CN115527078A - Data prediction method, device and equipment and model training method - Google Patents

Abstract

The invention provides a data prediction method, a data prediction device, data prediction equipment and a model training method. The method comprises the following steps: acquiring equipment working condition data and flame image data corresponding to each working condition point; generating a fusion feature set based on the equipment working condition data and the flame image data; and inputting the fusion feature set into a thermal variable prediction model to obtain target thermal variable data. The method leads flame image data into fusion characteristics in a centralized manner, so that the real operation conditions of equipment at each working condition point reflected by the flame image data are applied to the prediction process of a thermal variable prediction model, thereby improving the time delay problem existing in the working condition data of the equipment, relieving the unstable condition of data acquisition in a variable-load working condition environment, enhancing the robustness and stability of the model, improving the prediction precision of the thermal variable data under the variable-load working condition, optimizing the energy utilization rate of a thermal power generating unit under peak and frequency modulation, adjusting the smoke temperature at the outlet of a boiler, and reducing the emission of pollutants.

Description

Data prediction method, device and equipment and model training method

Technical Field

The invention relates to the technical field of energy, in particular to a data prediction method, a data prediction device, data prediction equipment and a model training method.

Background

At present, fossil energy is still dominant in consuming energy. However, with the large amount of new energy power grid connection, the thermal power generating unit participates in peak shaving frequency modulation, so that the working condition load of the thermal power generating unit fluctuates, and the challenge is brought to the prediction of the thermal power variable data in the thermal power generating unit.

Taking a thermal power plant mainly for coal-fired power generation as an example, in order to optimize a coal-fired thermal power unit, the discharge amount of nitrogen oxides (NOx) and the temperature of flue gas at the outlet of a boiler (referred to as boiler outlet flue gas temperature for short) are usually predicted. In order to achieve the standard emission, the emission of nitrogen oxides (NOx) in the flue gas and the temperature of the flue gas at the outlet of the boiler need to be predicted, so that the combustion system and the desulfurization system of the boiler are optimized based on the prediction result, the energy utilization rate of a unit is improved, and the emission of NOx is reduced. However, in the related art, it is often difficult to determine the real-time working condition of the system due to the fact that the acquired data is delayed or unstable, and the accuracy of the prediction result of the thermal variable data is affected.

Therefore, a thermal variable prediction scheme suitable for the variable load working condition is urgently needed to be provided, and the scheme is used for more accurately predicting data such as NOx emission amount and boiler outlet smoke temperature under the variable load working condition.

Disclosure of Invention

The embodiment of the invention provides a data prediction method, a data prediction device, data prediction equipment and a model training method, which are used for accurately predicting thermotechnical variable data. For example, emission data of pollutants in the thermal power generating unit and the temperature of smoke at the outlet of the boiler are predicted.

In a first aspect, an embodiment of the present invention provides a data prediction method, where the method includes:

acquiring equipment working condition data and flame image data corresponding to each working condition point;

generating a fusion feature set based on the equipment working condition data and the flame image data;

and inputting the fusion feature set into a thermal variable prediction model to obtain target thermal variable data.

In a second aspect, an embodiment of the present invention provides a data prediction apparatus, including:

the data acquisition module is used for acquiring equipment working condition data and flame image data corresponding to each working condition point;

the feature fusion module is used for generating a fusion feature set based on the equipment working condition data and the flame image data;

and the prediction module is used for inputting the fusion feature set into a thermal variable prediction model to obtain target thermal variable data.

In a third aspect, an embodiment of the present invention provides an electronic device, which includes a processor and a memory, where the memory stores executable codes, and when the executable codes are executed by the processor, the processor is enabled to implement at least the data prediction method in the first aspect.

In a fourth aspect, embodiments of the present invention provide a non-transitory machine-readable storage medium having executable code stored thereon, which when executed by a processor of an electronic device, causes the processor to implement at least the data prediction method of the first aspect.

In a fifth aspect, an embodiment of the present invention provides a model training method, where the method includes:

acquiring equipment working condition data and flame image data corresponding to each working condition point;

generating a fusion feature set based on the equipment working condition data and the flame image data;

inputting the fusion feature set into an initial thermal variable prediction model, and performing iterative training on the initial thermal variable prediction model to obtain a thermal variable prediction model for predicting target thermal variable data.

In a sixth aspect, an embodiment of the present invention provides a model training apparatus, including:

the data acquisition module is used for acquiring equipment working condition data and flame image data corresponding to each working condition point;

the feature fusion module is used for generating a fusion feature set based on the equipment working condition data and the flame image data;

and the training module is used for inputting the fusion feature set into an initial thermal variable prediction model, and performing iterative training on the initial thermal variable prediction model to obtain a thermal variable prediction model for predicting target thermal variable data.

In a seventh aspect, an embodiment of the present invention provides an electronic device, which includes a processor and a memory, where the memory stores executable code, and when the executable code is executed by the processor, the processor is enabled to implement at least the model training method in the fifth aspect.

In an eighth aspect, an embodiment of the present invention provides a non-transitory machine-readable storage medium having executable code stored thereon, which, when executed by a processor of an electronic device, causes the processor to implement at least the model training method in the fifth aspect.

In a ninth aspect, an embodiment of the present invention provides a data prediction method, where the method includes:

acquiring equipment working condition data and flame image data corresponding to each working condition point;

generating a fusion feature set based on the equipment working condition data and the flame image data;

inputting the fusion feature set into an initial thermal variable prediction model, and performing iterative training on the initial thermal variable prediction model to obtain a thermal variable prediction model;

and predicting target thermal variable data by adopting the thermal variable prediction model.

In a tenth aspect, an embodiment of the present invention provides an electronic device, which includes a processor and a memory, where the memory stores executable code thereon, and when the executable code is executed by the processor, the processor is enabled to implement at least the data prediction method in the ninth aspect.

In an eleventh aspect, embodiments of the present invention provide a non-transitory machine-readable storage medium having stored thereon executable code, which when executed by a processor of an electronic device, causes the processor to implement at least the data prediction method of the ninth aspect.

In a twelfth aspect, embodiments of the present invention provide a measurement device, such as a two-phase flow mass flow measurement device, for use in the detection of two-phase flow mass flow in a primary air duct. The data (such as the mass flow data of the two-phase flow and the pulverized coal in the air conveying pipeline) detected by the device is applied to the modeling process of the thermal variable prediction model to improve the modeling precision of the thermal variable prediction model.

According to the technical scheme provided by the embodiment of the invention, the device working condition data and the flame image data corresponding to each working condition point are firstly obtained. The device working condition data can reflect the change condition of multiple monitoring indexes in the device operation process, and the flame image data further reflects the real operation condition of the device at each working condition point, so that the time delay problem existing in the device working condition data can be improved through a fusion characteristic set generated based on the device working condition data and the flame image data, the unstable condition of data acquisition in a variable load working condition environment is relieved, the robustness and the stability of the model are enhanced, the thermal variable prediction model can more accurately predict target thermal variable data based on the fusion characteristic set, the prediction precision of the thermal variable data (such as NOx emission and boiler outlet smoke temperature) under the variable load working condition is improved, the energy utilization rate of a thermal power unit under the peak-modulated frequency modulation of the thermal power unit is optimized, the boiler outlet smoke temperature is adjusted, and the pollutant emission is reduced.

Drawings

The above and other objects, features and advantages of exemplary embodiments of the present invention will become readily apparent from the following detailed description read in conjunction with the accompanying drawings. Several embodiments of the invention are illustrated by way of example, and not by way of limitation, in the figures of the accompanying drawings and in which:

fig. 1 is a schematic flow chart of a data prediction method according to an embodiment of the present invention;

FIG. 2 is a schematic diagram of a thermal variable prediction model according to an embodiment of the present invention;

FIG. 3 is a schematic structural diagram of a BilSTM model according to an embodiment of the present invention;

fig. 4 is a schematic structural diagram of a data prediction apparatus according to an embodiment of the present invention;

fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present invention;

FIG. 6 is a schematic flow chart of a model training method according to an embodiment of the present invention;

FIG. 7 is a schematic structural diagram of a model training apparatus according to an embodiment of the present invention;

fig. 8 is a schematic structural diagram of another electronic device according to an embodiment of the present invention;

FIG. 9 is a flow chart illustrating another data prediction method according to an embodiment of the present invention;

FIG. 10 is a schematic diagram of another data prediction apparatus according to an embodiment of the present invention;

fig. 11 is a schematic structural diagram of another electronic device according to an embodiment of the present invention.

In the drawings, the same or corresponding reference numerals indicate the same or corresponding parts.

Detailed Description

The principles and spirit of the present invention will be described with reference to a number of exemplary embodiments. It is understood that these embodiments are given solely for the purpose of enabling those skilled in the art to better understand and to practice the invention, and are not intended to limit the scope of the invention in any way. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

As will be appreciated by one skilled in the art, embodiments of the present invention may be embodied as a system, apparatus, device, method, or computer program product. Thus, the present invention may be embodied in the form of: entirely hardware, entirely software (including firmware, resident software, micro-code, etc.), or a combination of hardware and software.

According to the embodiment of the invention, a data prediction method, a data prediction device, data prediction equipment and a model training method are provided. Moreover, any number of elements in the drawings are by way of example and not by way of limitation, and any nomenclature is used solely for differentiation and not by way of limitation.

The principles and spirit of the present invention are explained in detail below with reference to several representative embodiments of the invention.

The inventor finds that with the large amount of new energy power grid connection, the thermal power generating unit participates in peak shaving frequency modulation, so that the working condition load of the thermal power generating unit fluctuates, the operation cost is increased, and the challenge is brought to the prediction of thermal variable data in the thermal power generating unit.

In the case of a coal-fired unit, the main pollutant in the flue gas discharged by the optimized coal-fired unit is NOx. In order to reach the standard emission, the emission data of NOx in the smoke needs to be predicted, so that the energy utilization rate of a unit is optimized based on the prediction result, and the emission of NOx is reduced. Besides the NOx emission, the boiler outlet flue gas temperature is also an important data index to be predicted when optimizing the unit. However, in the related art, it is often difficult to determine the real-time working condition of the system due to the fact that the acquired data is delayed or unstable, and the like, and the accuracy of the prediction result of the thermal variable data is affected.

In summary, a thermal variable prediction scheme applicable to the variable load working condition is urgently needed to be provided for more accurately predicting NOx emission data and boiler outlet smoke temperature under the variable load working condition.

In order to overcome at least one technical problem in the prior art, the invention provides a data prediction method, a data prediction device, data prediction equipment and a model training method. The data prediction method at least comprises the following steps: and acquiring equipment working condition data and flame image data corresponding to each working condition point, thereby generating a fusion feature set based on the equipment working condition data and the flame image data. And finally, inputting the fusion feature set introduced with the flame image data into a thermal variable prediction model to obtain target thermal variable data.

In the data prediction method, the flame image data is introduced into the fusion characteristic set, so that the real operation condition of the equipment at each working condition point reflected by the flame image data can be applied to the prediction process of a thermotechnical variable prediction model, the delay problem existing in the working condition data of the equipment is solved, the unstable condition of data acquisition in a variable load working condition environment is alleviated, and the robustness and the stability of the model are enhanced. Finally, the thermal variable prediction model can more accurately predict target thermal variable data based on the fusion feature set, the prediction precision of the thermal variable data under the variable load working condition is improved, the unit energy utilization rate under the peak and frequency modulation of the thermal power unit is optimized, the smoke temperature at the outlet of the boiler is adjusted, and the pollutant emission is reduced.

It is understood that the principles of the model training method, apparatus, medium, and device are similar to the data prediction method and will not be described herein again.

Having described the general principles of the invention, various non-limiting embodiments of the invention are described in detail below.

The technical scheme provided by the embodiment of the invention can be executed by an electronic device, and the electronic device can be a terminal device such as a PC (personal computer), a notebook computer and the like, and can also be a server. The server may be a physical server including an independent host, or may also be a virtual server carried by a host cluster, or may also be a cloud server.

The technical scheme provided by the embodiment of the invention can be suitable for various processing scenes of the thermal variable data, in particular to prediction scenes of the thermal variable data. For example, a scenario for predicting one or more thermal variable data (such as NOx emissions and boiler outlet flue gas temperature) under variable load conditions. For example, the prediction scene of the pollutant emission amount in the flue gas emitted by the coal burning unit, the prediction scene of the flue gas oxygen content in the thermal power generating unit, and the like. For example, a prediction scene of the boiler outlet smoke temperature in a thermal power generating unit.

In the following, a technical solution for predicting thermal variable data according to an exemplary embodiment of the present invention is described with reference to the accompanying drawings in conjunction with an application scenario. It should be noted that the above application scenarios are merely illustrated for the convenience of understanding the spirit and principles of the present invention, and the embodiments of the present invention are not limited in this respect. Rather, embodiments of the present invention may be applied to any scenario where applicable.

The embodiment of the invention provides a data prediction method, as shown in fig. 1, the data prediction method is applied to prediction of target thermal variable data, and the data prediction method at least comprises the following steps:

101. acquiring equipment working condition data and flame image data corresponding to each working condition point;

102. generating a fusion feature set based on the equipment working condition data and the flame image data;

103. and inputting the fusion feature set into a thermal variable prediction model to obtain target thermal variable data.

In the data prediction method shown in fig. 1, the device operating condition data is data that reflects the device operating condition, for example, data corresponding to multiple monitoring indexes when the device is operating. The flame image data further reflects the real operation condition of the equipment at each working condition point, for example, the flame image data can reflect the flame combustion condition in the boiler. Optionally, prior to 101, flame image data is acquired from the scene.

For a thermal power generating unit, assuming that thermal variable data is pollutant emission data in flue gas, the equipment working condition data comprises at least one item of thermal process data related to the pollutant emission data. Aiming at target thermal variable data needing to be predicted, equipment working condition data and flame image data corresponding to each working condition point can be collected firstly, and therefore a fusion feature set is generated based on the equipment working condition data and the flame image data. By introducing the flame image data into the fusion characteristic set, the real operation condition of the equipment at each working condition point reflected by the flame image data can be applied to the prediction process of a thermal variable prediction model, so that the time delay problem existing in the working condition data of the equipment can be conveniently improved, the unstable condition of data acquisition in a variable load working condition environment can be alleviated, and the robustness and the stability of the model can be enhanced. Finally, the thermal variable prediction model can more accurately predict target thermal variable data based on the fusion feature set, the prediction precision of the thermal variable data under the variable load working condition is improved, the unit energy utilization rate under the peak and frequency regulation of the thermal power unit is optimized, and the pollutant emission is reduced. Similarly, the method can be used for more accurately predicting the boiler outlet smoke temperature based on the fusion feature set so as to optimize the unit energy utilization rate under the peak and frequency regulation of the thermal power unit under the variable load working condition and adjust the boiler outlet smoke temperature.

The thermal variable prediction model provided by the embodiment of the invention is suitable for various working conditions, particularly variable load working conditions. Taking a thermal power generating unit as an example, under the influence of peak load regulation and frequency modulation, the thermal power generating unit is under a variable load working condition, so that thermal variable data often fluctuate. Alternatively, the operating point may be a point in time or a period of time during which the device operating data is collected.

Therefore, when designing the network structure of the thermal variable prediction model, the input data type and the data scale of the model are often required to be considered, and a neural network and a model depth suitable for the current application scenario are selected.

In practical application, the network structure of the thermal variable prediction model may be a single neural network or a hybrid neural network. It will be appreciated that in a hybrid neural network, a plurality of neural networks are typically combined in a predetermined manner, thereby allowing different types of neural networks to exert their respective advantages to improve the overall performance of the model. The combination of the neural networks includes series connection, parallel connection, and other connection ways, which are not limited herein. For example, a plurality of neural networks are sequentially superposed to be connected in series

Hybrid neural networks of the type. For example, different branches of the model are respectively formed by multiple neural networks, and outputs of the different branches are combined to construct a parallel hybrid neural network.

In an optional embodiment, the thermal variable prediction model is formed by mixing and constructing a Partial Least Squares (PLS) model, a Deep Convolutional Neural Network (DCNN) model and a Bi-directional Long Short-Term Memory (BiLSTM) model, so as to adapt to an analysis scenario of multi-source data related to thermal variable data prediction.

In practical application, a network structure of a thermotechnical variable prediction model constructed by mixing the three models is shown in fig. 2. In fig. 2, the PLS model and the DCNN model are connected in parallel, and feature extraction and mixing are performed on multi-source input data, so that a fusion result based on the multi-source input data is obtained and input into the BiLSTM model, and the BiLSTM model predicts target thermal variable data based on the fusion result. In fig. 2, the multi-source input data includes device operating condition data a, device operating condition data b, and flame image data. For example, assuming that the device operating condition data a is historical thermal variable data, the device operating condition data b may be various types of thermal process data related to the historical thermal variable data.

In the embodiment of the present invention, the multi-source data related to the prediction of the thermal variable includes, but is not limited to: equipment working condition data and flame image data. The device working condition data is numerical data, and the flame image data is image data. Similar to the above description, the device operating condition data can reflect the operating condition data of the device, and the flame image data can further reflect the actual operating condition of the device at each operating point.

First, the device condition data related to the embodiment of the present invention will be described.

The equipment working condition data is data used for reflecting the equipment running condition, such as data corresponding to multiple monitoring indexes during the equipment running. Optionally, the equipment working condition data includes thermal variable data and thermal process data related to the thermal variable data. Wherein the thermal variable data comprises historical thermal variable data.

For example, assuming that the plant is a thermal power generating unit, the thermal process data affecting pollutant emissions (such as NOx) may be selected from the plant operating condition data according to a relevant mechanism, such as: load of the unit (L) _oad ) Total coal quantity (F) _a ) Total boiler air volume (M) _a ) Furnace pressure (P) _a )，Water-coal ratio (W) _c ) Oxygen content of flue gas (Y) _O2 ) Total water supply (W) _a ) Wind coal ratio (W) _o ) Pressure (P) of primary hot air pressure in the main pipe _r ) Total hot primary air temperature (T) _w ) Total secondary air volume (W) of boiler _e ) Coal powder mass flow (C) of each primary air pipeline of 4 powder making systems _a 、C _b 、C _c 、C _d ) Instantaneous coal supply (C) of coal feeder _co ) Primary air quantity (W) of coal mill _y ) 4 coal mill rotational speed (R) _a 、R _b 、R _c 、R _d ) Main steam temperature (T) _a ) Main steam pressure (P) _c ) Secondary air temperature (T) on both sides _oa 、T _ob ) Temperature (T) of exhaust gas _p ) 6 secondary air baffle opening degree (S) _a 、S _b 、S _c 、S _d 、S _e 、S _f ) Opening degree of 4-layer burn-up damper (S) _oa 、S _ob 、S _oc 、S _od ). The 36-dimensional thermal process data can be acquired by a sensor or other monitoring equipment carried by equipment and a monitoring system. For example, the thermal process data may be acquired from a Distributed Control System (DCS).

It should be noted that the 36-dimensional thermal process data also has an influence on the boiler outlet smoke temperature, and therefore, the thermal process data can also be used in the prediction process of the boiler outlet smoke temperature.

In fact, the monitoring dimension of the thermal process data can also be set according to the actual application scenario, and is not limited to the number given in the above example. The monitoring quantity of the rotating speed of the coal mill, the opening degree of the secondary air baffle and the opening degree of the over-fire air baffle can be set according to actual equipment configuration or monitoring requirements.

It can be understood that the target thermal variable data to be predicted in one application scenario may be discharge data, such as pollutant discharge data and boiler outlet smoke temperature, or may be one or a combination of the above thermal process data in another application scenario, which is not limited in the embodiment of the present invention. In practical applications, the thermal variable data includes historical thermal variable data. In the present document, a type of data to which data to be predicted belongs in the device working condition data is referred to as thermal variable data, the collected data is referred to as historical thermal variable data, and the data to be predicted by the technical scheme provided by the present invention is referred to as target thermal variable data. In different scenes, the thermal variable data to be predicted can be emission data in equipment working condition data, can also be boiler outlet smoke temperature in the equipment working condition data, and can also be one or a combination of the multiple thermal process data.

In practical application, time delay exists during the data acquisition of the equipment working condition, and unstable conditions easily occur in the data acquisition under the variable load working condition, so that the real conditions of the equipment running under the current working condition point can not be reflected by the equipment working condition data, and the accuracy of the model prediction result is influenced.

In order to ensure the accuracy of the model prediction result, in addition to the above-described device operating condition data, the flame image data corresponding to each operating condition point needs to be acquired in 101. For example, in a thermal power generating unit, the flame image data may be furnace flame map data corresponding to each operating point. The flame image data is acquired by an image acquisition device or module, the specific device type or module type is determined by the actual application scene, and the embodiment of the invention is not limited.

Optionally, the flame image data corresponding to each operating point may be a single image or an image group formed by multiple images. For example, it is still assumed that the operating device is a thermal power generating unit, and in the thermal power generating unit, the flame image data corresponding to a certain operating point may be one image used for representing the combustion condition of the flame in the furnace, or may be multiple images acquired from different angles in the thermal power generating unit at the same time or in the same time period.

Optionally, the number of flame image data corresponds to the number of operating points so as to reflect the device operating conditions at each operating point. For example, the number of the flame images is consistent with the number of the operating points, or the group number of the flame image groups is consistent with the number of the operating points.

After the multi-source input data (i.e., the device operating condition data and the flame image data) acquired in 101 are introduced, in 102, a corresponding fusion feature set can be generated based on the multi-source input data corresponding to each operating condition point.

Specifically, it is assumed that the equipment working condition data includes historical thermal variable data and various thermal process data related to the historical thermal variable data. Based on this, 102 can be implemented as:

performing characteristic selection on historical thermal variable data and various thermal process data to obtain a numerical variable characteristic set; carrying out feature selection on flame image data to obtain an image type variable feature set; and fusing the numerical variable feature set and the image variable feature set to obtain a fused feature set.

Through the steps, the multi-source input data corresponding to each working condition point can be fused into a fusion feature set after feature extraction, and therefore, all factors influencing target thermotechnical variables in the operation process of equipment can be reflected more comprehensively through the fusion feature set.

In the following, a feature extraction manner for different types of data in multi-source input data is described with reference to specific examples.

For the flame image data, in 102, the step of performing feature selection on the flame image data corresponding to each operating point to obtain an image type variable feature set may be implemented as follows:

extracting key image features from the flame image data corresponding to each working point; and flattening the key image features to obtain flame image feature vectors in the image type variable feature set, wherein the flame image feature vectors are one-dimensional feature vectors.

Specifically, the flame image data corresponding to each operating point is an image type variable, so that the DCNN model can be used for feature selection of the flame image data. Optionally, the DCNN model is composed of a plurality of convolution layers for performing convolution operations on the flame image data and a pooling layer for performing pooling operations on the flame image data. The method can extract key image features from the flame image data through a plurality of convolution layers and pooling layers, and reduce the dimensions of the key image features. The activation function may be a ReLU function. Since the flame image data is two-dimensional data, an optional input format for the flame image data is (P, X, Y), where P is the number of layers of the convolution pooling layer, and X and Y represent the number of rows and columns of the two-dimensional data, respectively. Alternatively, the number of convolution pooling layers may be set to 3, such as P set to 3.

Furthermore, assuming that the processed key image feature is a matrix, in this case, the matrix needs to be flattened to obtain a one-dimensional feature vector corresponding to the matrix, where the one-dimensional feature vector is a flame image feature vector corresponding to the current operating point. Finally, after the key image features corresponding to the operating point are subjected to flattening treatment, the flame image feature vectors corresponding to the operating point form an image type variable feature set.

For example, one of the alternative embodiments may be implemented as: it is assumed that the flame image data includes a plurality of flame images. Based on the method, a plurality of flame images are input into a convolutional neural network model (such as a DCNN model) to obtain an image feature matrix, wherein the convolutional neural network model comprises a plurality of convolutional layers, a plurality of pooling layers, a ReLU function and a full connection layer. Wherein for each flame image, the following operations are performed through a convolutional neural network model: performing convolution calculation on each input flame image and a convolution kernel, and processing a convolution calculation result by adopting a ReLU function; and inputting the convolution calculation result processed by the ReLU function into a pooling layer to obtain an image feature matrix containing key image features. And converting the image feature matrix into a corresponding one-dimensional feature vector, and taking the corresponding one-dimensional feature vector as the flame image feature vector in the image type variable feature set.

In addition to the DCNN model described above, other networks or algorithms may be used in 102 to perform feature selection on the flame image data, such as a Visual Geometry Group Network (VGG) model or Alex Net model.

For the equipment working condition data, the method for selecting the thermal process data in the embodiment of the present invention may be a statistical method and a machine learning method, including but not limited to: one or more of Partial principal component analysis, kernel principal component analysis, partial Least Squares (PLS), feature selection (Relief) algorithm, and convolutional neural network.

In the case of PLS, the main principle of the algorithm is: when the maximum interpretation variance is extracted from the independent variable data, the correlation between dependent variables corresponding to the independent variable data is maximized. Based on the principle, PLS is adopted to determine the contribution degree of various thermal process data to the thermal variable data, so that the thermal process data with larger contribution degree is screened out to be used for the subsequent prediction process of the target thermal variable data.

Based on the above principle, in an optional embodiment, in 102, feature selection is performed on historical thermal variable data and multiple thermal process data corresponding to each operating point to obtain a numerical variable feature set, which can be implemented as follows:

calculating principal components in various thermal process data by adopting a PLS model; determining the number of the principal components to be extracted based on the principal component calculation result and a cross validity principle; determining contribution degrees of various thermal process data to historical thermal variable data, wherein the greater the contribution degree is, the greater the correlation with the historical thermal variable data is; selecting thermal process data with contribution degree meeting preset conditions from various thermal process data based on the number of main components to be extracted; and taking the selected thermal process data and historical thermal as a numerical variable characteristic set.

In the above steps, the thermotechnical process data more related to the historical thermotechnical variable data can be effectively screened out according to the contribution degree of various thermotechnical process data to the thermotechnical variable data, so that the quantity of the thermotechnical process data is effectively compressed, the complexity of a thermotechnical variable prediction model is simplified, and the prediction accuracy and the generalization capability of the thermotechnical variable prediction model are improved.

Still take a thermal power generating unit as an example, and assume historical thermal variable data as the content of pollutants in flue gas. It is assumed that each set of equipment condition data includes historical contaminant content (i.e., historical thermal variable data) and thermal process data related to the historical contaminant content.

Specifically, in the thermal power generating unit, feature selection is performed on numerical variables (namely historical thermal variable data and various thermal process data corresponding to each working point) based on the PLS.

Wherein, it is assumed that the input variables required to be input into the PLS model are thermal process data collected from the thermal power generating unit, including but not limited to the following data: load of the unit (L) _oad ) Total coal quantity (F) _a ) Total boiler air volume (M) _a ) Furnace pressure (P) _a ) Water to coal ratio (W) _c ) Oxygen content of flue gas (Y) _O2 ) Total water supply (W) _a ) Wind coal ratio (W) _o ) Pressure (P) of primary hot air pressure in the main pipe _r ) Total hot primary air temperature (T) _w ) Total secondary air volume (W) of boiler _e ) Coal powder mass flow (C) of each primary air pipeline of 4 powder making systems _a 、C _b 、C _c 、C _d ) Instantaneous coal supply (C) of coal feeder _co ) Primary air quantity (W) of coal mill _y ) 4 rotational speed (R) of coal mill _a 、R _b 、R _c 、R _d ) Main steam temperature (T) _a ) Main steam pressure (P) _c ) Two side secondary air temperature (T) _oa 、T _ob ) Temperature (T) of exhaust gas _p ) 6 secondary air baffle opening degree (S) _a 、S _b 、S _c 、S _d 、S _e 、S _f ) Opening degree of 4-layer burn-up damper (S) _oa 、S _ob 、S _oc 、S _od ). Assume that the output variables of the PLS model are: the amount of NOx emissions.

Based on the assumption, feature engineering can be carried out on a training sample set constructed based on numerical variables (namely equipment working condition data); furthermore, the contribution degree of various thermal process data to historical thermal variable data is analyzed by the method for selecting the thermal process data, so that a final input/output variable set is determined according to the contribution degree

Wherein x is _i ∈R ^{M ×p} ，y _i ∈R ^M×q P is the number of input variables (i.e., thermal process data) and q is the output variable (i.e., thermal process data)) The number of the cells.

Of course, the output variable of the PLS model may also be the boiler outlet flue gas temperature, such as a boiler outlet flue gas temperature prediction scenario, and the implementation process is similar to the process of treating the NOx emission, and is not expanded here.

Based on the PLS algorithm presented above, the main steps of feature engineering are as follows:

first, calculating a first pair of principal components as t ₁ And u ₁ Making the characteristic information capable of characterizing the variable set contained therein as much as possible, the calculation expression is as follows:

in the formula, E ₀ 、F ₀ The matrices are normalized for the set of variables, respectively.

Second, calculating

And

inner product of (2), and then t is obtained ₁ And u ₁ Covariance of (c) Cov (t) ₁ ,u ₁ ) The process conversion expression is found as follows:

third, calculate t ₁ The calculation process of the regression model of (3) is as follows:

in the formula, alpha ₁ And beta ₁ As a regression vector, calculated by the following expression:

the fourth step, by finding E ₁ 、F ₁ In place of E ₀ 、F ₀ And repeating the first step to the third step, and calculating each main component successively until the final number of the main components is obtained.

And determining the number of the main components to be extracted finally according to a cross validity principle. Specifically, the number of the final principal components to be extracted is determined based on a cross validity principle, and a calculation formula is as follows:

in the formula, y _i In order to be a true input value,

for the prediction value of sample i after h components are extracted,

to eliminate the predicted value of the sample point i to the sample point i. Satisfy the above

Characterizing principal component t _h The marginal contribution to the historical thermal variable data is significant.

Of course, in an alternative embodiment, the Variable Projection Importance indicator (VIP) may be used to determine the contribution value (i.e., contribution degree) of each of the above-mentioned various thermal process data to the NOx emission.

In particular when

Time-cross effectiveness, in which case the number of thermal processes to be extracted is determinedAccording to the quantity; further, calculating the contribution value of each of the plurality of thermal process data to the NOx emission, namely:

in formula 7, p is the number of independent variables, and m is the number of main components; r (y; t) _h ) Is y and t _h Y is equipment working condition data in thermal process data, t _h Historical thermal variable data is obtained, and h is obtained; w is a _hk Is a weight vector w _h The kth component of (1).

Optionally, in order to simplify the complexity of the thermal variable prediction model and improve the prediction accuracy of the thermal variable prediction model, part of thermal process data with VIP values smaller than a preset threshold may be removed, and the remaining thermal process data may be used as thermal process data for training the feature selection model for the numerical variable.

In practical applications, in addition to the above PLS model, other algorithms or models can be used to perform feature selection on the numerical variables (i.e., the device operating condition data), such as the above-listed partial principal component analysis method, kernel principal component analysis method, relief algorithm, convolutional neural network model, and the like, and are not expanded here.

Through the characteristic extraction mode, the numerical variable characteristic set and the image variable characteristic set can be obtained from multi-source input data, so that the multi-source input data can be fully fused, and the real working condition of equipment can be reflected. In practical application, besides the combination of the DCNN model and the PLS model, the multi-source input data can be subjected to feature extraction through other single models or mixed models, and the feature extraction is not expanded here.

Then, after the numerical variable feature set and the image variable feature set are obtained, in 102, the numerical variable feature set and the image variable feature set are fused to obtain a fusion feature set. One of the ways to obtain the fusion feature set may be specifically implemented as:

normalizing each feature vector in the numerical variable feature set and the image variable feature set; and splicing the processed numerical variable characteristic vector and the flame image characteristic vector to obtain a fusion characteristic vector in a fusion characteristic set.

Specifically, taking the combination of the DCNN model and the PLS model as an example, the feature vectors output by the two models are normalized and then spliced together to form a fusion feature vector, which is used as the input of the thermal variable prediction model.

In an optional embodiment, the following formula is adopted to normalize each feature vector in the numerical variable feature set and the image variable feature set to obtain a feature vector to be spliced:

wherein, y _scaled Is a feature vector to be spliced, y is any one feature vector in the numerical variable feature set and the image variable feature set, y _min Is the minimum value, y, of the set of numerical or image-based variable features _max Is the maximum value of the numerical variable feature set or the image variable feature set.

In practical application, y of numerical variable feature set _min And y _max Can be determined from the data samples. Y of image type variable feature set _min Is typically set at 0,y _max Typically set at 255.R is the maximum value of the preset zoom range, Q is the minimum value of the preset zoom range, and the preset zoom range is generally set to be between 0 and 1.

Through the steps, a numerical variable feature set and an image variable feature set can be obtained from multi-source input data, and a fusion feature set containing multi-source feature information is generated based on fusion of the two variable feature sets, so that the time delay problem of equipment working condition data is solved, the unstable data acquisition condition in the variable-load working condition environment is relieved, and the robustness and the stability of a model are enhanced.

Finally, after the fusion feature set is obtained, the fusion feature set can be input into a thermal variable prediction model to obtain target thermal variable data.

The thermal variable prediction model is obtained by adopting a fusion feature set to carry out iterative training. For example, a fusion feature set obtained by fusing a numerical variable feature sample set and an image variable feature sample set is adopted to perform iterative training of preset times on the thermotechnical variable prediction model. The predetermined number of times is 10.

Optionally, after the iterative training of the thermal variable prediction model is completed, the model may be quantitatively evaluated to further optimize the model. The quantitative evaluation method includes but is not limited to: average relative error, root mean square error.

In practical application, the thermal variable prediction model may be a single model or a mixed model. The thermal variable prediction model includes, but is not limited to, one or a combination of the following: a BilSTM model, a support vector machine, a BP neural network, a lifting tree model (XGboost) and a fuzzy tree.

Taking the example that the thermotechnical variable prediction model is the BilSTM model, the BilSTM model has time series characteristics and has better prediction effect on data acquired based on each working condition point.

Specifically, the thermotechnical variable prediction model comprises a forward long-short term memory unit and a reverse long-short term memory unit. The long and short term memory unit (LSTM) model enables the characteristic information with time sequence characteristics to be better transmitted backwards through the combination of the input gate, the forgetting gate, the output gate and the storage memory unit, and therefore the problems of gradient extinction and gradient explosion in the long sequence training process are effectively solved. Particularly, the BilSTM model can simultaneously record effective information in the characteristic data from the forward direction and the reverse direction, has stronger relevance, and can overcome the problem of inconsistent time delay among various characteristic data acquired at the same time, thereby better memorizing valuable information in the time sequence data.

The network structure of the BilSTM model provided by the embodiment of the invention is shown in FIG. 3. In fig. 3, the BiLSTM model includes: an input layer, an Embedding (Embedding) layer, a forward Long Short Term Memory (LSTM) layer, a reverse Long Short Term Memory (LSTM) layer, and an output layer. Fusion feature setThe fused feature vector of (1) { x } ₁ ，x ₂ ，x ₃ ，……，x _n Inputting the BiLSTM model, processing the BiLSTM model by an Embedding layer, a forward LSTM layer and a reverse LSTM layer, and outputting target thermal variable data (y) predicted by the BiLSTM model ₁ ，y ₂ ，y ₃ ，……，y _n }. Wherein, { h } ₁ ，h ₂ ，h ₃ ，……，h _n And { e (x) } and { e ₁ )，e(x ₂ )，e(x ₃ )，……，e(x _n ) Is an intermediate quantity.

Through the steps, the thermal variable prediction model can more accurately predict the target thermal variable data based on the fusion characteristic set, improve the prediction precision of the thermal variable data under the variable load working condition, optimize the unit energy utilization rate under the peak and frequency modulation of the thermal power unit, adjust the smoke temperature at the outlet of the boiler, reduce the pollutant emission and reduce the operation cost.

In the data prediction method shown in fig. 1, first, device operating condition data and flame image data corresponding to each operating condition point are obtained. Furthermore, the time delay problem of the equipment working condition data can be improved through a fusion characteristic set generated based on the equipment working condition data and the flame image data, the unstable condition of data acquisition in a variable load working condition environment is relieved, the robustness and the stability of the model are enhanced, the thermal variable prediction model can more accurately predict target thermal variable data (such as NOx emission amount and boiler outlet smoke temperature) based on the fusion characteristic set, the prediction precision of the thermal variable data under the variable load working condition is improved, the unit energy utilization rate under the peak-load modulation of the thermal power unit is optimized, the boiler outlet smoke temperature is adjusted, and the pollutant emission is reduced.

Having described a data prediction method exemplary of the present invention, an exemplary implementation of the apparatus is described. The data prediction device provided by the invention can be applied to any method provided by the embodiment corresponding to the figure 1. Referring to fig. 4, the data prediction apparatus includes at least:

the data acquisition module 401 is configured to acquire device operating condition data and flame image data corresponding to each operating condition point;

a feature fusion module 402 configured to generate a fusion feature set based on the device operating condition data and the flame image data;

and the prediction module 403 is configured to input the fusion feature set into a thermal variable prediction model to obtain target thermal variable data.

Optionally, the equipment working condition data includes historical thermal variable data and various thermal process data related to the historical thermal variable data.

Wherein, the feature fusion module 402 is specifically configured to: performing feature selection on the historical thermal variable data and the multiple thermal process data to obtain a numerical variable feature set; carrying out feature selection on the flame image data to obtain an image type variable feature set; and fusing the numerical variable feature set and the image variable feature set to obtain the fused feature set.

Optionally, the feature fusion module 402 is configured to fuse the numerical variable feature set and the image variable feature set to obtain a fusion feature set, and specifically configured to: normalizing each feature vector in the numerical variable feature set and the image variable feature set; and splicing the processed numerical variable characteristic vector and the flame image characteristic vector to obtain a fusion characteristic vector in the fusion characteristic set.

In practical application, optionally, the number of the flame image feature vectors in the image type variable feature set is used as the number of the fusion feature vectors in the fusion feature set. Alternatively, the number of the fused feature vectors in the fused feature set may be determined based on the number of the numerical variable feature vectors in the numerical variable feature set. Of course, the number of the fused feature vectors in the fused feature set can also be determined according to the number of the two feature vectors.

Optionally, the feature fusion module 402 normalizes each feature vector in the numerical variable feature set and the image variable feature set by using the following formula to obtain a feature vector to be spliced:

wherein, y _scaled Is a feature vector to be spliced, y is any one feature vector in the numerical variable feature set and the image variable feature set, y _min Is the minimum value, y, of the numerical variable feature set or the image variable feature set _max And the maximum value of the numerical variable feature set or the image variable feature set is obtained. R and Q are the maximum value and the minimum value of a preset scaling range. The formula here is similar to formula 8, and as explained above, is not expanded here.

Optionally, the feature fusion module 402 is configured to, when performing feature selection on the flame image data to obtain an image type variable feature set, specifically: extracting key image features from the flame image data; and flattening the key image features to obtain flame image feature vectors in an image type variable feature set, wherein the flame image feature vectors are one-dimensional feature vectors.

Wherein, optionally, the flame image data comprises a plurality of flame images.

Based on this, when the feature fusion module 402 extracts the key image features from the flame image data, it is specifically configured to: inputting the flame images into a convolutional neural network model to obtain the image feature matrix, wherein the convolutional neural network model comprises a plurality of convolutional layers, a plurality of pooling layers, a ReLU function and a full connection layer. Wherein for each flame image, the following operations are performed by the convolutional neural network model: performing convolution calculation on each input flame image and a convolution kernel, and processing a convolution calculation result by adopting a ReLU function; and inputting the convolution calculation result processed by the ReLU function into a pooling layer to obtain an image characteristic matrix containing the key image characteristics.

Further, the feature fusion module 402 is specifically configured to, when flattening the key image features to obtain a flame image feature vector in an image type variable feature set: and converting the image characteristic matrix into a corresponding one-dimensional characteristic vector, and taking the corresponding one-dimensional characteristic vector as a flame image characteristic vector in the image type variable characteristic set.

Optionally, the feature fusion module 402 is configured to perform feature selection on the historical thermal variable data and the multiple thermal process data to obtain a numerical variable feature set, and is specifically configured to: calculating principal components in various thermal process data by adopting a PLS model; determining the number of the principal components to be extracted based on the principal component calculation result and a cross validity principle; determining contribution degrees of various thermal process data to the historical thermal variable data, wherein the greater the contribution degree is, the greater the correlation with the historical thermal variable data is; selecting thermal process data with contribution degree meeting preset conditions from the multiple thermal process data based on the number of main components to be extracted; and taking the selected thermal process data and the historical thermal as the numerical variable feature set.

Optionally, the thermal variable prediction model is obtained by performing iterative training by using the fusion feature set.

Optionally, the thermotechnical variable prediction model comprises a forward long-short term memory unit and a reverse long-short term memory unit. For example, the thermal variable prediction model may be composed of an input layer, an Embedding layer, a forward LSTM layer, a backward LSTM layer, and an output layer.

It should be noted that the embodiment provided in fig. 4 is similar to the embodiment provided in fig. 1, and the similarities are mutually referred to and will not be expanded here.

Having described the data prediction method and apparatus of the exemplary embodiments of the present invention, the present invention provides an exemplary medium having stored thereon computer-executable instructions operable to cause a computer to implement any one of the corresponding exemplary embodiments of the present invention of fig. 1 for a data prediction method.

Having described the method, medium, and apparatus for data prediction according to the exemplary embodiment of the present invention, next, referring to fig. 5, an exemplary computing device 50 provided by the present invention is described, where the computing device 50 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, and the Memory 502 includes a Random Access Memory (RAM) 5021, a cache Memory 5022, a Read-Only Memory (ROM) 5023, and a storage unit array 5025 made up of at least one storage unit 5024. The memory 502 is used for storing programs or instructions executed by the processing unit 501; the processing unit 501 is configured to execute the data prediction method according to any one of the exemplary embodiments of the present invention shown in fig. 1 according to a program or an instruction stored in the memory 502; the I/O interface 505 is used for receiving or transmitting data under the control of the processing unit 501.

Fig. 6 is a flowchart of a model training method according to an embodiment of the present invention, and as shown in fig. 6, the model training method may include the following steps:

601. acquiring equipment working condition data and flame image data corresponding to each working condition point;

602. generating a fusion feature set based on the equipment working condition data and the flame image data;

603. inputting the fusion feature set into an initial thermal variable prediction model, and performing iterative training on the initial thermal variable prediction model to obtain a thermal variable prediction model for predicting target thermal variable data.

Optionally, the thermotechnical variable prediction model comprises a forward long-short term memory unit and a reverse long-short term memory unit.

The execution process of step 601 to step 603 may refer to the description in the foregoing other embodiments, which is not described herein again. It can be understood that, in the iterative training process, after the step 603 is finished, the process jumps to the step 601, and the steps 601 to 603 are repeatedly executed until the iterative training is finished. The end condition of the iterative training is, for example, that the training number exceeds a preset number, or the model is evaluated through a preset evaluation, but may also be other conditions, which are not limited herein.

Fig. 7 is a schematic structural diagram of a model training apparatus according to an embodiment of the present invention, and as shown in fig. 7, the apparatus includes: a data acquisition module 701, a feature fusion module 702, and a training module 703.

The data acquisition module 701 is used for acquiring equipment working condition data and flame image data corresponding to each working condition point;

a feature fusion module 702, configured to generate a fusion feature set based on the device operating condition data and the flame image data;

a training module 703, configured to input the fusion feature set into an initial thermal variable prediction model, and perform iterative training on the initial thermal variable prediction model to obtain a thermal variable prediction model for predicting target thermal variable data.

The model training apparatus shown in fig. 7 may perform the model training method illustrated in the foregoing embodiment of fig. 6, and parts not described in detail in this embodiment may refer to the related description of the foregoing embodiment, which is not repeated herein.

Having described the model training method and apparatus of the exemplary embodiments of this invention, the present invention provides an exemplary medium having stored thereon computer-executable instructions operable to cause a computer to implement any one of the corresponding exemplary embodiments of this invention of FIG. 6 for a model training method.

Having described methods, media, and apparatus for model training in accordance with exemplary embodiments of the invention, reference is now made to FIG. 8, which illustrates an exemplary computing device 80, wherein computing device 80 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, and wherein memory 802 includes a random access memory 8021, a cache memory 8022, a read-only memory 8023, and a storage unit array 8025 of at least one storage unit 8024. The memory 802 is used for storing programs or instructions executed by the processing unit 801; the processing unit 801 is configured to execute the model training method according to any one of the exemplary embodiments of the present invention corresponding to fig. 6 according to a program or an instruction stored in the memory 802; the I/O interface 805 is used for receiving or transmitting data under the control of the processing unit 801.

Fig. 9 is a flowchart of another data prediction method according to an embodiment of the present invention, and as shown in fig. 9, the model training method may include the following steps:

901. acquiring equipment working condition data and flame image data corresponding to each working condition point;

902. generating a fusion feature set based on the equipment working condition data and the flame image data;

903. inputting the fusion feature set into an initial thermal variable prediction model, and performing iterative training on the initial thermal variable prediction model to obtain a thermal variable prediction model;

904. and predicting target thermal variable data by adopting the thermal variable prediction model.

The execution process of step 901 and step 904 may refer to the description in the foregoing other embodiments, and is not described herein again.

Fig. 10 is a schematic structural diagram of a data prediction apparatus according to an embodiment of the present invention, as shown in fig. 10, the apparatus includes: a data acquisition module 1001, a feature fusion module 1002, a training module 1003, and a prediction module 1004.

The data acquisition module 1001 is used for acquiring equipment working condition data and flame image data corresponding to each working condition point;

a feature fusion module 1002, configured to generate a fusion feature set based on the device operating condition data and the flame image data;

a training module 1003, configured to input the fusion feature set into an initial thermal variable prediction model, and perform iterative training on the initial thermal variable prediction model to obtain a thermal variable prediction model;

the prediction module 1004 is configured to predict target thermal variable data by using the thermal variable prediction model.

The model training apparatus shown in fig. 10 can perform the data prediction method illustrated in the foregoing embodiment shown in fig. 9, and for parts not described in detail in this embodiment, reference may be made to the related description of the foregoing embodiment, which is not repeated herein.

Having described the model training method and apparatus of the exemplary embodiments of this invention, the present invention provides an exemplary medium having stored thereon computer-executable instructions operable to cause a computer to implement any of the corresponding exemplary embodiments of this invention of FIG. 8 for a data prediction method.

Having described methods, media, and apparatus for model training in accordance with exemplary embodiments of the invention, reference is now made to fig. 11, which illustrates an exemplary computing device 110, where computing device 110 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, and where memory 1102 includes a random access memory 11021, a cache memory 11022, a read-only memory 11023, and a storage unit array 11025 formed of at least one piece of storage unit 11024. The memory 1102 is used for storing programs or instructions executed by the processing unit 1101; the processing unit 1101 is configured to execute the data prediction method according to any one of the exemplary embodiments of the present invention corresponding to fig. 9 according to the program or the instructions stored in the memory 1102; the I/O interface 1105 is used for receiving or transmitting data under the control of the processing unit 1101.

Exemplary embodiments of the present invention also provide a measurement device, such as a two-phase flow mass flow measurement device. The data (such as two-phase flow and mass flow data of coal dust in the air conveying pipeline) detected by the device can be applied to the modeling process of the thermal variable prediction model to improve the modeling precision of the thermal variable prediction model. Specifically, the device can acquire the combustion quality of the boiler fuel through two-phase flow measurement of pulverized coal and air in the primary air pipe, and provides detection data and reference indexes for a thermal variable prediction model, so that the modeling precision of the model is improved.

Of course, in another embodiment, the apparatus also dynamically equalizes, distributes adjustments based on sensed data to improve combustion quality of boiler fuel, providing a more efficient fuel utilization scheme.

It should be noted that although in the above detailed description several units/modules or sub-units/modules of the apparatus are mentioned, such a division is merely exemplary and not mandatory. Indeed, the features and functionality of two or more of the units/modules described above may be embodied in one unit/module according to embodiments of the invention. Conversely, the features and functions of one unit/module described above may be further divided into embodiments by a plurality of units/modules.

Moreover, while the operations of the method of the invention are depicted in the drawings in a particular order, this does not require or imply that the operations must be performed in this particular order, or that all of the illustrated operations must be performed, to achieve desirable results. Additionally or alternatively, certain steps may be omitted, multiple steps combined into one step execution, and/or one step broken down into multiple step executions.

While the spirit and principles of the invention have been described with reference to several particular embodiments, it is to be understood that the invention is not limited to the disclosed embodiments, nor is the division of aspects, which is for convenience only as the features in such aspects may not be combined to benefit. The invention is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.

Claims (13)

1. A method of data prediction, comprising:

acquiring equipment working condition data and flame image data corresponding to each working condition point;

generating a fusion feature set based on the equipment working condition data and the flame image data;

and inputting the fusion feature set into a thermal variable prediction model to obtain target thermal variable data.

2. The method of claim 1, wherein the equipment condition data comprises historical thermal variable data, a plurality of thermal process data related to the historical thermal variable data;

generating a fusion feature set based on the device operating condition data and the flame image data, including:

performing characteristic selection on the historical thermal variable data and the multiple thermal process data to obtain a numerical variable characteristic set;

carrying out feature selection on the flame image data to obtain an image type variable feature set;

and fusing the numerical variable feature set and the image variable feature set to obtain the fused feature set.

3. The method according to claim 2, wherein fusing the numerical variable feature set and the image variable feature set to obtain the fused feature set comprises:

normalizing each feature vector in the numerical variable feature set and the image variable feature set;

and splicing the processed numerical variable characteristic vector and the flame image characteristic vector to obtain a fusion characteristic vector in the fusion characteristic set.

4. The method according to claim 3, wherein the normalization processing is performed on each feature vector in the numerical variable feature set and the image variable feature set by using the following formula to obtain feature vectors to be spliced:

wherein, y _scaled Is a feature vector to be spliced, y is any one feature vector in the numerical variable feature set and the image variable feature set, y _min Is the minimum value, y, of the numerical variable feature set or the image variable feature set _max And the maximum value of the numerical variable feature set or the image variable feature set is obtained.

5. The method of claim 2, wherein the feature selection of the flame image data resulting in an image-based variable feature set comprises:

extracting key image features from the flame image data;

and flattening the key image features to obtain flame image feature vectors in the image type variable feature set.

6. The method of claim 5, wherein the flame image data comprises a plurality of flame images;

extracting key image features from the flame image data, including:

inputting the flame images into a convolutional neural network model to obtain the image characteristic matrix, wherein the convolutional neural network model comprises a plurality of convolutional layers, a plurality of pooling layers, a ReLU function and a full connection layer;

for each flame image, the following operations are carried out through the convolutional neural network model:

performing convolution calculation on each input flame image and a convolution kernel, and processing a convolution calculation result by adopting a 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;

flattening the key image features to obtain flame image feature vectors in an image type variable feature set, wherein the flattening comprises the following steps:

and converting the image characteristic matrix into a corresponding one-dimensional characteristic vector, and taking the corresponding one-dimensional characteristic vector as a flame image characteristic vector in the image type variable characteristic set.

7. The method of claim 2, wherein the performing feature selection on the historical thermal variable data and the plurality of thermal process data to obtain a numerical variable feature set comprises:

calculating principal components in various thermal process data by adopting a PLS model;

determining the number of the principal components to be extracted based on the principal component calculation result and a cross validity principle;

determining contribution degrees of various thermotechnical process data to the historical thermotechnical variable data, wherein the greater the contribution degree, the greater the correlation with the historical thermotechnical variable data;

selecting thermal process data with contribution degree meeting preset conditions from the multiple thermal process data based on the number of main components to be extracted;

and taking the selected thermal process data and the historical thermal as the numerical variable feature set.

8. The method according to claim 1, wherein the thermal variable prediction model is iteratively trained using the fused feature set.

9. The method of claim 1, wherein the thermal variable prediction model comprises a forward long short term memory unit and a reverse long short term memory unit.

10. A data prediction apparatus, comprising:

the data acquisition module is used for acquiring equipment working condition data and flame image data corresponding to each working condition point;

the feature fusion module is used for generating a fusion feature set based on the equipment working condition data and the flame image data;

and the prediction module is used for inputting the fusion feature set into a thermal variable prediction model to obtain target thermal variable data.

11. An electronic device, comprising: a memory, a processor; wherein the memory has stored thereon executable code which, when executed by the processor, causes the processor to perform a data prediction method as claimed in any one of claims 1 to 9.

12. A method of model training, comprising:

acquiring equipment working condition data and flame image data corresponding to each working condition point;

generating a fusion feature set based on the equipment working condition data and the flame image data;

inputting the fusion feature set into an initial thermal variable prediction model, and performing iterative training on the initial thermal variable prediction model to obtain a thermal variable prediction model for predicting target thermal variable data.

13. A method of data prediction, comprising:

acquiring equipment working condition data and flame image data corresponding to each working condition point;

generating a fusion feature set based on the equipment working condition data and the flame image data;

inputting the fusion feature set into an initial thermal variable prediction model, and performing iterative training on the initial thermal variable prediction model to obtain a thermal variable prediction model;

and predicting target thermal variable data by adopting the thermal variable prediction model.

Images