Method, device and equipment for predicting multi-physical field parameters in combustion and medium

Abstract

The invention discloses a multi-physical field parameter prediction method, device and equipment in combustion and a medium, relates to the technical field of industrial equipment combustion control, and can improve the precision and robustness of multi-physical field parameter prediction. According to the scheme, the method comprises the steps that current physical environment parameters of industrial equipment are obtained in real time, wherein the current physical environment parameters comprise thermal field parameters, fluid field parameters, electromagnetic field parameters and environment parameters; the current physical environment parameters are input into a preset target prediction model, predicted values corresponding to the current physical environment parameters are obtained, the predicted values are used for indicating the combustion states corresponding to the current physical environment parameters, and the target prediction model is obtained based on sample set training; the sample set comprises time sequence simulation data of a plurality of different physical fields under different combustion conditions, environmental parameters and corresponding real value training; and inputting the predicted value into a preset mapping function to obtain the combustion performance of the corresponding industrial equipment.

Description

Technical Field

[0001] The present application relates to the technical field of industrial equipment combustion control, and in particular to a method, device, equipment and medium for predicting multi-physical field parameters in combustion. Background Art

[0002] With the increasing demand in the industrial field for coupled modeling and optimization of multiple physical fields (such as thermal fields, fluid fields, electromagnetic fields, etc.), accurate prediction and control of these physical field parameters are of great significance for improving the combustion efficiency of industrial equipment, reducing energy consumption, and ensuring system safety.

[0003] However, traditional modeling methods often face challenges such as high data dimension, strong feature diversity, and strong nonlinearity when dealing with the complex coupling relationship of multiple physical fields, which makes the prediction accuracy and computational efficiency unable to meet the needs of modern industrial scenarios. Summary of the Invention

[0004] The present application provides a method, device, equipment and medium for predicting multi-physical field parameters in combustion, which can improve the accuracy and robustness of multi-physical field parameter prediction, thereby improving the intelligence and accuracy of combustion control of industrial equipment.

[0005] In order to achieve the above objectives, this application adopts the following technical solutions:

[0006] In a first aspect of an embodiment of the present application, a method for predicting multi-physics field parameters in combustion is provided, the method comprising:

[0007] Real-time acquisition of current physical environment parameters of industrial equipment, including thermal field parameters, fluid field parameters, electromagnetic field parameters, and environmental parameters;

[0008] Inputting the current physical environment parameters into a preset target prediction model to obtain a predicted value corresponding to the current physical environment parameters, wherein the predicted value is used to indicate the combustion state corresponding to the current physical environment parameters, wherein the target prediction model is trained based on a sample set, wherein the sample set includes: time series simulation data of multiple different physical fields under different combustion conditions, environmental parameters, and corresponding real values;

[0009] The predicted value is input into a preset mapping function to obtain the combustion performance of the corresponding industrial equipment.

[0010] As a possible implementation, before obtaining the current physical environment parameters of the current industrial combustion process in real time, the method further includes:

[0011] obtaining the sample set;

[0012] Inputting the simulation data and environmental parameters in the sample set into the initial prediction model in sequence;

[0013] Using the initial prediction model to perform a training process on each input simulation data and the environmental parameters until the prediction error of the initial prediction model is less than a preset threshold, thereby obtaining the target prediction model;

[0014] Wherein, one of the training processes includes:

[0015] After extracting and fusing the features of the simulation data and the environmental parameters in the time series dimension, a time series fusion matrix is obtained;

[0016] After extracting and fusing frequency domain dimension features of the time series fusion matrix, a frequency domain fusion matrix is obtained;

[0017] Obtaining the predicted value according to the time series fusion matrix and the frequency domain fusion matrix;

[0018] The prediction error is obtained according to the predicted value and the true value corresponding to the predicted value.

[0019] As a possible implementation manner, before extracting and fusing the features of the simulation data and the environmental parameters in the time series dimension, the method further includes:

[0020] Connecting and flattening the simulation data and the environmental parameters to obtain an input vector;

[0021] Performing multi-scale decomposition on the input vector to obtain multiple subsequence vectors of different time scales;

[0022] Normalizing each of the subsequence vectors to obtain a plurality of corresponding target subvectors;

[0023] Extracting and fusing time-series features of the simulation data and the environmental parameters includes:

[0024] Performing time-series feature extraction and fusion on the multiple target sub-vectors.

[0025] As a possible implementation manner, the extracting and fusing features of the multiple target sub-vectors in the time series dimension includes:

[0026] After extracting the temporal features of each target sub-vector using a first multi-layer perceptron, a corresponding first feature matrix is obtained;

[0027] Using a second multi-layer perceptron, extract the interaction information between the first feature matrices at different scales to obtain a second feature matrix;

[0028] Perform weighted fusion on the second feature matrix to obtain the time series fusion matrix.

[0029] As a possible implementation, after extracting and fusing the frequency domain dimension features of the time series fusion matrix, a frequency domain fusion matrix is obtained, including:

[0030] Performing time series segmentation on the time series fusion matrix to obtain multiple sub-fusion matrices;

[0031] Performing frequency domain conversion on the multiple sub-fusion matrices to obtain multiple frequency domain matrices;

[0032] performing noise filtering on the plurality of frequency domain matrices to obtain a plurality of corresponding target frequency domain matrices;

[0033] Using a third multi-layer perceptron, extracting frequency domain features of the plurality of target frequency domain matrices to obtain a plurality of frequency domain feature matrices;

[0034] A plurality of the frequency domain feature matrices are weightedly fused to obtain the frequency domain fusion matrix.

[0035] As a possible implementation manner, obtaining corresponding prediction values according to the time series fusion matrix and the frequency domain fusion matrix includes:

[0036] A fourth multi-layer perceptron is used to perform feature extraction on the time series fusion matrix and the frequency domain fusion matrix to obtain the predicted value.

[0037] As a possible implementation, after inputting the predicted value into a preset mapping function to obtain the corresponding combustion performance of the industrial equipment, the method further includes:

[0038] Control information of the industrial equipment is generated based on the combustion performance.

[0039] In a second aspect of an embodiment of the present application, a device for predicting multi-physics field parameters in combustion is provided, the device comprising:

[0040] An acquisition module is used to acquire the current physical environment parameters of the industrial equipment in real time, wherein the current physical environment parameters include thermal field parameters, fluid field parameters, electromagnetic field parameters and environmental parameters;

[0041] A prediction module is configured to input the current physical environment parameters into a preset target prediction model to obtain a predicted value corresponding to the current physical environment parameters, wherein the predicted value is used to indicate the combustion state corresponding to the current physical environment parameters, wherein the target prediction model is trained based on a sample set including: time series simulation data of multiple different physical fields under different combustion conditions, environmental parameters, and corresponding real values;

[0042] The determination module is used to input the predicted value into a preset mapping function to obtain the combustion performance of the corresponding industrial equipment.

[0043] According to a third aspect of an embodiment of the present application, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the method for predicting multi-physical field parameters in combustion according to the first aspect of an embodiment of the present application is implemented.

[0044] In a fourth aspect of an embodiment of the present application, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, it implements the method for predicting multi-physical field parameters in combustion described in the first aspect of the embodiment of the present application.

[0045] The beneficial effects of the technical solutions provided in the embodiments of the present application include at least:

[0046] The embodiment of the present application provides a method for predicting multi-physical field parameters in combustion, which obtains the current physical environment parameters of industrial equipment in real time, and the current physical environment parameters include: thermal field parameters, fluid field parameters, electromagnetic field parameters and environmental parameters; the current physical environment parameters are input into a preset target prediction model to obtain predicted values corresponding to the current physical environment parameters, and the predicted values are used to indicate the combustion state corresponding to the current physical environment parameters. The target prediction model is obtained based on sample set training, and the sample set includes: time series simulation data of multiple different physical fields under different combustion conditions, environmental parameters and corresponding real values trained; the predicted values are input into a preset mapping function to obtain the combustion performance of the corresponding industrial equipment, which can improve the accuracy and robustness of multi-physical field parameter prediction, and thus improve the intelligence and accuracy of combustion control of industrial equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A flowchart of a method for predicting multi-physics field parameters in combustion provided in an embodiment of the present application;

[0048] Figure 2 A structural diagram of a multi-physics field parameter prediction device for combustion provided in an embodiment of the present application;

[0049] Figure 3 A schematic diagram of the internal structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0051] In the following, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present disclosure, unless otherwise specified, "plurality" means two or more.

[0052] Additionally, the use of “based on” or “according to” is intended to be open and inclusive, in that a process, step, calculation, or other action “based on” or “according to” one or more conditions or values may, in practice, be based on additional conditions or beyond values.

[0053] With the increasing demand in the industrial field for coupled modeling and optimization of multiple physical fields (such as thermal fields, fluid fields, electromagnetic fields, etc.), accurate prediction and control of these physical field parameters are of great significance for improving equipment efficiency, reducing energy consumption, and ensuring system safety.

[0054] However, traditional modeling methods often face challenges such as high data dimension, strong feature diversity, and strong nonlinearity when dealing with the complex coupling relationship of multiple physical fields, which makes the prediction accuracy and computational efficiency unable to meet the needs of modern industrial scenarios.

[0055] Existing multi-physics modeling methods often only use information at a single scale, making it difficult to fully capture multi-level dynamic features. This application introduces a multi-expert network and multi-scale decomposition to effectively extract information from different time windows (different scales), and combines multi-granularity time series data to fully consider the impact of each scale on the final prediction results. The design of the multi-expert network can dynamically adjust the importance of interactions between features of different scales, ensuring that features of different granularities can be fully integrated, thereby improving the prediction accuracy of the model.

[0056] In addition, traditional time series prediction methods often ignore the frequency domain characteristics of time series, resulting in a decrease in model performance when faced with non-stationary data. By introducing frequency domain multi-scale modeling technology, this application can effectively process non-stationary signals in the frequency domain and use frequency domain transformation methods such as Fourier transform to extract the frequency components in the signal. This enables the model to better cope with non-stationary and complex fluctuating time series data, and enhances the model's adaptability to changes in multiple physical fields in complex industrial scenarios.

[0057] In practical industrial applications, multi-physics field data is often affected by noise and uncertainty, resulting in reduced prediction accuracy. Existing technologies typically rely on data cleaning and post-processing to address this issue. This application introduces a learnable frequency domain mask to effectively filter noise in the frequency domain and utilizes a noise regularization loss to sparsify frequency domain features, thereby improving the model's robustness and generalization capabilities.

[0058] At the same time, when processing information of different granularities, multi-expert networks often cause the weights of certain scales to be too large or too small, affecting the stability and accuracy of the model. To address this issue, this application introduces a balance loss term, which effectively balances the weights between features at different scales, preventing polarization and thus improving the performance of multi-expert networks in fusing data at different scales.

[0059] The deep learning approach, which combines a multi-expert network with frequency-domain multiscale modeling, not only handles complex multiphysics data but also enables efficient model training through a jointly designed loss function (including the primary task loss, balance loss, and noise regularization loss). While maintaining high prediction accuracy, it avoids the slow training and overfitting problems associated with excessive parameters in traditional methods.

[0060] Compared to existing technologies, this application combines the advantages of multi-expert networks and frequency-domain multi-scale modeling. It not only offers significant advantages in handling complex coupled relationships across multiple physical fields, but also better adapts to non-stationary data, improving the model's predictive accuracy, robustness, and generalization capabilities. This makes this method valuable for industrial applications, particularly in scenarios requiring high-precision, coupled multi-physics modeling, providing a more effective solution.

[0061] This application provides a method for predicting multi-physics parameters in combustion, such as Figure 1 As shown, the method includes the following steps:

[0062] Step 101: Acquire current physical environment parameters of industrial equipment in real time, where the current physical environment parameters include thermal field parameters, fluid field parameters, electromagnetic field parameters, and environmental parameters.

[0063] Among them, environmental parameters include: ambient temperature, ambient pressure and ambient humidity, etc.

[0064] Optionally, the process of obtaining the current physical environment parameters may be:

[0065] Thermal field parameters are obtained by collecting the thermal field distribution of the physical system using devices such as thermocouples, infrared thermometers, and thermal imagers. Actual fluid flow data is collected using devices such as flowmeters, pressure sensors, and particle image velocimetry (PIV) to obtain fluid field parameters. Electromagnetic field intensity data is obtained using devices such as fluxmeters, Hall effect sensors, and electric field detectors to obtain electromagnetic field parameters. Environmental condition data is collected using devices such as temperature and humidity sensors, barometers, and weather stations. Environmental parameters are obtained from meteorological departments or online APIs such as OpenWeatherMap.

[0066] Step 102: Input the current physical environment parameters into a preset target prediction model to obtain a prediction value corresponding to the current physical environment parameters, where the prediction value is used to indicate the combustion state corresponding to the current physical environment parameters. The target prediction model is trained based on a sample set, which includes: time series simulation data of multiple different physical fields under different combustion conditions, environmental parameters, and corresponding real values.

[0067] Step 103: Input the predicted value into a preset mapping function to obtain the corresponding combustion performance of the industrial equipment.

[0068] The embodiment of the present application provides a method for predicting multi-physical field parameters in combustion, which obtains the current physical environment parameters of industrial equipment in real time, and the current physical environment parameters include: thermal field parameters, fluid field parameters, electromagnetic field parameters and environmental parameters; the current physical environment parameters are input into a preset target prediction model to obtain predicted values corresponding to the current physical environment parameters, and the predicted values are used to indicate the combustion state corresponding to the current physical environment parameters. The target prediction model is obtained based on sample set training, and the sample set includes: time series simulation data of multiple different physical fields under different combustion conditions, environmental parameters and corresponding real values trained; the predicted values are input into a preset mapping function to obtain the combustion performance of the corresponding industrial equipment, which can improve the accuracy and robustness of multi-physical field parameter prediction, and thus improve the intelligence and accuracy of combustion control of industrial equipment.

[0069] Optionally, before obtaining the current physical environment parameters of the current industrial combustion process in real time, the method further includes:

[0070] obtaining the sample set;

[0071] Inputting the simulation data and environmental parameters in the sample set into the initial prediction model in sequence;

[0072] Using the initial prediction model to perform a training process on each input simulation data and the environmental parameters until the prediction error of the initial prediction model is less than a preset threshold, thereby obtaining the target prediction model;

[0073] Wherein, one of the training processes includes:

[0074] After extracting and fusing the features of the simulation data and the environmental parameters in the time series dimension, a time series fusion matrix is obtained;

[0075] After extracting and fusing frequency domain dimension features of the time series fusion matrix, a frequency domain fusion matrix is obtained;

[0076] Obtaining the predicted value according to the time series fusion matrix and the frequency domain fusion matrix;

[0077] The prediction error is obtained according to the predicted value and the true value corresponding to the predicted value.

[0078] Specifically, before extracting and fusing the time series features of the simulation data and the environmental parameters, the method further includes:

[0079] Connecting and flattening the simulation data and the environmental parameters to obtain an input vector;

[0080] Performing multi-scale decomposition on the input vector to obtain multiple subsequence vectors of different time scales;

[0081] Normalizing each of the subsequence vectors to obtain a plurality of corresponding target subvectors;

[0082] Correspondingly, the above-mentioned time-series feature extraction and fusion of the simulation data and the environmental parameters include:

[0083] Performing time-series feature extraction and fusion on the multiple target sub-vectors.

[0084] In the actual execution process, it is assumed that the simulation data and the environment parameters are expressed in matrix or tensor form, where:

[0085] X′ thermal Represents thermal field simulation data, X′ fluid represents the fluid field simulation data, X′ electromagnetic Represents electromagnetic field simulation data, X′ environment Represents environmental parameter data. The dimension of each data can be expressed as Where T i is the number of time steps, D i is the dimension of the data at each time step.

[0086] Thermal field simulation data can be generated using thermal field analysis software to generate thermal field distribution data. Fluid field simulation data can be simulated using fluid mechanics simulation software to obtain fluid flow velocity, pressure, eddy currents, etc. Electromagnetic field simulation data can be simulated using electromagnetic field simulation software to simulate electromagnetic field distribution.

[0087] The simulation data X′ of each physical field i and environmental parameter data X′ environment Concatenate and flatten to get the final input vector:

[0088] X all =Flatten(X′ thermal ,X′ fluid ,X′ electromagnetic ,X′ environment )

[0089] So, X all It is the input vector mentioned above, with dimension T total ×D total , where T total is the total number of time steps for all physics data, D total is the total dimension of all features.

[0090] Perform multi-scale decomposition on the input vector to extract features at different time scales. all Divide and generate multiple subsequence vectors of different time windows Where m represents different scales (window sizes), and each window size ω m Corresponding to different time granularities, wavelet transform, sliding window and other methods can be used for decomposition. The mathematical expression is:

[0091]

[0092] Among them, t i is the data index at time i, ranging from 1 to T (the total length of the time series), w m is the time window size corresponding to the mth scale, indicating the length of each subsequence, and M is the number of scales.

[0093] It can be understood that wavelet transform can decompose time series data into multiple frequency components and capture features at different frequencies (scales). Through wavelet transform, the time series will be decomposed into a low-frequency part and multiple high-frequency parts, each high-frequency part corresponds to a different time scale. Wavelet transform is often used to process non-stationary data because it can provide both time and frequency information. The sliding window method divides the sequence into multiple subsequences by sliding the window on the time series. The length of each subsequence is w m, which can cover the entire time series. For each window, its features are calculated and multi-scale information is obtained during the sliding process.

[0094] The process of normalizing each of the subsequence vectors to obtain a plurality of corresponding target subvectors may be:

[0095] In each subsequence vector, the Z-Score normalization formula is first used to normalize each dimension of the feature separately, and then all data are normalized to the maximum and minimum. This method can map the data to the interval [0,1] to obtain the target subvector corresponding to each subsequence vector. This ensures that the scales of different physical field simulation data and environmental parameter data are consistent, avoiding a certain feature from having a disproportionate impact on the model.

[0096] After obtaining the multi-scale decomposition and normalized target subvectors obtained through the aforementioned data preprocessing, a deep learning model combining a multi-expert network and frequency-domain multi-scale modeling was constructed. This model aims to integrate temporal features of different granularities and the interaction information between variables, and further enhances the model's ability to model non-stationary data through frequency-domain multi-scale modeling.

[0097] Optionally, the feature extraction and fusion of the time series dimension of the multiple target sub-vectors includes: using a first multi-layer perceptron to extract the time series features of each target sub-vector to obtain a corresponding first feature matrix; using a second multi-layer perceptron to extract the interaction information between each first feature matrix at different scales to obtain a second feature matrix; and performing weighted fusion on the second feature matrix to obtain the time series fusion matrix.

[0098] Assume the normalized target subvector is (data from scale m), whose size is w m ,Right now:

[0099]

[0100] here, represents the target sub-vector set at scale m, w m is the window size at that scale, and M represents the number of scales. These feature vectors will serve as the input of the multi-expert network.

[0101] The Multi-Expert Network extracts multi-granularity information by dynamically selecting and adjusting the importance of interactions between granularities at different scales. The design of the Multi-Expert Network includes the following key modules:

[0102] For each scale m, we extract the temporal features of each granularity through the first multi-layer perceptron (MLP). Given the input We obtain the corresponding time series feature representation through MLP:

[0103]

[0104] in, It represents the time series features extracted at scale m, that is, the first feature matrix, and MLP1 is the neural network layer used for time series feature extraction.

[0105] To extract the interaction information between different scales, we use a second MLP module and introduce a scaling factor α m , used to dynamically adjust the relative importance between different scales:

[0106]

[0107] in, is the feature representation between variables, that is, the second feature matrix, MLP2 is the neural network layer that extracts the interaction features between variables, and α m is the scaling factor, which indicates the importance of scales of different granularities when fusion occurs.

[0108] The features of all scales m are weighted and fused, and the weights are calculated by the Softmax function to represent the importance of each scale. We perform weighted fusion between granularities using the following formula:

[0109]

[0110] Here, h fusion It is the feature representation after fusion between granularities, that is, the time series fusion matrix. The Softmax operation is used to assign weights to each scale, and the sum of these weights is 1.

[0111] Optionally, after extracting and fusing the frequency domain dimension features of the time series fusion matrix, a frequency domain fusion matrix is obtained, including:

[0112] The time series fusion matrix is segmented into time series to obtain multiple sub-fusion matrices; the multiple sub-fusion matrices are converted into frequency domain to obtain multiple frequency domain matrices; the multiple frequency domain matrices are noise filtered to obtain multiple corresponding target frequency domain matrices; the frequency domain features of the multiple target frequency domain matrices are extracted using a third multi-layer perceptron to obtain multiple frequency domain feature matrices; the multiple frequency domain feature matrices are weightedly fused to obtain the frequency domain fusion matrix.

[0113] In actual implementation, the above process can be:

[0114] Introducing a learnable frequency domain mask M freq To filter out the noise component. The mask is learned by the network and multiplied with the frequency domain data to obtain a clear frequency domain feature:

[0115]

[0116] in, It is the frequency domain feature after noise filtering, that is, the target frequency domain matrix.

[0117] Finally, an MLP is used to extract frequency domain features and map them to the predicted frequency domain representation. It is the frequency domain data after noise filtering, and the frequency domain feature matrix The determination process is:

[0118]

[0119] The frequency domain features at all scales are weighted and fused to obtain the final frequency domain fusion matrix h freq_fusion Expressed as:

[0120]

[0121] Optionally, obtaining corresponding prediction values according to the time series fusion matrix and the frequency domain fusion matrix includes:

[0122] A fourth multi-layer perceptron is used to perform feature extraction on the time series fusion matrix and the frequency domain fusion matrix to obtain the predicted value.

[0123] The final output of the model is the predicted value Y of the multiphysics parameter pred , combining time series features and frequency domain features, the output form is as follows:

[0124] Y pred =MLP4(h fusion ,h freq_fusion )

[0125] The loss function of the model is designed as follows:

[0126] Main task loss (prediction error ):

[0127]

[0128] Among them, Y in the sample set true is the true value.

[0129] Balance loss:

[0130]

[0131] Noise regularization loss:

[0132]

[0133] Total loss:

[0134]

[0135] By minimizing the total loss, the model parameters are optimized to obtain high-accuracy multi-physics parameter predictions.

[0136] The current physical environment parameters output by the trained prediction model are combined with the control and optimization problems in actual application scenarios to achieve real-time control and system optimization.

[0137] Optionally, after inputting the predicted value into a preset mapping function to obtain the corresponding combustion performance of the industrial equipment, the method further includes:

[0138] Control information of the industrial equipment is generated based on the combustion performance.

[0139] In the actual execution process, the predicted value Y under the current physical environment parameters is predicted by the trained prediction model. pred After the predictions are made, the next step is to correlate the predicted values with the performance indicators in the actual application. By establishing a correlation model, we can map the predicted physical field parameters to the optimization objectives of the actual industrial process, such as the ignition efficiency and combustion performance of the boiler. Specifically, the quantitative relationship between the parameters and the performance indicators can be established through the following model:

[0140] P me t r i c =g map (Y pred )

[0141] Among them, P metric Indicates the performance indicators (such as ignition time, combustion heat value, etc.) obtained based on the predicted value, g map It is a mapping function used to indicate the relationship between the predicted value and the performance indicator. It is obtained by fitting the predicted value at multiple historical times with multiple corresponding performance indicators.

[0142] The performance indicators derived from this mapping model can help analyze and predict system operating conditions, providing decision support for actual production processes. By correlating the predicted values with the performance indicators, the system can adjust the actual system operating conditions through feedback control. This step automatically adjusts the operating parameters of industrial equipment such as boilers through real-time monitoring and optimized control algorithms.

[0143] For example, based on the predicted values, the burner power, fuel mixture ratio, and air flow rate are adjusted. The control input can be expressed by the following formula:

[0144] U contro l=h control (P metric )

[0145] Among them, U control It represents the output of the feedback control system, such as the adjusted igniter power, fuel ratio, etc. control It is the control function obtained by associating the model and the optimization algorithm.

[0146] Furthermore, to ensure the effectiveness of the control system, a real-time feedback mechanism is usually used to dynamically adjust the control strategy to cope with the nonlinear characteristics and uncertainties of the system. The real-time update of the control system can be achieved using the following update formula:

[0147] U control (t+1)=U control (t)+ΔU control (t)

[0148] Among them, U control (t) is the control input at time t, ΔU control (t) is the adjustment calculated by the control algorithm to optimize the system performance at the next moment.

[0149] In the real-time feedback control process, the introduction of optimization algorithms will further improve the control effect. For example, policy optimization based on deep reinforcement learning can dynamically adjust the operating conditions of the system to achieve optimal energy efficiency and performance. The optimization objective can be expressed as follows:

[0150]

[0151] Among them, J opt is the optimization objective function, P target is the desired performance indicator target, w1 and w2 are parameters for adjusting the weights in the optimization process, and T is the time interval considered in the optimization process.

[0152] The optimization process adjusts the control strategy based on real-time information provided by feedback systems and associated models to ensure that the system always maintains optimal operation under dynamically changing conditions.

[0153] Combining these steps ultimately creates a complete intelligent multi-physics prediction and optimization system. This system acquires multi-physics data in real time, performs deep learning predictions, maps the predictions to actual control objectives, and then adjusts system operating parameters through optimized control algorithms, thereby optimizing and improving system performance. This method has high application value and is particularly suitable for complex industrial scenarios, such as optimizing the control of systems like boilers and engines. It can significantly improve system efficiency, reduce energy consumption, and ensure safety.

[0154] The present application provides a device for predicting multi-physics field parameters in combustion, the device comprising:

[0155] An acquisition module 11 is used to acquire the current physical environment parameters of the industrial equipment in real time, wherein the current physical environment parameters include thermal field parameters, fluid field parameters, electromagnetic field parameters and environmental parameters;

[0156] A prediction module 12 is configured to input the current physical environment parameters into a preset target prediction model to obtain a prediction value corresponding to the current physical environment parameters, wherein the prediction value is used to indicate the combustion state corresponding to the current physical environment parameters. The target prediction model is trained based on a sample set, wherein the sample set includes: time series simulation data of multiple different physical fields under different combustion conditions, environmental parameters, and corresponding real values;

[0157] The determination module 13 is configured to input the predicted value into a preset mapping function to obtain the corresponding combustion performance of the industrial equipment.

[0158] In one embodiment, the apparatus further comprises a training module 14, wherein the training module 14 is configured to:

[0159] obtaining the sample set;

[0160] Inputting the simulation data and environmental parameters in the sample set into the initial prediction model in sequence;

[0161] Using the initial prediction model to perform a training process on each input simulation data and the environmental parameters until the prediction error of the initial prediction model is less than a preset threshold, thereby obtaining the target prediction model;

[0162] Wherein, one of the training processes includes:

[0163] After extracting and fusing the features of the simulation data and the environmental parameters in the time series dimension, a time series fusion matrix is obtained;

[0164] After extracting and fusing frequency domain dimension features of the time series fusion matrix, a frequency domain fusion matrix is obtained;

[0165] Obtaining the predicted value according to the time series fusion matrix and the frequency domain fusion matrix;

[0166] The prediction error is obtained according to the predicted value and the true value corresponding to the predicted value.

[0167] In one embodiment, the training module 14 is further configured to:

[0168] Connecting and flattening the simulation data and the environmental parameters to obtain an input vector;

[0169] Performing multi-scale decomposition on the input vector to obtain multiple subsequence vectors of different time scales;

[0170] Normalizing each of the subsequence vectors to obtain a plurality of corresponding target subvectors;

[0171] The training module 14 is specifically used to:

[0172] Performing time-series feature extraction and fusion on the multiple target sub-vectors.

[0173] The training module 14 is specifically used to:

[0174] After extracting the temporal features of each target sub-vector using a first multi-layer perceptron, a corresponding first feature matrix is obtained;

[0175] Using a second multi-layer perceptron, extract the interaction information between the first feature matrices at different scales to obtain a second feature matrix;

[0176] Perform weighted fusion on the second feature matrix to obtain the time series fusion matrix.

[0177] In one embodiment, the training module 14 is specifically configured to:

[0178] Performing time series segmentation on the time series fusion matrix to obtain multiple sub-fusion matrices;

[0179] Performing frequency domain conversion on the multiple sub-fusion matrices to obtain multiple frequency domain matrices;

[0180] performing noise filtering on the plurality of frequency domain matrices to obtain a plurality of corresponding target frequency domain matrices;

[0181] Using a third multi-layer perceptron, extracting frequency domain features of the plurality of target frequency domain matrices to obtain a plurality of frequency domain feature matrices;

[0182] A plurality of the frequency domain feature matrices are weightedly fused to obtain the frequency domain fusion matrix.

[0183] In one embodiment, the training module 14 is specifically configured to:

[0184] A fourth multi-layer perceptron is used to perform feature extraction on the time series fusion matrix and the frequency domain fusion matrix to obtain the predicted value.

[0185] In one embodiment, the determination module is further configured to generate control information of the industrial equipment according to the combustion performance.

[0186] The multi-physical field parameter prediction device in combustion provided in the embodiment of the present application can execute the above-mentioned multi-physical field parameter prediction method embodiment in combustion. Its implementation principle and technical effects are similar and will not be elaborated here.

[0187] The specific definitions of the multi-physics field parameter prediction device for combustion can be found in the definitions of the multi-physics field parameter prediction method for combustion above and will not be repeated here. The various modules in the above-mentioned multi-physics field parameter prediction device for combustion can be implemented in whole or in part through software, hardware, or a combination thereof. The above-mentioned modules can be embedded in or independent of the processor of the electronic device in hardware form, or can be stored in the memory of the electronic device in software form, so that the processor can call and execute the corresponding operations of each of the above modules.

[0188] The execution subject of the multi-physical field parameter prediction method in combustion provided in the embodiment of the present application can be an electronic device, which can be a server, a computer device, an industrial combustion equipment, and the industrial combustion equipment can include industrial boiler equipment and combustion equipment, etc.

[0189] Figure 3 This is a schematic diagram of the internal structure of an electronic device provided in an embodiment of the present application. Figure 3 As shown, the electronic device includes a processor and memory connected via a system bus. The processor is used to provide computing and control capabilities. The memory may include a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and a computer program. The computer program can be executed by the processor to implement the steps of the multi-physics field parameter prediction method for combustion provided in each of the above embodiments. The internal memory provides a cached operating environment for the operating system and computer program stored in the non-volatile storage medium.

[0190] Those skilled in the art will understand that Figure 3 The internal structure diagram of the electronic device shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the electronic device to which the solution of the present application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0191] In another embodiment of the present application, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method for predicting multi-physical field parameters in combustion in the embodiment of the present application are implemented.

[0192] In another embodiment of the present application, a computer program product is also provided, which includes computer instructions. When the computer instructions are run on the multi-physical field parameter prediction device in combustion, the multi-physical field parameter prediction device in combustion executes each step of the multi-physical field parameter prediction method in the method flow shown in the above method embodiment.

[0193] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using a software program, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer execution instructions are loaded and executed on a computer, the process or function according to the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website, computer, server or data center to another website, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more media that can be integrated. The available medium may be a magnetic medium (eg, a floppy disk, a hard disk, a magnetic tape), an optical medium (eg, a DVD), or a semiconductor medium (eg, a solid state disk (SSD)).

[0194] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0195] The above embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A method for predicting multi-physics field parameters in combustion, characterized in that: The method comprises: Real-time acquisition of current physical environment parameters of industrial equipment, including thermal field parameters, fluid field parameters, electromagnetic field parameters, and environmental parameters; Inputting the current physical environment parameters into a preset target prediction model to obtain a predicted value corresponding to the current physical environment parameters, wherein the predicted value is used to indicate the combustion state corresponding to the current physical environment parameters, wherein the target prediction model is trained based on a sample set, wherein the sample set includes: time series simulation data of multiple different physical fields under different combustion conditions, environmental parameters, and corresponding real values; The predicted value is input into a preset mapping function to obtain the combustion performance of the corresponding industrial equipment.

2. The method according to claim 1, characterized in that Before obtaining the current physical environment parameters of the current industrial combustion process in real time, the method further includes: obtaining the sample set; Inputting the simulation data and environmental parameters in the sample set into the initial prediction model in sequence; Using the initial prediction model to perform a training process on each input simulation data and the environmental parameters until the prediction error of the initial prediction model is less than a preset threshold, thereby obtaining the target prediction model; Wherein, one of the training processes includes: After extracting and fusing the features of the simulation data and the environmental parameters in the time series dimension, a time series fusion matrix is obtained; After extracting and fusing frequency domain dimension features of the time series fusion matrix, a frequency domain fusion matrix is obtained; Obtaining the predicted value according to the time series fusion matrix and the frequency domain fusion matrix; The prediction error is obtained according to the predicted value and the true value corresponding to the predicted value.

3. The method according to claim 2, characterized in that Before extracting and fusing the features of the simulation data and the environmental parameters in a time series dimension, the method further includes: Connecting and flattening the simulation data and the environmental parameters to obtain an input vector; Performing multi-scale decomposition on the input vector to obtain multiple subsequence vectors of different time scales; Normalizing each of the subsequence vectors to obtain a plurality of corresponding target subvectors; Extracting and fusing time-series features of the simulation data and the environmental parameters includes: Performing time-series feature extraction and fusion on the multiple target sub-vectors.

4. The method according to claim 3, characterized in that The extracting and fusing features of the plurality of target sub-vectors in the time series dimension includes: After extracting the temporal features of each target sub-vector using a first multi-layer perceptron, a corresponding first feature matrix is obtained; Using a second multi-layer perceptron, extract the interaction information between the first feature matrices at different scales to obtain a second feature matrix; Perform weighted fusion on the second feature matrix to obtain the time series fusion matrix.

5. The method according to claim 3, characterized in that After extracting and fusing the frequency domain dimension features of the time series fusion matrix, a frequency domain fusion matrix is obtained, including: Performing time series segmentation on the time series fusion matrix to obtain multiple sub-fusion matrices; Performing frequency domain conversion on the multiple sub-fusion matrices to obtain multiple frequency domain matrices; performing noise filtering on the plurality of frequency domain matrices to obtain a plurality of corresponding target frequency domain matrices; Using a third multi-layer perceptron, extracting frequency domain features of the plurality of target frequency domain matrices to obtain a plurality of frequency domain feature matrices; A plurality of the frequency domain feature matrices are weightedly fused to obtain the frequency domain fusion matrix.

6. The method according to claim 2, characterized in that The obtaining corresponding prediction values according to the time series fusion matrix and the frequency domain fusion matrix includes: A fourth multi-layer perceptron is used to perform feature extraction on the time series fusion matrix and the frequency domain fusion matrix to obtain the predicted value.

7. The method according to claim 1, characterized in that After inputting the predicted value into a preset mapping function to obtain the corresponding combustion performance of the industrial equipment, the method further includes: Control information of the industrial equipment is generated based on the combustion performance.

8. A device for predicting multi-physics field parameters in combustion, characterized in that: The device comprises: An acquisition module is used to acquire the current physical environment parameters of the industrial equipment in real time, wherein the current physical environment parameters include thermal field parameters, fluid field parameters, electromagnetic field parameters and environmental parameters; A prediction module is configured to input the current physical environment parameters into a preset target prediction model to obtain a predicted value corresponding to the current physical environment parameters, wherein the predicted value is used to indicate the combustion state corresponding to the current physical environment parameters, wherein the target prediction model is trained based on a sample set including: time series simulation data of multiple different physical fields under different combustion conditions, environmental parameters, and corresponding real values; The determination module is used to input the predicted value into a preset mapping function to obtain the combustion performance of the corresponding industrial equipment.

9. An electronic device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the method for predicting multi-physical field parameters in combustion according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that A computer program is stored thereon, and when the computer program is executed by a processor, the method for predicting multi-physical field parameters in combustion according to any one of claims 1 to 7 is implemented.

Images