forms: entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.
[0050] According to embodiments of the present invention, a data prediction method, apparatus, device, and model training method are proposed. Furthermore, the number of any elements in the accompanying drawings is for illustrative purposes only and not as a limitation, and any naming is for distinguishing purposes only and has no limiting meaning.
[0051] The principles and spirit of the present invention will be explained in detail below with reference to several representative embodiments.
[0052] The inventors discovered that with the large-scale grid connection of new energy power, thermal power units are increasingly involved in peak shaving and frequency regulation, causing fluctuations in the operating load of thermal power units. This not only increases operating costs but also poses challenges to the prediction of thermal variables in thermal power units.
[0053] For coal-fired power units, the main pollutant in the flue gas emissions is NOx. To meet emission standards, it is necessary to predict NOx emissions in the flue gas so that the unit's energy utilization rate can be optimized based on the prediction results, thereby reducing NOx emissions. In addition to NOx emissions, boiler outlet flue gas temperature is also an important data indicator that needs to be predicted when optimizing the unit. However, in related technologies, due to data lag or unstable data acquisition, it is often difficult to determine the real-time operating status of the above systems, affecting the accuracy of the predicted thermal variable data.
[0054] In summary, there is an urgent need to propose a thermal variable prediction scheme applicable to variable load conditions, in order to more accurately predict NOx emission data and boiler outlet flue gas temperature under variable load conditions.
[0055] To overcome at least one technical problem existing in the prior art, this invention proposes a data prediction method, apparatus, device, and model training method. The data prediction method includes at least: acquiring equipment operating condition data and flame image data corresponding to each operating point; thereby generating a fused feature set based on the equipment operating condition data and flame image data; and finally, inputting the fused feature set incorporating flame image data into a thermal variable prediction model to obtain the target thermal variable data.
[0056] In the aforementioned data prediction method, flame image data is incorporated into the fused feature set. This allows the actual operating conditions of the equipment at various operating points, reflected in the flame image data, to be applied to the prediction process of the thermal variable prediction model. This improves the time delay problem of equipment operating condition data, mitigates the instability of data acquisition in variable load environments, and enhances the robustness and stability of the model. Ultimately, this enables the thermal variable prediction model to more accurately predict target thermal variable data based on the fused feature set, improves the prediction accuracy of thermal variable data under variable load conditions, optimizes the energy utilization rate of thermal power units under peak shaving and frequency regulation, adjusts boiler outlet flue gas temperature, and reduces pollutant emissions.
[0057] It is understandable that the principles of model training methods, devices, media, and equipment are similar to those of data prediction methods, and will not be elaborated here.
[0058] After introducing the basic principles of the present invention, various non-limiting embodiments of the present invention will be described in detail below.
[0059] The technical solution provided in this invention can be executed by an electronic device, which can be a terminal device such as a PC or laptop, or a server. The server can be a physical server containing an independent host, a virtual server hosted by a host cluster, or a cloud server.
[0060] The technical solutions provided by the embodiments of this invention can be applied to various scenarios involving the processing of thermal variable data, especially scenarios involving the prediction of thermal variable data. For example, scenarios involving the prediction of one or more thermal variable data (such as NOx emissions and boiler outlet flue gas temperature) under varying load conditions. For example, scenarios involving the prediction of pollutant emissions in flue gas from coal-fired units, or the prediction of oxygen content in flue gas from thermal power units. For example, scenarios involving the prediction of boiler outlet flue gas temperature in thermal power units.
[0061] The following describes a technical solution for predicting thermal variable data according to an exemplary embodiment of the present invention, with reference to the accompanying drawings and application scenarios. It should be noted that the above application scenarios are shown only to facilitate understanding of the spirit and principles of the present invention, and the embodiments of the present invention are not limited in any way. Rather, the embodiments of the present invention can be applied to any applicable scenario.
[0062] This invention provides a data prediction method, such as... Figure 1 As shown, this data prediction method is applied to the prediction of target thermal variable data, and the data prediction method includes at least the following steps:
[0063] 101. Obtain equipment operating condition data and flame image data corresponding to each operating point;
[0064] 102. Generate a fused feature set based on equipment operating condition data and flame image data;
[0065] 103. Input the fused feature set into the thermal variable prediction model to obtain the target thermal variable data.
[0066] Figure 1 In the data prediction method shown, equipment condition data reflects the equipment's operating status, such as data corresponding to multiple monitoring indicators during equipment operation. Flame image data further reflects the actual operating status of the equipment at various operating points; for example, flame image data can reflect the combustion status of the flame in a boiler. Optionally, flame image data is collected from the field before step 101.
[0067] Taking thermal power units as an example, assuming the thermal variable data is pollutant emission data in flue gas, the equipment operating condition data includes at least one thermal process data related to the pollutant emission data. For the target thermal variable data to be predicted, equipment operating condition data and flame image data corresponding to each operating point can be collected first, thereby generating a fused feature set based on the equipment operating condition data and flame image data. By introducing flame image data into the fused feature set, the actual operating conditions of the equipment at each operating point reflected by the flame image data can be applied to the prediction process of the thermal variable prediction model. This facilitates the improvement of the time delay problem existing in equipment operating condition data, mitigates the instability of data acquisition in variable load environments, and enhances the robustness and stability of the model. Ultimately, this enables the thermal variable prediction model to more accurately predict the target thermal variable data based on the fused feature set, improves the prediction accuracy of thermal variable data under variable load conditions, optimizes the unit energy utilization rate under peak shaving and frequency regulation of thermal power units, and reduces pollutant emissions. Similarly, the above method can be used to more accurately predict boiler outlet flue gas temperature based on the fused feature set, so as to optimize the unit energy utilization rate under peak shaving and frequency regulation of thermal power units and adjust the boiler outlet flue gas temperature under variable load conditions.
[0068] The thermal variable prediction model provided in this invention is applicable to various operating conditions, especially variable load conditions. Taking thermal power units as an example, due to the influence of peak shaving and frequency regulation, thermal power units are under variable load conditions, causing frequent fluctuations in thermal variable data. Optionally, the operating point can be a point in time or a period of time for collecting equipment operating data.
[0069] The network structure design of the thermal variable prediction model affects the prediction accuracy of the thermal variable. Therefore, when designing the network structure of the thermal variable prediction model, it is often necessary to consider the data type and scale of the input data of the model, and select a neural network and model depth that are suitable for the current application scenario.
[0070] In practical applications, the network structure of a thermal variable prediction model can be a single neural network or a hybrid neural network. It's understandable that a hybrid neural network typically combines multiple neural networks in a predetermined manner, allowing different types of neural networks to leverage their respective strengths to improve the overall performance of the model. The combination of multiple neural networks can be done in series, parallel, or other connection methods, which are not limited here. For example, multiple neural networks can be sequentially superimposed as a series connection.
[0071] Hybrid neural networks can be constructed in parallel. For example, different branches of a model can be formed by multiple neural networks, and the outputs of these branches can be merged to create a parallel hybrid neural network.
[0072] In one optional embodiment, the thermal variable prediction model is constructed by combining a partial least squares method (PLS) model, a deep convolutional neural network (DCNN) model, and a bidirectional long short-term memory (BiLSTM) model to adapt to the analysis scenario of multi-source data related to the prediction of thermal variable data.
[0073] In practical applications, the network structure of the thermal variable prediction model constructed by combining the above three models is as follows: Figure 2 As shown. In Figure 2 In this model, the PLS model and the DCNN model are connected in parallel to extract and mix features from multi-source input data, thereby obtaining a fusion result based on the multi-source input data. This fusion result is then input into the BiLSTM model, enabling the BiLSTM model to predict the target thermal variable data based on this fusion result. Figure 2In this context, the multi-source input data includes equipment operating condition data a, equipment operating condition data b, and flame image data. For example, assuming that equipment operating condition data a is historical thermal variable data, then equipment operating condition data b can be various thermal process data related to the historical thermal variable data.
[0074] In this embodiment of the invention, the multi-source data related to the prediction of thermal variables includes, but is not limited to: equipment operating condition data and flame image data. The equipment operating condition data is numerical data, while the flame image data is image data. Similar to the above description, equipment operating condition data reflects the operating status of the equipment, while flame image data further reflects the actual operating status of the equipment at various operating points.
[0075] First, let's introduce the equipment operating data involved in the embodiments of this invention.
[0076] Equipment operating condition data reflects the operational status of equipment, such as data corresponding to multiple monitoring indicators during equipment operation. Optionally, equipment operating condition data includes thermal variable data and related thermal process data. The thermal variable data includes historical thermal variable data.
[0077] For example, assuming the equipment is a thermal power unit, thermal process data affecting pollutant emissions (such as NOx) can be selected from the equipment operating data according to relevant mechanisms. For instance, the unit load (L... oad Total coal quantity (F) a Total boiler air volume (M) a Furnace pressure (P) a ), water-to-coal ratio (W c ), flue gas oxygen content (Y) O2 Total water supply (W) a ), coal-to-air ratio (W) o ), Hot primary air pressure main pipe pressure (P) r ), Total temperature of primary hot air (T) w Total secondary air volume of boiler (W) e ), the mass flow rate of pulverized coal in the primary air ducts of each of the four pulverizing systems (C a C b C c C d Instantaneous coal feed rate (C) of the coal feeder co ), primary air volume of coal mill (W) y ), the rotational speed of the four coal mills (R) a R b R c R d ), main steam temperature (T a ), main steam pressure (P) c Secondary air temperature on both sides (T)oa T ob ), flue gas temperature (T) p ), 6 secondary air damper openings (S) a S b S c S d S e S f ), 4-layer burnout wind baffle opening (S) oa S ob S oc S od The aforementioned 36-dimensional thermal process data can be collected by sensors or other monitoring devices and systems mounted on the equipment. For example, the thermal process data can be obtained from a distributed control system (DCS).
[0078] It is worth noting that the aforementioned 36-dimensional thermal process data also affect the boiler outlet flue gas temperature. Therefore, this thermal process data can also be used in the prediction of boiler outlet flue gas temperature.
[0079] In fact, the monitoring dimensions of the aforementioned thermal process data can be set according to the actual application scenario, and are not limited to the numbers given in the examples above. For example, the number of monitoring parameters such as coal mill speed, secondary air damper opening, and burnout air damper opening can be set according to the actual equipment configuration or monitoring needs.
[0080] It is understood that the target thermal variable data to be predicted in one application scenario can be emission data, such as pollutant emission data or boiler outlet flue gas temperature, or it can be one or a combination of the aforementioned thermal process data in another application scenario. This is not limited in the embodiments of the present invention. In practical applications, thermal variable data includes historical thermal variable data. In this document, the type of data to be predicted from the equipment operating condition data is referred to as thermal variable data; the collected data of this type is referred to as historical thermal variable data; and the data to be predicted using the technical solution provided by the present invention is referred to as target thermal variable data. In different scenarios, the thermal variable data to be predicted can be emission data from the equipment operating condition data, boiler outlet flue gas temperature from the equipment operating condition data, or one or a combination of the aforementioned thermal process data.
[0081] In practical applications, there is a time delay when collecting equipment operating condition data, and the data collection under variable load conditions is prone to instability. Both of these may cause the equipment operating condition data to fail to reflect the actual operating condition of the equipment at the current operating point, thereby affecting the accuracy of the model prediction results.
[0082] To address the aforementioned issues and ensure the accuracy of the model's predictions, in addition to the equipment operating condition data described above, step 101 also requires acquiring flame image data corresponding to each operating point. For example, in a thermal power unit, the flame image data could be furnace flame pattern data corresponding to each operating point. The aforementioned flame image data is acquired by image acquisition equipment or modules; the specific equipment or module type is determined by the actual application scenario, and this embodiment of the invention does not impose any limitations.
[0083] Optionally, the flame image data corresponding to each operating point can be a single image or a group of multiple images. For example, assuming the operating equipment is a thermal power unit, the flame image data corresponding to a certain operating point in the thermal power unit can be a single image representing the combustion status of the furnace flame, or it can be multiple images collected from different angles within the thermal power unit at the same time or during the same time period.
[0084] Optionally, the number of flame image data points corresponds to the number of operating points to reflect the equipment's operating status at each operating point. For example, the number of flame images may match the number of operating points, or the number of flame image groups may match the number of operating points.
[0085] After introducing the multi-source input data (i.e., equipment operating data and flame image data) obtained in step 101, step 102 shows that the corresponding fusion feature set can be generated based on the multi-source input data corresponding to each operating point.
[0086] Specifically, it is assumed that the equipment operating condition data includes historical thermal variable data, as well as various thermal process data related to the historical thermal variable data. Based on this, 102 can be implemented as follows:
[0087] Feature selection is performed on historical thermal variable data and various thermal process data to obtain a numerical variable feature set; feature selection is performed on flame image data to obtain an image variable feature set; the numerical variable feature set and the image variable feature set are fused to obtain a fused feature set.
[0088] Through the above steps, the multi-source input data corresponding to each operating point can be fused into a fused feature set after feature extraction. Thus, the fused feature set can more comprehensively reflect the various factors affecting the target thermal variables during equipment operation.
[0089] The following section uses specific examples to introduce feature extraction methods for different types of data in multi-source input data.
[0090] For flame image data, step 102, which involves feature selection of the flame image data corresponding to each working point to obtain the image-type variable feature set, can be implemented as follows:
[0091] Key image features are extracted from the flame image data corresponding to each working point; the key image features are flattened to obtain the flame image feature vector in the image variable feature set, wherein the flame image feature vector is a one-dimensional feature vector.
[0092] Specifically, the flame image data corresponding to each working point is an image-type variable. Therefore, a DCNN model can be used to select features from the flame image data. Optionally, the DCNN model consists of multiple convolutional layers and pooling layers. The convolutional layers are used to perform convolution operations on the flame image data, and the pooling layers are used to perform pooling operations on the flame image data. Through multiple convolutional and pooling layers, key image features can be extracted from the flame image data, and the dimensionality of these key image features can be reduced. The ReLU function can be used as the activation function. Since the flame image data is two-dimensional, an optional input format for the flame image data is (P, X, Y), where P is the number of convolutional and pooling layers, and X and Y represent the number of rows and columns of the two-dimensional data, respectively. Optionally, the number of convolutional and pooling layers can be set to 3, such as setting P to 3.
[0093] Furthermore, assuming the key image features after the above processing are a matrix, this matrix needs to be flattened to obtain a one-dimensional feature vector. This one-dimensional feature vector is the flame image feature vector corresponding to the current operating point. Finally, after the key image features corresponding to each operating point are flattened, the flame image feature vectors corresponding to each operating point are used to form an image-type variable feature set.
[0094] For example, one alternative embodiment can be implemented as follows: Assume the flame image data includes multiple flame images. Based on this, the multiple flame images are input into a convolutional neural network model (e.g., a DCNN model) to obtain an image feature matrix. The convolutional neural network model includes multiple convolutional layers, multiple pooling layers, a ReLU function, and fully connected layers. Specifically, for each flame image, the convolutional neural network model performs the following operations: convolving each input flame image with a convolutional kernel, and processing the convolution result using the ReLU function; inputting the ReLU-processed convolution result into a pooling layer to obtain an image feature matrix containing key image features. Furthermore, the image feature matrix is converted into a corresponding one-dimensional feature vector, and this one-dimensional feature vector is used as the flame image feature vector in the image-type variable feature set.
[0095] In addition to the DCNN model mentioned above, other networks or algorithms can also be used in 102 to perform feature selection on flame image data, such as the Visual Geometry Group Network (VGG) model and the AlexNet model.
[0096] For equipment operating condition data, the method for selecting thermal process data in this embodiment of the invention can be statistical method and machine learning method, including but not limited to: partial principal component analysis, kernel principal component analysis, partial least squares method (PLS), feature selection (Relief) algorithm, convolutional neural network or one or more of these.
[0097] Taking PLS as an example, the main principle of this algorithm is to maximize the correlation between the independent variable and the dependent variable when extracting the maximum explanatory variance of the independent variable data. Based on the above principle, this paper uses PLS to determine the contribution of various thermal process data to the thermal variable data, thereby selecting the thermal process data with a greater contribution for subsequent prediction of the target thermal variable data.
[0098] Based on the above principles, in an optional embodiment, in step 102, feature selection is performed on the historical thermal variable data and various thermal process data corresponding to each operating point to obtain a numerical variable feature set, which can be implemented as follows:
[0099] The principal components of various thermal process data are calculated using a PLS model. Based on the principal component calculation results and the principle of cross-validation, the number of principal components to be extracted is determined. The contribution of various thermal process data to historical thermal variable data is determined, where the greater the contribution, the greater the correlation with historical thermal variable data. Based on the number of principal components to be extracted, thermal process data whose contribution meets the preset conditions are selected from various thermal process data. The selected thermal process data and historical thermal processes are used as the feature set of numerical variables.
[0100] In the above steps, by assessing the contribution of various thermal process data to thermal variable data, we can effectively filter out thermal process data that are more relevant to historical thermal variable data. This effectively reduces the amount of thermal process data, simplifies the complexity of the thermal variable prediction model, and improves the prediction accuracy and generalization ability of the thermal variable prediction model.
[0101] Taking thermal power units as an example again, let's assume that the historical thermal variable data is the pollutant content in the flue gas. Let's assume that each set of equipment operating condition data includes the historical pollutant content (i.e., historical thermal variable data) and the thermal process data related to the historical pollutant content.
[0102] Specifically, in thermal power units, feature selection is performed on numerical variables (i.e., historical thermal variable data corresponding to each operating point and various thermal process data) based on PLS.
[0103] Here, it is assumed that the input variables to the PLS model are thermal process data collected from the thermal power unit, including but not limited to the following data: unit load (L oad Total coal quantity (F) a Total boiler air volume (M) a Furnace pressure (P) a ), water-to-coal ratio (W c ), flue gas oxygen content (Y) O2 Total water supply (W) a ), coal-to-air ratio (W) o ), Hot primary air pressure main pipe pressure (P) r ), Total temperature of primary hot air (T) w Total secondary air volume of boiler (W) e ), the mass flow rate of pulverized coal in the primary air ducts of each of the four pulverizing systems (C a C b C c C d Instantaneous coal feed rate (C) of the coal feeder co ), primary air volume of coal mill (W) y ), the rotational speed of the four coal mills (R) a R b R c R d ), main steam temperature (T a ), main steam pressure (P) c Secondary air temperature on both sides (T) oa T ob ), flue gas temperature (T) p ), 6 secondary air damper openings (S) a S b S c S d S e S f ), 4-layer burnout wind baffle opening (S) oa S ob S oc S od Assume the output variable of this PLS model is NOx emissions.
[0104] Based on the above assumptions, feature engineering can be performed on the training sample set constructed based on numerical variables (i.e., equipment operating condition data). Furthermore, the contribution of various thermal process data to historical thermal variable data can be analyzed using the aforementioned method for selecting thermal process data, so as to determine the final input / output variable set based on the degree of contribution. Where, x i ∈R M ×p y i ∈R M×qp represents the number of input variables (i.e., thermal process data), and q represents the number of output variables (i.e., thermal variable data).
[0105] Of course, the output variable of the PLS model can also be the boiler outlet flue gas temperature, such as the boiler outlet flue gas temperature prediction scenario. The specific implementation process is similar to the processing of NOx emissions, which will not be elaborated here.
[0106] Based on the PLS algorithm described above, the main steps of feature engineering are as follows:
[0107] The first step is to calculate the first pair of principal components, t1 and u1, to contain as much characteristic information as possible that can represent the variable set. The calculation expression is as follows:
[0108]
[0109]
[0110] In the formula, E0 and F0 are the standardized matrices of the variable sets, respectively.
[0111] The second step is to calculate. and The inner product of t1 and u1 is then used to calculate the covariance Cov(t1,u1). The calculation process is transformed into the following expression:
[0112]
[0113] The third step is to calculate the regression model for t1. The calculation process is shown below:
[0114]
[0115] In the formula, α1 and β1 are regression vectors, calculated using the following expression:
[0116]
[0117] The fourth step involves replacing E0 and F0 with E1 and F1, and repeating the first to third steps to calculate each principal component in turn, until the final number of principal components is obtained.
[0118] The final number of principal components to be extracted is determined based on the cross-validation principle. Specifically, the calculation formula for determining the final number of principal components to be extracted based on the cross-validation principle is as follows:
[0119]
[0120] In the formula, y i For actual input values, Let h be the predicted value for sample i after extracting h components. To remove the predicted value of sample point i for sample point i. The above conditions must be met. Characterizing principal component t h The marginal contribution to historical thermal variable data is significant.
[0121] Of course, in an alternative embodiment, the Variable Importance In Projection (VIP) index can be used to determine the contribution value (i.e., degree of contribution) of each of the above-mentioned thermal process data to NOx emissions.
[0122] Specifically, when The time-crossing validity is considered, determining the number of thermal process data points to be extracted; subsequently, the contribution of each thermal process data point to NOx emissions is calculated, i.e.:
[0123]
[0124] In Formula 7, p is the number of independent variables, m is the number of principal components; r(y; t) h ) for y and t h The correlation coefficient, y represents the equipment operating condition data in the thermal process data, and t h For historical thermal variable data, h is; w hk For the weight vector w h The kth component.
[0125] Optionally, to simplify the complexity of the thermal variable prediction model and improve its prediction accuracy, some thermal process data with VIP values less than a preset threshold can be removed, and the remaining thermal process data can be used as thermal process data for training the feature selection model for numerical variables.
[0126] In practical applications, in addition to the PLS model mentioned above, other algorithms or models can be used to perform feature selection on numerical variables (i.e., equipment operating data), such as partial principal component analysis, kernel principal component analysis, Relief algorithm, convolutional neural network model, etc., which will not be elaborated here.
[0127] The above feature extraction methods can obtain numerical and image-based variable feature sets from multi-source input data, enabling the full integration of multi-source input data to reflect the actual operating conditions of the equipment. In practical applications, in addition to the combination of DCNN and PLS models, other single or hybrid models can also be used to extract features from multi-source input data, which will not be elaborated here.
[0128] Furthermore, after obtaining the numerical variable feature set and the image variable feature set, step 102 involves fusing the numerical variable feature set and the image variable feature set to obtain a fused feature set. One method for obtaining the fused feature set can be specifically implemented as follows:
[0129] Normalize each feature vector in the numerical variable feature set and the image variable feature set; then concatenate the processed numerical variable feature vector with the flame image feature vector to obtain the fused feature vector in the fused feature set.
[0130] Specifically, taking the combination of DCNN and PLS models as an example, the feature vectors output by the two models are normalized and then concatenated to form a fused feature vector, which is used as the input to the thermal variable prediction model.
[0131] In one optional embodiment, the feature vectors in the numerical variable feature set and the image variable feature set are normalized using the following formula to obtain the feature vectors to be concatenated:
[0132]
[0133] Among them, y scaled Let y be the feature vector to be concatenated, and let y be any feature vector from the numerical variable feature set and the image variable feature set. min y represents the minimum value of either the numerical or graphical feature set. max It represents the maximum value of the feature set of numerical variables or the feature set of graphical variables.
[0134] In practical applications, the y-value of the feature set of numerical variables min and y max It can be determined from the data sample. The y-value of the image-type variable feature set. min It is usually set to 0, y max It is usually set to 255. R is the maximum value of the preset scaling range, and Q is the minimum value of the preset scaling range. The preset scaling range is generally set to between 0 and 1.
[0135] Through the above steps, numerical variable feature sets and image variable feature sets can be obtained from multi-source input data. Based on the fusion of the above two variable feature sets, a fused feature set containing multi-source feature information can be generated to improve the time delay problem of equipment operating condition data, alleviate the unstable data acquisition situation in variable load operating conditions, and enhance the robustness and stability of the model.
[0136] Finally, after obtaining the fused feature set, it can be input into the thermal variable prediction model to obtain the target thermal variable data.
[0137] The thermal variable prediction model is obtained through iterative training using a fused feature set. For example, the thermal variable prediction model is trained iteratively a predetermined number of times using a fused feature set obtained by fusing numerical variable feature samples and image variable feature samples. The predetermined number of times is, for example, 10 times.
[0138] Optionally, after completing the iterative training of the thermal variable prediction model, the model can be quantitatively evaluated to further optimize it. Quantitative evaluation methods include, but are not limited to, mean relative error and root mean square error.
[0139] In practical applications, the prediction model for thermal variables can be a single model or a hybrid model. Prediction models for thermal variables include, but are not limited to, one or a combination of the following: BiLSTM model, support vector machine, backpropagation neural network, boosting tree model (XGBoost), and fuzzy tree.
[0140] Taking the BiLSTM model as an example, which is a predictive model for thermal variables, the BiLSTM model has time series characteristics and has a good predictive effect on data collected from various operating points.
[0141] Specifically, the thermal variable prediction model includes forward long short-term memory (LSTM) units and backward long short-term memory (BSTM) units. The LSTM model, through a combination of input gates, forget gates, output gates, and storage memory units, allows time-series characteristic feature information to be better propagated backward, effectively solving the gradient vanishing and gradient exploding problems during long sequence training. In particular, the BiLSTM model can simultaneously record effective information from feature data in both forward and backward directions, exhibiting stronger correlation and overcoming the time delay inconsistency problem between various feature data collected at the same time, thus better remembering valuable information in time-series data.
[0142] The network structure of the BiLSTM model provided in this embodiment of the invention is as follows: Figure 3 As shown. In Figure 3 In BiLSTM, the model consists of: an input layer, an embedding layer, a forward Long Short-Term Memory (LSTM) layer, a backward Long Short-Term Memory (LSTM) layer, and an output layer. The fused feature vector {x1, x2, x3, ..., x...} is formed from the fused feature set. n The input BiLSTM model, after processing through the Embedding layer, forward LSTM layer, and backward LSTM layer, outputs the target thermal variable data {y1, y2, y3, ..., y4} predicted by the BiLSTM model. n}. Where, {h1, h2, h3, ..., h n} and {e(x1), e(x2), e(x3), ..., e(x)}n )} is an intermediate quantity.
[0143] Through the above steps, the thermal variable prediction model can more accurately predict the target thermal variable data based on the fused feature set, improve the prediction accuracy of thermal variable data under variable load conditions, optimize the energy utilization rate of thermal power units under peak shaving and frequency regulation, adjust the boiler outlet flue gas temperature, reduce pollutant emissions, and reduce operating costs.
[0144] Figure 1 The data prediction method shown first acquires equipment operating condition data and flame image data corresponding to each operating point. Then, by generating a fusion feature set based on the equipment operating condition data and flame image data, the time delay problem of equipment operating condition data can be improved, the instability of data acquisition in variable load environments can be mitigated, and the robustness and stability of the model can be enhanced. This allows the thermal variable prediction model to more accurately predict target thermal variable data (such as NOx emissions and boiler outlet flue gas temperature) based on the fusion feature set, improving the prediction accuracy of thermal variable data under variable load conditions, optimizing the unit energy utilization rate under peak shaving and frequency regulation of thermal power units, adjusting boiler outlet flue gas temperature, and reducing pollutant emissions.
[0145] After introducing an exemplary data prediction method provided by the present invention, the following describes an exemplary implementation of the apparatus. The data prediction apparatus provided by the present invention can be applied to... Figure 1 Any of the methods provided in the corresponding embodiments. See also Figure 4 The data prediction device includes at least:
[0146] The data acquisition module 401 is used to acquire equipment operating condition data and flame image data corresponding to each operating point;
[0147] Feature fusion module 402 is used to generate a fused feature set based on the equipment operating data and the flame image data;
[0148] The prediction module 403 is used to input the fused feature set into the thermal variable prediction model to obtain the target thermal variable data.
[0149] Optionally, the equipment operating condition data includes historical thermal variable data and various thermal process data related to the historical thermal variable data.
[0150] Specifically, the feature fusion module 402 is used to: perform feature selection on the historical thermal variable data and various thermal process data to obtain a numerical variable feature set; perform feature selection on the flame image data to obtain an image variable feature set; and fuse the numerical variable feature set and the image variable feature set to obtain the fused feature set.
[0151] Optionally, when the feature fusion module 402 fuses the numerical variable feature set and the image variable feature set to obtain the fused feature set, it is specifically used to: normalize each feature vector in the numerical variable feature set and the image variable feature set; and concatenate the processed numerical variable feature vector with the flame image feature vector to obtain the fused feature vector in the fused feature set.
[0152] In practical applications, optionally, the number of flame image feature vectors in the image-type variable feature set can be used as the number of fused feature vectors in the fusion feature set. Alternatively, the number of fused feature vectors in the fusion feature set can be determined based on the number of numerical variable feature vectors in the numerical variable feature set. Of course, the number of fused feature vectors in the fusion feature set can also be determined based on the number of both of the aforementioned feature vectors.
[0153] Optionally, the feature fusion module 402 normalizes each feature vector in the numerical variable feature set and the image variable feature set using the following formula to obtain the feature vector to be concatenated:
[0154]
[0155] Among them, y scaled Let y be the feature vector to be concatenated, and let y be any feature vector from the numerical variable feature set and the image variable feature set. min y is the minimum value of the numerical variable feature set or the image variable feature set. max R represents the maximum value of the numerical variable feature set or the image variable feature set. R and Q represent the maximum and minimum values of the preset scaling range. The formula here is similar to Formula 8; please refer to the above for details, which will not be elaborated here.
[0156] Optionally, when the feature fusion module 402 performs feature selection on the flame image data to obtain an image-type variable feature set, it is specifically used to: extract key image features from the flame image data; flatten the key image features to obtain a flame image feature vector in the image-type variable feature set, wherein the flame image feature vector is a one-dimensional feature vector.
[0157] Optionally, the flame image data includes multiple flame images.
[0158] Based on this, when the feature fusion module 402 extracts key image features from the flame image data, it specifically performs the following operations: inputting the multiple flame images into a convolutional neural network model to obtain the image feature matrix. The convolutional neural network model includes multiple convolutional layers, multiple pooling layers, a ReLU function, and a fully connected layer. Specifically, for each flame image, the convolutional neural network model performs the following operations: convolving each input flame image with a convolutional kernel and processing the convolution calculation result using the ReLU function; inputting the convolution calculation result processed by the ReLU function into a pooling layer to obtain an image feature matrix containing the key image features.
[0159] Furthermore, when the feature fusion module 402 flattens the key image features to obtain the flame image feature vector in the image variable feature set, it is specifically used to: convert the image feature matrix into the corresponding one-dimensional feature vector, and use the corresponding one-dimensional feature vector as the flame image feature vector in the image variable feature set.
[0160] Optionally, when the feature fusion module 402 performs feature selection on the historical thermal variable data and various thermal process data to obtain a numerical variable feature set, it specifically performs the following: calculates the principal components in the various thermal process data using a PLS model; determines the number of principal components to be extracted based on the principal component calculation results and the cross-validity principle; determines the degree of contribution of various thermal process data to the historical thermal variable data, wherein the greater the degree of contribution, the greater the correlation with the historical thermal variable data; selects thermal process data whose contribution degree meets the preset conditions from the various thermal process data based on the number of principal components to be extracted; and uses the selected thermal process data and the historical thermal process data as the numerical variable feature set.
[0161] Optionally, the thermal variable prediction model is obtained by iterative training using the fused feature set.
[0162] Optionally, the thermal variable prediction model includes forward long short-term memory units and backward long short-term memory units. For example, the thermal variable prediction model may consist of an input layer, an embedding layer, a forward LSTM layer, a backward LSTM layer, and an output layer.
[0163] It should be noted that, Figure 4 The provided embodiments and Figure 1 The provided embodiments are similar, and the similarities can be found in each other, so they will not be elaborated here.
[0164] Following the introduction of the data prediction method and apparatus according to exemplary embodiments of the present invention, the present invention next provides an exemplary medium storing computer-executable instructions that can be used to cause the computer to perform... Figure 1 The corresponding exemplary embodiments of the present invention are used for data prediction methods.
[0165] After introducing the data prediction method, medium, and apparatus according to exemplary embodiments of the present invention, the following references are made. Figure 5 This invention introduces an exemplary computing device 50, which includes a processing unit 501, a memory 502, a bus 503, an external device 504, an I / O interface 505, and a network adapter 506. The memory 502 includes a random access memory (RAM) 5021, a cache memory 5022, a read-only memory (ROM) 5023, and a memory cell array 5025 consisting of at least one memory cell 5024. The memory 502 stores programs or instructions executed by the processing unit 501; the processing unit 501 executes programs or instructions stored in the memory 502. Figure 1 The data prediction method described in any one of the exemplary embodiments of the present invention; the I / O interface 505 is used to receive or send data under the control of the processing unit 501.
[0166] Figure 6 A flowchart of a model training method provided in an embodiment of the present invention is shown below. Figure 6 As shown, the model training method may include the following steps:
[0167] 601. Obtain equipment operating condition data and flame image data corresponding to each operating point;
[0168] 602. Generate a fused feature set based on the equipment operating data and the flame image data;
[0169] 603. Input the fused feature set into the initial thermal variable prediction model, and iteratively train the initial thermal variable prediction model to obtain a thermal variable prediction model for predicting target thermal variable data.
[0170] Optionally, the thermal variable prediction model includes forward long short-term memory units and reverse long short-term memory units.
[0171] The execution process of steps 601 to 603 can be found in the descriptions of the other embodiments mentioned above, and will not be repeated here. It is understood that during iterative training, after step 603 is completed, the process must jump back to step 601 and repeat steps 601 to 603 until the iterative training ends. The termination condition for iterative training may be, for example, that the number of training iterations exceeds a preset number, or that the model passes a preset evaluation; other conditions are also possible and are not limited here.
[0172] Figure 7 This is a schematic diagram of the structure of a model training device provided in an embodiment of the present invention, as shown below. Figure 7 As shown, the device includes: a data acquisition module 701, a feature fusion module 702, and a training module 703.
[0173] The data acquisition module 701 is used to acquire equipment operating condition data and flame image data corresponding to each operating point;
[0174] Feature fusion module 702 is used to generate a fused feature set based on the equipment operating data and the flame image data;
[0175] The training module 703 is used to input the fused feature set into the initial thermal variable prediction model, and to iteratively train the initial thermal variable prediction model to obtain a thermal variable prediction model for predicting target thermal variable data.
[0176] Figure 7 The model training device shown can perform the aforementioned Figure 6 The model training method illustrated in this embodiment is not described in detail in this embodiment. For the parts not described in detail in this embodiment, please refer to the relevant descriptions in the foregoing embodiments. It will not be repeated here.
[0177] After introducing the model training method and apparatus of exemplary embodiments of the present invention, the present invention next provides an exemplary medium storing computer-executable instructions that can be used to cause the computer to perform... Figure 6 The corresponding exemplary embodiments of the present invention are used for model training methods.
[0178] After introducing the model training methods, media, and apparatus according to exemplary embodiments of the present invention, the following references are made. Figure 8 This invention introduces an exemplary computing device 80, which includes a processing unit 801, a memory 802, a bus 803, an external device 804, an I / O interface 805, and a network adapter 806. The memory 802 includes a random access memory 8021, a cache memory 8022, a read-only memory 8023, and a memory cell array 8025 composed of at least one memory cell 8024. The memory 802 stores programs or instructions executed by the processing unit 801; the processing unit 801 executes programs or instructions stored in the memory 802. Figure 6 The model training method described in any one of the exemplary embodiments of the present invention; the I / O interface 805 is used to receive or send data under the control of the processing unit 801.
[0179] Figure 9A flowchart of another data prediction method provided in an embodiment of the present invention is shown below. Figure 9 As shown, the model training method may include the following steps:
[0180] 901. Obtain equipment operating condition data and flame image data corresponding to each operating point;
[0181] 902. Generate a fused feature set based on the equipment operating data and the flame image data;
[0182] 903. Input the fused feature set into the initial thermal variable prediction model, and iteratively train the initial thermal variable prediction model to obtain the thermal variable prediction model.
[0183] 904. The target thermal variable data are predicted using the aforementioned thermal variable prediction model.
[0184] The execution process of steps 901 and 904 can be found in the descriptions of the other embodiments mentioned above, and will not be repeated here.
[0185] Figure 10 This is a schematic diagram of the structure of a data prediction device provided in an embodiment of the present invention, as shown below. Figure 10 As shown, the device includes: a data acquisition module 1001, a feature fusion module 1002, a training module 1003, and a prediction module 1004.
[0186] The data acquisition module 1001 is used to acquire equipment operating condition data and flame image data corresponding to each operating point;
[0187] Feature fusion module 1002 is used to generate a fused feature set based on the equipment operating data and the flame image data;
[0188] Training module 1003 is used to input the fused feature set into the initial thermal variable prediction model, and to iteratively train the initial thermal variable prediction model to obtain the thermal variable prediction model.
[0189] The prediction module 1004 is used to predict the target thermal variable data using the thermal variable prediction model.
[0190] Figure 10 The model training device shown can perform the aforementioned Figure 9 The data prediction method illustrated in this embodiment is not described in detail here. For the parts not described in detail in this embodiment, please refer to the relevant descriptions in the foregoing embodiments. It will not be repeated here.
[0191] After introducing the model training method and apparatus of exemplary embodiments of the present invention, the present invention next provides an exemplary medium storing computer-executable instructions that can be used to cause the computer to perform... Figure 8 The corresponding exemplary embodiments of the present invention are used for data prediction methods.
[0192] After introducing the model training methods, media, and apparatus according to exemplary embodiments of the present invention, the following references are made. Figure 11 This invention introduces an exemplary computing device 110, which includes a processing unit 1101, a memory 1102, a bus 1103, an external device 1104, an I / O interface 1105, and a network adapter 1106. The memory 1102 includes a random access memory 11021, a cache memory 11022, a read-only memory 11023, and a memory cell array 11025 composed of at least one memory cell 11024. The memory 1102 stores programs or instructions executed by the processing unit 1101; the processing unit 1101 executes programs or instructions stored in the memory 1102. Figure 9 The data prediction method described in any one of the exemplary embodiments of the present invention; the I / O interface 1105 is used to receive or send data under the control of the processing unit 1101.
[0193] An exemplary embodiment of the present invention also provides a measuring device, such as a two-phase flow mass flow measuring device. The data detected by this device (such as the mass flow data of two-phase flow and pulverized coal in the air conveying duct) can be applied in the modeling process of a thermal variable prediction model to improve the modeling accuracy. Specifically, by measuring the two-phase flow of pulverized coal and air in the primary air duct, the device can determine the combustion quality of boiler fuel, providing detection data and reference indicators for the thermal variable prediction model, thereby improving the modeling accuracy.
[0194] Of course, in another embodiment, the device also dynamically balances and adjusts the distribution based on detection data to improve the combustion quality of boiler fuel and provide a more efficient fuel utilization solution.
[0195] It should be noted that although several units / modules or sub-units / modules of the device have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of the present invention, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided and embodied by multiple units / modules.
[0196] Furthermore, although the operations of the method of the present invention are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.
[0197] While the spirit and principles of the invention have been described with reference to several specific embodiments, it should be understood that the invention is not limited to the disclosed specific embodiments, and the division of aspects does not imply that features in these aspects cannot be combined for benefit; such division is merely for ease of description. The invention is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.
Claims
1. A data prediction method, characterized in that, include: Acquire equipment operating condition data and flame image data corresponding to each operating point; A fusion feature set is generated based on the equipment operating data and the flame image data; The fused feature set is input into the thermal variable prediction model to obtain the target thermal variable data; The step of generating a fusion feature set based on the equipment operating data and the flame image data includes: Numerical variable feature sets and image variable feature sets are obtained from the equipment operating condition data and the flame image data, respectively. Normalize each feature vector in the numerical variable feature set and the image variable feature set; The processed numerical variable feature vector is concatenated with the flame image feature vector to obtain the fused feature vector in the fused feature set.
2. The method according to claim 1, characterized in that, The equipment operating data includes historical thermal variable data and various thermal process data related to the historical thermal variable data; Numerical variable feature sets and image variable feature sets are obtained from the equipment operating condition data and the flame image data, respectively, including: Feature selection is performed on the historical thermal variable data and various thermal process data to obtain a numerical variable feature set; Feature selection is performed on the flame image data to obtain an image-type variable feature set.
3. The method according to claim 1, characterized in that, The following formula is used to normalize each feature vector in the numerical variable feature set and the image variable feature set to obtain the feature vector to be concatenated: ； in, The feature vectors to be concatenated Let be any feature vector from the numerical variable feature set and the image variable feature set. The minimum value of the numerical variable feature set or the image variable feature set. R is the maximum value of the numerical variable feature set or the image variable feature set, R is the maximum value of the preset scaling range, and Q is the minimum value of the preset scaling range.
4. The method according to claim 2, characterized in that, Feature selection is performed on the flame image data to obtain an image-type variable feature set, including: Extract key image features from the flame image data; The key image features are flattened to obtain the flame image feature vector in the image-type variable feature set.
5. The method according to claim 4, characterized in that, The flame image data includes multiple flame images; Extracting key image features from the flame image data includes: The multiple flame images are input into a convolutional neural network model to obtain the image feature matrix. The convolutional neural network model includes multiple convolutional layers, multiple pooling layers, a ReLU function, and a fully connected layer. For each flame image, the convolutional neural network model performs the following operations: Each input flame image is convolved with a convolution kernel, and the convolution result is processed using the ReLU function. The convolution calculation result processed by the ReLU function is input into the pooling layer to obtain an image feature matrix containing the key image features; The key image features are flattened to obtain a flame image feature vector from the image-type variable feature set, including: The image feature matrix is converted into a corresponding one-dimensional feature vector, and the corresponding one-dimensional feature vector is used as the flame image feature vector in the image-type variable feature set.
6. The method according to claim 2, characterized in that, Feature selection is performed on the historical thermal variable data and various thermal process data to obtain a numerical variable feature set, including: The principal components in various thermal process data were calculated using the PLS model. Based on the principal component calculation results and the principle of cross-validation, the number of principal components to be extracted is determined; Determine the contribution of various thermal process data to the historical thermal variable data, wherein the greater the contribution, the greater the correlation with the historical thermal variable data; Based on the number of principal components to be extracted, thermal process data whose contribution level meets the preset conditions are selected from the various thermal process data. The selected thermal process data and the historical thermal work data are used as the feature set of the numerical variables.
7. The method according to claim 1, characterized in that, The thermal variable prediction model is obtained by iterative training using the fused feature set.
8. The method according to claim 1, characterized in that, The thermal variable prediction model includes forward long short-term memory units and reverse long short-term memory units.
9. A data prediction apparatus for implementing the steps of the method according to any one of claims 1-8, characterized in that, include: The data acquisition module is used to acquire equipment operating condition data and flame image data corresponding to each operating point. The feature fusion module is used to generate a fused feature set based on the equipment operating data and the flame image data; The prediction module is used to input the fused feature set into the thermal variable prediction model to obtain the target thermal variable data.
10. An electronic device, characterized in that, include: A memory and a processor; wherein the memory stores executable code, and when the executable code is executed by the processor, the processor performs the data prediction method as described in any one of claims 1 to 8.
11. A model training method for training a thermal variable prediction model according to any one of claims 1-8, characterized in that, include: Acquire equipment operating condition data and flame image data corresponding to each operating point; A fusion feature set is generated based on the equipment operating data and the flame image data; The fused feature set is input into the initial thermal variable prediction model, and the initial thermal variable prediction model is iteratively trained to obtain a thermal variable prediction model for predicting target thermal variable data.
12. A data prediction method, characterized in that, include: Acquire equipment operating condition data and flame image data corresponding to each operating point; A fusion feature set is generated based on the equipment operating data and the flame image data; The fused feature set is input into the initial thermal variable prediction model, and the initial thermal variable prediction model is iteratively trained to obtain the thermal variable prediction model. The aforementioned thermal variable prediction model is used to predict the target thermal variable data; The step of generating a fusion feature set based on the equipment operating data and the flame image data includes: Numerical variable feature sets and image variable feature sets are obtained from the equipment operating condition data and the flame image data, respectively. Normalize each feature vector in the numerical variable feature set and the image variable feature set; The processed numerical variable feature vector is concatenated with the flame image feature vector to obtain the fused feature vector in the fused feature set.