Flame stability determination method and device

Abstract

The invention discloses a flame stability determination method and device. The method comprises the steps that physical field data related to the combustion process of target equipment is collected; extracting feature vectors of the temperature field data, the electric field data and the airflow field data to obtain a target temperature field feature vector, a target electric field feature vector and a target airflow field feature vector; combining the intra-group scatter matrix and the inter-group scatter matrix into a three-dimensional tensor comprising two channels through a dimension conversion layer of the first model, and compressing the high-dimensional feature data corresponding to the three-dimensional tensor into low-dimensional feature data; and determining a flame stability value of the target equipment according to the low-dimensional feature data through the multi-layer sensing layer of the first model. According to the scheme, multi-physical field information is fused, the first model is adopted for feature extraction, key data influencing flame stability can be effectively extracted, dimensionality reduction is carried out on high-dimensional complex data by optimizing the structure and the training method of the neural network, and the fidelity of key parameters in the dimensionality reduction process is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of stoves, and in particular to a method and device for determining flame stability. Background Art

[0002] Plasma cookers are highly efficient and environmentally friendly energy devices widely used in high-temperature heating, material processing, metallurgy, and other fields. During operation, flame stability directly impacts system efficiency and safety. Traditional flame stability control methods rely primarily on empirical rules and real-time monitoring data. However, these methods are limited in addressing complex multi-physics field interactions and cannot accurately reflect the system's dynamic changes and complex physical coupling relationships.

[0003] Although some studies have attempted to optimize plasma stoves using artificial intelligence and deep learning technologies, the following key issues remain:

[0004] 1. Complexity of high-dimensional data processing: The operation of a plasma cooker involves the coupling of multiple physical fields (such as temperature, electric, and airflow), resulting in high data dimensionality. Traditional dimensionality reduction methods often fail to effectively capture the complex relationships between these multiple physical fields, resulting in the loss of important physical properties or increased errors. While deep neural networks can process high-dimensional data, maintaining physical consistency and accuracy during feature extraction and dimensionality reduction remains a challenge.

[0005] 2. Deficiencies in multi-physics feature fusion: Existing models often fail to fully exploit the spatial-temporal coupling between different physical fields. Due to the varying spatiotemporal distributions of these fields, traditional feature fusion methods often fail to accurately capture this information. Consequently, the reduced low-dimensional features fail to effectively represent the key characteristics of the system, impacting the model's predictive accuracy.

[0006] 3. Accuracy of flame stability prediction: Existing prediction models for ignition success rate and flame stability often rely on simple linear relationships or empirical models when dealing with complex multi-physics coupling. Although neural networks such as multilayer perceptrons (MLPs) can model nonlinear relationships to a certain extent, avoiding overfitting and improving model generalization in high-dimensional feature spaces remain challenges. Summary of the Invention

[0007] The present invention aims to at least solve the technical problems existing in the prior art. To this end, the present invention provides a flame stability determination method in a first aspect, the method comprising:

[0008] Collecting physical field data related to the combustion process in the combustion chamber of the target device, the physical field data including temperature field data, electric field data, and airflow field data in the combustion chamber of the target device; the temperature field data, the electric field data, and the airflow field data are respectively the temperature value of each sampling point in the combustion chamber at each sampling time, the electric field intensity values in the x, y, and z directions, and the airflow velocity values in the x, y, and z directions; the target device is a device using plasma heating;

[0009] Extracting the characteristic vectors of the temperature field data, the electric field data, and the airflow field data respectively through the feature extraction layer of the first model to obtain a target temperature field characteristic vector, a target electric field characteristic vector, and a target airflow field characteristic vector;

[0010] Calculating the mean of each of the target temperature field eigenvectors, the target electric field eigenvectors, and the target airflow field eigenvectors through the matrix calculation layer of the first model, and determining the temperature field group scatter matrix, the electric field group scatter matrix, and the airflow field group scatter matrix using the mean values, respectively;

[0011] Determining, through the matrix calculation layer, a joint eigenvector of each sampling point according to the target temperature field eigenvector, the target electric field eigenvector, and the target airflow field characteristic corresponding to each sampling point, and determining an inter-group scatter matrix of the sampling points according to the mean of the joint eigenvector;

[0012] Combining the intra-group scatter matrix and the inter-group scatter matrix into a three-dimensional tensor including two channels through a dimensionality conversion layer of the first model, and compressing high-dimensional feature data corresponding to the three-dimensional tensor into low-dimensional feature data;

[0013] The flame stability value of the target device is determined according to the low-dimensional feature data through the multi-layer perception layer of the first model.

[0014] Optionally, respectively extracting feature vectors of the temperature field data, the electric field data, and the airflow field data includes:

[0015] Applying three-dimensional convolution to the temperature values to extract spatial correlation features of the temperature values in a local area, and using a time series model to extract hidden features of the spatial correlation features at the current sampling moment, and concatenating the hidden features at each sampling moment to obtain a target temperature field feature vector;

[0016] Performing a weighted summation on the electric field intensity values in the x, y, and z directions to obtain a fused electric field eigenvalue; inputting the electric field eigenvalue into a multi-channel convolutional network to obtain a convolved feature tensor; flattening the feature tensor into a global vector along the spatial dimension to obtain a target electric field eigenvector;

[0017] The airflow velocity values in the x, y, and z directions are weighted and summed to obtain a fused airflow field eigenvalue; the airflow field eigenvalue is fused with the target electric field eigenvector to obtain a joint eigenvalue; the joint eigenvalue is input into the multi-channel convolutional network, and the high-order features of the joint features are extracted to obtain a joint high-order eigenvalue; the joint high-order eigenvalue is flattened into a global vector to obtain a target airflow field eigenvector.

[0018] Optionally, the respectively calculating the mean of each of the target temperature field eigenvectors, the target electric field eigenvectors, and the target airflow field eigenvectors, and respectively using the mean to determine the temperature field group scatter matrix, the electric field group scatter matrix, and the airflow field group scatter matrix, comprises:

[0019] Calculating the mean of each of the target temperature field characteristic vectors, the target electric field characteristic vectors, and the target airflow field characteristic vectors respectively to obtain a temperature field mean vector, an electric field mean vector, and an airflow field mean vector;

[0020] respectively calculating the difference vectors between each target temperature field characteristic vector and the temperature field mean vector, and forming a temperature field group scatter matrix from each difference vector;

[0021] respectively calculating the difference vectors between each target electric field characteristic vector and the electric field mean vector, and forming an electric field group scatter matrix from each difference vector;

[0022] The difference vectors between each target airflow field characteristic vector and the airflow field mean vector are calculated respectively, and the airflow field group scatter matrix is composed of each difference vector.

[0023] Optionally, determining a joint eigenvector of each sampling point according to the target temperature field eigenvector, the target electric field eigenvector, and the target airflow field characteristic corresponding to each sampling point, and determining an inter-group scatter matrix of the sampling points according to a mean of the joint eigenvector includes:

[0024] splicing the target temperature field feature vector, the target electric field feature vector, and the target airflow field feature vector corresponding to each sampling point to obtain a joint feature vector for each sampling point;

[0025] Calculating the mean of the joint eigenvectors of each of the sampling points to obtain a total mean vector;

[0026] The difference between the temperature field mean vector and the total mean vector, the difference between the electric field mean vector and the total mean vector, and the difference between the airflow field mean vector and the total mean vector are calculated respectively to obtain the inter-group scatter matrix of the sampling points.

[0027] Optionally, merging the intra-group scatter matrix and the inter-group scatter matrix into a three-dimensional tensor including two channels, and compressing high-dimensional feature data corresponding to the three-dimensional tensor into low-dimensional feature data, includes:

[0028] Merging the intra-group scatter matrix and the inter-group scatter matrix into a three-dimensional tensor with two channels through a channel expansion operation;

[0029] Performing a first convolution operation on the three-dimensional tensor to learn a local feature relationship between the inter-group scatter matrix and the intra-group scatter matrix to obtain a first feature map;

[0030] Performing a second-layer convolution operation on the first feature map to learn high-order features of the first feature map to obtain a second-layer feature map;

[0031] A global average pooling operation and a full connection operation are performed on the second-layer feature map to compress the high-dimensional features of the second-layer feature map into low-dimensional features.

[0032] Optionally, the loss function of the first model during training is:

[0033]

[0034] Among them, L contrastive Represents the loss value, Y pred (i) and Y pred (j) represents the prediction results of the first model at time i and time j, respectively, ||Y pred (i)-Y pred (j)|| represents Y pred (i) and Y pred (j) The Euclidean distance, m represents the preset boundary value.

[0035] Optionally, after determining the flame stability of the plasma-heating device according to the low-dimensional feature data, the method further includes:

[0036] If the flame stability value is less than a preset safety threshold, the voltage value and the current value are adjusted.

[0037] A second aspect of the present invention provides a flame stability determination device, comprising:

[0038] a data acquisition module for acquiring physical field data related to the combustion process in the combustion chamber of the target device, the physical field data including temperature field data, electric field data, and airflow field data in the combustion chamber of the target device; the temperature field data, the electric field data, and the airflow field data are respectively the temperature value of each sampling point in the combustion chamber at each sampling moment, the electric field intensity values in the x, y, and z directions, and the airflow velocity values in the x, y, and z directions; the target device is a device using plasma heating;

[0039] a feature extraction module, configured to extract the feature vectors of the temperature field data, the electric field data, and the airflow field data respectively through the feature extraction layer of the first model, and obtain a target temperature field feature vector, a target electric field feature vector, and a target airflow field feature vector;

[0040] an intra-group matrix determination module, configured to calculate the mean of each of the target temperature field eigenvectors, the target electric field eigenvectors, and the target airflow field eigenvectors through the matrix calculation layer of the first model, and to determine the temperature field intra-group scatter matrix, the electric field intra-group scatter matrix, and the airflow field intra-group scatter matrix using the mean values;

[0041] an inter-group matrix determination module, configured to determine, through the matrix calculation layer, a joint eigenvector of each sampling point according to the target temperature field eigenvector, the target electric field eigenvector, and the target airflow field characteristic corresponding to each sampling point, and determine an inter-group scatter matrix of the sampling points according to the mean of the joint eigenvector;

[0042] a compression module, configured to combine the intra-group scatter matrix and the inter-group scatter matrix into a three-dimensional tensor including two channels through a dimensionality conversion layer of the first model, and compress high-dimensional feature data corresponding to the three-dimensional tensor into low-dimensional feature data;

[0043] A flame stability determination module is used to determine the flame stability value of the target device according to the low-dimensional feature data through the multi-layer perception layer of the first model.

[0044] A third aspect of the present invention provides a device for plasma heating, comprising:

[0045] Controller;

[0046] a memory for storing instructions executable by the controller;

[0047] Wherein, the controller is configured to execute the instructions to implement the flame stability determination method as described in the first aspect.

[0048] A fourth aspect of the present invention provides a computer-readable storage medium. When instructions in the computer-readable storage medium are executed by a controller of a device using plasma heating, the device using plasma heating can perform the flame stability determination method described in the first aspect.

[0049] The embodiments of the present invention have the following beneficial effects:

[0050] In an embodiment of the present invention, physical field data related to the combustion process in the combustion chamber of the target device are collected, and the physical field data include temperature field data, electric field data and airflow field data in the combustion chamber of the target device; through the feature extraction layer of the first model, the characteristic vectors of the temperature field data, the electric field data and the airflow field data are extracted respectively to obtain the target temperature field characteristic vector, the target electric field characteristic vector and the target airflow field characteristic vector; through the matrix calculation layer of the first model, the mean of each target temperature field characteristic vector, the target electric field characteristic vector and the target airflow field characteristic vector are calculated respectively, and the mean is used to determine the scatter matrix within the temperature field group, the electric field group and the target airflow field characteristic vector respectively. The internal scatter matrix and the airflow field group scatter matrix; through the matrix calculation layer, the joint feature vector of each sampling point is determined according to the target temperature field feature vector, the target electric field feature vector and the target airflow field feature corresponding to each sampling point, and the inter-group scatter matrix of the sampling point is determined according to the mean of the joint feature vector; through the dimensionality conversion layer of the first model, the internal scatter matrix and the inter-group scatter matrix are merged into a three-dimensional tensor including two channels, and the high-dimensional feature data corresponding to the three-dimensional tensor is compressed into low-dimensional feature data; through the multi-layer perception layer of the first model, the flame stability value of the target device is determined according to the low-dimensional feature data. This solution can effectively extract the key data affecting flame stability by fusing the information of multiple physical fields such as temperature field, electric field, and airflow field, and using the first model for feature extraction, and by optimizing the structure and training method of the neural network, the high-dimensional complex data is reduced in dimensionality to ensure the fidelity of the dimensionality reduction process to key parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a flowchart of a method for determining flame stability provided by an embodiment of the present invention;

[0052] Figure 2 This is a structural block diagram of a flame stability determination device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0053] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0054] In the following, the terms "first" and "second" are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, the features defined as "first" and "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, "multiple" means two or more. In addition, the use of "based on" or "according to" means openness and inclusiveness, because the process, steps, calculations or other actions "based on" or "according to" one or more of the conditions or values may be based on additional conditions or values beyond the stated in practice.

[0055] Figure 1 This is a flowchart of the steps of a method for determining flame stability provided by an embodiment of the present invention.

[0056] like Figure 1 As shown, the method includes the following steps:

[0057] Step 101: Collect physical field data related to the combustion process in the combustion chamber of the target device, wherein the physical field data includes temperature field data, electric field data, and airflow field data in the combustion chamber of the device using plasma heating; the temperature field data, the electric field data, and the airflow field data are respectively the temperature value of each sampling point in the combustion chamber at each sampling moment, the electric field strength values in the x, y, and z directions, and the airflow velocity values in the x, y, and z directions; the target device is a device using plasma heating.

[0058] In the embodiments of the invention, the target device is a device that uses plasma heating. Plasma-heating devices refer to devices that use the plasma heating principle as a heat source. Such devices include, but are not limited to, stoves, electric heaters, wall-mounted boilers, electric ovens, portable stoves, and the like. Plasma heating is based on the principle that high voltage at the end of the plasma breaks through air to generate an open flame, which then heats the device.

[0059] First, multi-physics field data related to the combustion process is collected within the target device's combustion chamber, including temperature, electric, and airflow fields. A complete multi-physics model is established through physical experiments and simulation tools, providing foundational data for subsequent simplification and optimization.

[0060] During combustion, the temperature field within the combustion chamber exhibits highly nonlinear variations, with complex temporal and spatial distributions. Consequently, these physical field data typically vary in three dimensions (x, y, z) and time, resulting in high-dimensional data. For example, the temperature field may vary over time or based on spatial location.

[0061] During plasma ignition, changes in the electric field intensity directly affect the ionization rate of the airflow, thereby altering the local temperature distribution. Consequently, there is a strong physical coupling between the temperature field, the electric field, and the airflow field. Changes in the electric field affect the airflow distribution, which in turn alters the local temperature field. This interdependence complicates feature extraction and analysis.

[0062] The airflow field within a combustion chamber may contain thousands of sampling points, each corresponding to velocity components in the x, y, and z directions, with a total feature dimension of up to 3 x 1000. Therefore, each physical field typically contains millions or even more sampling points, resulting in an extremely large number of features. Dimensionality reduction requires extracting key information from complex, high-dimensional data while ensuring physical consistency.

[0063] The temperature field data is the temperature value of each sampling point in the combustion chamber at each sampling moment, which can be expressed as T(x, y, z, t). It reflects the temperature distribution at different positions in the combustion chamber and the change of temperature over time, describing the heat conduction and combustion dynamics during the combustion process.

[0064] The electric field data is the electric field intensity value in the x, y, and z directions at each sampling point at each sampling moment, which can be expressed as E(x, y, z, t). The three directional components of the electric field are E x ,E y ,E z , which represents the directionality in space, and the component of the electric field intensity that changes with time is E(t), which reflects the dynamic behavior of the electric field.

[0065] The airflow field data is the airflow velocity value in the x, y, and z directions at each sampling point at each sampling moment, which can be expressed as V(x, y, z, t). The three directional components of the airflow are v x ,v y ,v z , which represents the velocity direction and magnitude of the airflow. The time-varying component of the airflow is v(t), which describes the transient behavior of the fluid motion.

[0066] Step 102: extract the characteristic vectors of the temperature field data, the electric field data and the airflow field data respectively through the feature extraction layer of the first model to obtain a target temperature field characteristic vector, a target electric field characteristic vector and a target airflow field characteristic vector.

[0067] The first model in this solution is a model obtained through training for converting high-dimensional complex physical field data into low-dimensional data. The first model includes a feature extraction layer for extracting features from multi-physical field data; a matrix calculation layer of the first model for calculating the intra-group scatter matrix and the inter-group scatter matrix; a dimension conversion layer of the first model for merging the intra-group scatter matrix and the inter-group scatter into a three-dimensional tensor including two channels, and compressing the high-dimensional feature data corresponding to the three-dimensional tensor into low-dimensional feature data.

[0068] The inputs to the first model are the temperature field data T(x, y, z, t), electric field data E(x, y, z, t), and airflow field data V(x, y, z, t) in step 101. The outputs of the feature extraction layer are the target temperature field feature vector, target electric field feature vector, and target airflow field feature vector.

[0069] As an optional embodiment, step 102 includes:

[0070] Step 1021: Apply three-dimensional convolution to the temperature values to extract spatial correlation features of the temperature values in the local area, and use a time series model to extract hidden features of the spatial correlation features at the current sampling moment, and splice the hidden features at each sampling moment to obtain a target temperature field feature vector;

[0071] Apply three-dimensional convolution to the temperature value T(x, y, z) at each moment to extract the spatial correlation features in the local area. The specific extraction method is as follows:

[0072] T s ′=σ(W T *T+b T )

[0073] Where T represents the input temperature field tensor T(x,y,z), W T represents the convolution kernel, which is used to learn the association between spatial positions, b T represents the bias, σ represents the activation function (such as ReLU function), which is used to introduce nonlinear features, T s ′ represents the target temperature field feature vector output after convolution.

[0074] LSTM (Long Short-Term Memory) is used to capture the dynamic changes of the temperature field over time. The specific method is as follows:

[0075] h t =LSTM(T s ′(t),h t-1 )

[0076] Among them, T s ′ represents the features of each time step after convolution, ht represents the hidden state of the current time step, h t-1 represents the hidden state at the previous time step.

[0077] The last hidden state h of the time series model t Summarized as the dynamic characteristics of the temperature field:

[0078] T′=Concat(h1,h2,…,h T )

[0079] Wherein, T' represents the temperature field output vector, that is, the target temperature field feature vector.

[0080] Step 1022: perform weighted summation on the electric field intensity values in the x, y, and z directions to obtain a fused electric field eigenvalue; input the electric field eigenvalue into a multi-channel convolutional network to obtain a convolved feature tensor; flatten the feature tensor into a global vector along the spatial dimension to obtain a target electric field eigenvector.

[0081] Specifically, for the three directional components of the electric field E x ,E y ,E z Perform weighted summation, specifically:

[0082] E′ i =α x E i,x +α y E i,y +α z E i,z

[0083] Among them, E i,x ,E i,y ,E i,z Represents the components of the electric field in the x, y, and z directions. x ,α y ′,α z (α x +α y +α z =1) represents the weighting coefficient, which is used to adjust the contribution of the three directional components of the electric field. i Represents the electric field characteristic value after fusion.

[0084] The fused electric field eigenvalue E′ i Enter the multi-channel convolutional network, specifically:

[0085] E′ s =σ(W E *E′ i +b E )

[0086] Among them, WE Represents the convolution kernel, which is used to extract the local pattern of electric field features in space. i Represents the characteristic value of the electric field after fusion. b E Represents the convolution bias term, which is used to adjust the convolution result. σ represents the activation function (such as ReLU), which is used to introduce nonlinear characteristics and enhance the network expression ability. E′ s Represents the feature tensor after convolution.

[0087] E′ s Flattening along the spatial dimension into a global vector, we get the global features, specifically:

[0088] E′=Flatten(E′ s )

[0089] Flatten represents a flattening operation, which converts a high-dimensional tensor into a one-dimensional vector. E' represents the output vector of the electric field layer, i.e., the target electric field feature vector.

[0090] Step 1023: perform weighted summation on the airflow velocity values in the x, y, and z directions to obtain a fused airflow field eigenvalue; fuse the airflow field eigenvalue with the target electric field eigenvector to obtain a joint eigenvalue; input the joint eigenvalue into the multi-channel convolutional network, extract the high-order features of the joint features, and obtain a joint high-order eigenvalue; flatten the joint high-order eigenvalue into a global vector to obtain a target airflow field eigenvector.

[0091] The airflow velocity values in the x, y, and z directions are weighted and summed to obtain the fused airflow field characteristic value, specifically:

[0092] V i ′=β x v i,x +β y v i,y +β z v i,z

[0093] Among them, v i,x ,v i,y ,v i,z represents the velocity component of the airflow field V(x,y,z,t) in the (x,y,z) directions, β x ,β y ,β z (β x +β y +β z =1) represents the weighting coefficient, which is used to adjust the contribution of airflow in the x, y, and z directions. i Represents the characteristics of the airflow field after fusion.

[0094] The airflow field characteristic value V′ i Combined with the target electric field eigenvector E', a joint high-order eigenvalue is formed, specifically:

[0095] V′ coupled =Concat(V′ i ,E′)

[0096] Among them, V′ coupled Represents the joint features of airflow and electric field, that is, the joint high-order eigenvalue. Concat represents the feature concatenation operation.

[0097] Use multi-layer convolutional networks to further extract coupling features:

[0098] V′ s =σ(W V *V′ coupled +b V )

[0099] Among them, W V Represents the convolution kernel, which is used to learn the coupling relationship between airflow and electric field features. coupled Represents the joint eigenvalue of the airflow field and electric field characteristics. b V Represents the convolution bias term. σ represents the activation function, which means that the convolution output introduces nonlinear features. V′ s Represents the feature tensor after convolution, that is, the joint high-order eigenvalue.

[0100] V′ s Flatten along the spatial dimension, specifically:

[0101] V′=Flatten(V′ s )

[0102] Where V' represents the output target airflow field feature vector. Flatten represents a flattening operation, which is used to convert a multi-dimensional tensor into a one-dimensional vector.

[0103] Step 103: Calculate the mean of each of the target temperature field eigenvectors, the target electric field eigenvectors, and the target airflow field eigenvectors through the matrix calculation layer of the first model, and use the mean values to determine the temperature field group scatter matrix, the electric field group scatter matrix, and the airflow field group scatter matrix.

[0104] The degree of dispersion of data of different categories can be evaluated by calculating the within-group scatter matrix (Within-Group ScatterMatrix) and the between-group scatter matrix (Between-Group Scatter Matrix).

[0105] The within-group scatter matrix indicates the degree of dispersion of the data within each category. If the within-group scatter matrix is large, it means that the differences between the data in the same group are large, and the data are more dispersed within each category.

[0106] As an optional embodiment, step 103 includes:

[0107] Step 1031: Calculate the mean of each of the target temperature field characteristic vectors, the target electric field characteristic vectors, and the target airflow field characteristic vectors to obtain a temperature field mean vector, an electric field mean vector, and an airflow field mean vector;

[0108] Step 1032: Calculate the difference vectors between each target temperature field feature vector and the temperature field mean vector, and form a temperature field group scatter matrix based on the difference vectors.

[0109] Step 1033: Calculate the difference vectors between each target electric field feature vector and the electric field mean vector, and form an electric field group scatter matrix based on the difference vectors.

[0110] Step 1034: Calculate the difference vectors between each target airflow field feature vector and the airflow field mean vector respectively, and form an airflow field group scatter matrix based on each difference vector.

[0111] In steps 1031 to 1034, the intra-group scatter matrix describes the distribution of samples of each category relative to its mean. The calculation method of the intra-group scatter matrix is:

[0112]

[0113] Among them, S W Represents the intra-group scatter matrix. C represents the total number of categories. For example, temperature field, electric field, and airflow field are three categories, and the total number of categories is 3. j Represents the feature vector of the jth sample. i Represents the mean vector of category Ci. x′ j -μ x Represents sample x′ j and the category mean μ i The difference vector of (x′ j -μ i )(x′ j -μ i ) T Represents the outer product, which is used to calculate the dispersion contribution of a single sample relative to the deviation of the class center.

[0114] Step 104: Determine the joint eigenvector of each sampling point through the matrix calculation layer according to the target temperature field eigenvector, the target electric field eigenvector, and the target airflow field characteristics corresponding to each sampling point, and determine the inter-group scatter matrix of the sampling points according to the mean of the joint eigenvector.

[0115] The inter-group scatter matrix indicates the degree of dispersion of data between different categories. If the inter-group scatter matrix is large, it means that the data between different categories are very different and the data are distributed relatively far between categories.

[0116] The between-group scatter matrix describes the distribution between category means.

[0117] As an optional embodiment, step 104 includes:

[0118] Step 1041: Concatenate the target temperature field feature vector, the target electric field feature vector, and the target airflow field feature vector corresponding to each sampling point to obtain a joint feature vector for each sampling point;

[0119] Step 1042: Calculate the mean of the joint feature vectors of the sampling points to obtain a total mean vector;

[0120] Step 1043: Calculate the difference between the temperature field mean vector and the total mean vector, the difference between the electric field mean vector and the total mean vector, and the difference between the airflow field mean vector and the total mean vector to obtain the inter-group scatter matrix of the sampling points.

[0121] In steps 1041 to 1043, the inter-group scatter matrix is calculated as follows:

[0122]

[0123] Where: S B represents the between-group scatter matrix. C represents the total number of categories. N i Indicates the number of samples of category Ci. μ i represents the mean vector of category Ci. μ represents the total mean vector.

[0124] Among them, the calculation method of the total mean vector is:

[0125]

[0126] Where N represents the total number of samples. j represents the joint eigenvector.

[0127] Step 105: Through the dimensionality conversion layer of the first model, the intra-group scatter matrix and the inter-group scatter are merged into a three-dimensional tensor including two channels, and the high-dimensional feature data corresponding to the three-dimensional tensor is compressed into low-dimensional feature data.

[0128] First, a channel expansion operation is used to combine the intra-group scatter matrix and the inter-group scatter matrix into a three-dimensional tensor consisting of two channels, which is convenient for processing by the multi-layer perception layer. Then, a global pooling and fully connected layer are used to compress the high-dimensional feature data corresponding to the three-dimensional tensor into low-dimensional feature data.

[0129] As an optional embodiment, step 105 includes:

[0130] Step 1051: Combine the intra-group scatter matrix and the inter-group scatter matrix into a three-dimensional tensor with two channels through a channel expansion operation.

[0131] This step is expressed as follows:

[0132] Input Tensor=Stack(S W ,S B )∈R d′×d′×2

[0133] Among them, S W Represents the intra-group scatter matrix, which indicates the compactness of the distribution of samples within the same category, S B represents the inter-group scatter matrix, which indicates the degree of separation between different categories. Stack represents the channel expansion operation. d' represents the feature dimension.

[0134] S W and S B Merge into a 3D tensor with 2 channels to facilitate processing by multi-layer perception layers.

[0135] Step 1052: Perform a first layer convolution operation on the three-dimensional tensor to learn the local feature relationship between the inter-group scatter matrix and the intra-group scatter matrix to obtain a first feature map.

[0136] The first layer of convolution operation learns the local feature relationship between the intra-group scatter matrix and the inter-group scatter matrix, which can be expressed as:

[0137] H1=σ(W1*Input Tensor+b1)

[0138] Where H1 represents the first-layer feature map, W1 represents the first-layer convolution kernel, and σ represents the activation function. Input Tensor represents the input tensor, i.e., the three-dimensional tensor after the merger in step 1052. b1 represents the bias of the first convolution layer.

[0139] Step 1053: Perform a second-layer convolution operation on the first feature map to learn high-order features of the first feature map to obtain a second-layer feature map.

[0140] The second layer of convolution operation further learns the high-order feature relationship, which can be expressed as:

[0141] H2=σ(W2*H1+b2)

[0142] Among them, H2 represents the second layer feature map, H1 represents the first layer feature map, W2 represents the second layer convolution kernel, and b2 represents the bias of the second layer convolution layer.

[0143] Step 1054: Perform a global average pooling operation and a full connection operation on the second-layer feature map to compress the high-dimensional features of the second-layer feature map into low-dimensional features.

[0144] Global Average Pooling (GAP) is used to compress high-dimensional features into low-dimensional features, which can be expressed as follows:

[0145] Z=GAP(H2)

[0146] GAP represents the global average pooling operation, which is used to compress the spatial features of each channel of H2 into a single value to form a global feature vector. H2 represents the second-layer feature map, and Z represents the feature vector after pooling.

[0147] This step compresses the high-dimensional feature map H2 through global pooling to obtain low-dimensional global features.

[0148] The fully connected layer further compresses the feature dimensions of the low-dimensional global features to generate a low-dimensional representation. It can be expressed as:

[0149] Y=σ(W fc Z+b fc )

[0150] Among them, W fc represents the weight of the fully connected layer, b fc represents the bias term, and Y represents the low-dimensional feature.

[0151] Step 106: Determine the flame stability value of the target device according to the low-dimensional feature data through the multi-layer perception layer of the first model.

[0152] Assuming that flame stability S is a function of multiple physical factors, these factors can be represented by the low-dimensional features Y(t) compressed from the high-dimensional features by the deep neural network in the previous step. A regression model based on a multi-layer perceptron can be used to describe the relationship between flame stability and low-dimensional features:

[0153] S(t)=g(Y(t),ω)

[0154] Where S(t)∈[0,1] represents flame stability, g(·) is a 3-layer multilayer perceptron model, Y(t) represents low-dimensional features, and ω represents the weight of the multilayer perceptron.

[0155] The low-dimensional feature data Y(t) in step 105 is input into the multi-layer perception layer to obtain the flame stability value S(t) of the target device.

[0156] As an optional embodiment, after step 106, the method further includes:

[0157] Step 107: If the flame stability value is less than a preset safety threshold, adjust the voltage and current values.

[0158] Predetermine the safety threshold S of flame stability safe , such as a lower limit of a stability value, such as 0.5. If the flame stability is lower than the safety threshold, it means that the flame is unstable, which may lead to reduced efficiency or safety hazards.

[0159] The stability threshold judgment result can indicate whether the operating parameters of the stove need to be adjusted. If the flame stability S(t) is lower than the safety threshold S safe , you can adjust the operating parameters, such as current, voltage, air flow speed, etc.

[0160] It can also be calculated based on the flame stability S(t) and the stability threshold S safe , optimize the stove's control parameters. For example, increase the current or adjust the air flow rate to improve flame stability.

[0161] In addition, based on energy-saving considerations, if the flame stability is too high and the operation is at unnecessarily high power, the current or airflow can be appropriately reduced to save energy and improve efficiency.

[0162] As an optional embodiment, the loss function of the first model during training is:

[0163]

[0164] Among them, L contrastive Represents the loss value, Y pred (i) and Y pred (j) represents the prediction results of the first model at time i and time j, respectively, ||Y pred (i)-Y pred (j)|| represents Y pred (i) and Y pred (j) The Euclidean distance, m represents the preset boundary value.

[0165] In this embodiment of the present invention, the goal of training the first model is to maximize the separation of the inter-group scatter matrix and minimize the compactness of the intra-group scatter matrix by optimizing the network parameters, while retaining the core information of the input features. The entire training process includes the following steps:

[0166] S1. Data preparation: Stack the dataset as the input tensor of the network.

[0167] S2. Forward propagation: Passing forward through each layer of the neural network, finally compressing high-dimensional features through the convolutional neural network, and generating low-dimensional feature representations through the fully connected layer.

[0168] S3. Loss calculation: Calculate the error of the current network based on the designed loss function.

[0169] S4. Backpropagation and optimization: Update network parameters, including convolution kernels, biases, and fully connected layer weights, through gradient descent or optimization algorithms (such as Adam).

[0170] S5. Validation and Adjustment: Evaluate network performance on the validation set, observe the classification ability or separation effect of low-dimensional feature representation, and adjust hyperparameters to improve the generalization ability of the model.

[0171] S6. Iterative training: Repeat the above steps until the network loss function converges or the set training termination condition is reached.

[0172] For the design of the loss function of the first model, we can use a contrast loss function. The output of the first model is a low-dimensional representation that integrates multi-physics field information such as temperature, electric field, and airflow. In order to ensure that similar time series are close to each other in the low-dimensional space, we design a contrast loss function. The details are as follows:

[0173]

[0174] Among them, L contrastive Represents the loss value, Y pred (i) and Y pred (j) represents the prediction results of the first model at time i and time j respectively. These results are low-dimensional global feature vectors that contain joint information of multiple physical fields. pred (i)-Y pred (j)|| represents Y pred (i) and Y pred The Euclidean distance of (j) is used to measure the similarity or difference between the two. m is a preset "boundary" value. Usually, m is set to an appropriate hyperparameter based on the scale and physical background of the data.

[0175] max(0,m-||Ypred (i)-Y pred (j) represents a larger value. The max function ensures that the loss will only increase when the distance is less than m.

[0176] In summary, the flame stability determination method provided by the embodiment of the present invention collects physical field data related to the combustion process in the combustion chamber of the target device, and the physical field data includes temperature field data, electric field data and airflow field data in the combustion chamber of the target device; through the feature extraction layer of the first model, the characteristic vectors of the temperature field data, the electric field data and the airflow field data are extracted respectively to obtain the target temperature field characteristic vector, the target electric field characteristic vector and the target airflow field characteristic vector; through the matrix calculation layer of the first model, the mean of each target temperature field characteristic vector, the target electric field characteristic vector and the target airflow field characteristic vector are calculated respectively, and the mean values are used to determine the dispersion within the temperature field group respectively. The invention relates to a method for determining the flame stability of a target device by using a multi-layer perception layer. The method comprises the following steps: first, a distribution matrix, an intra-group scatter matrix of an electric field, and an intra-group scatter matrix of an airflow field; through the matrix calculation layer, the joint feature vector of each sampling point is determined based on the target temperature field feature vector, the target electric field feature vector, and the target airflow field feature corresponding to each sampling point, and the inter-group scatter matrix of the sampling point is determined based on the mean of the joint feature vector; through the dimensionality conversion layer of the first model, the intra-group scatter matrix and the inter-group scatter matrix are merged into a three-dimensional tensor including two channels, and the high-dimensional feature data corresponding to the three-dimensional tensor are compressed into low-dimensional feature data; through the multi-layer perception layer of the first model, the flame stability value of the target device is determined based on the low-dimensional feature data. This solution can effectively extract key data affecting flame stability by fusing multi-physical field information such as temperature field, electric field, and airflow field, and using the first model for feature extraction. Furthermore, by optimizing the structure and training method of the neural network, the high-dimensional complex data is reduced in dimensionality to ensure the fidelity of the dimensionality reduction process to key parameters.

[0177] Figure 2 This is a structural block diagram of a flame stability determination device provided by an embodiment of the present invention. Figure 2 As shown, the device 200 includes:

[0178] A data acquisition module 201 is configured to acquire physical field data related to the combustion process in a combustion chamber of a target device, wherein the physical field data includes temperature field data, electric field data, and airflow field data in the combustion chamber of the target device; the temperature field data, the electric field data, and the airflow field data are respectively the temperature value of each sampling point in the combustion chamber at each sampling moment, the electric field intensity values in the x, y, and z directions, and the airflow velocity values in the x, y, and z directions; the target device is a device that uses plasma heating;

[0179] A feature extraction module 202 is configured to extract feature vectors of the temperature field data, the electric field data, and the airflow field data respectively through a feature extraction layer of the first model to obtain a target temperature field feature vector, a target electric field feature vector, and a target airflow field feature vector;

[0180] An intra-group matrix determination module 203 is configured to calculate the mean of each of the target temperature field eigenvectors, the target electric field eigenvectors, and the target airflow field eigenvectors through the matrix calculation layer of the first model, and to determine the temperature field intra-group scatter matrix, the electric field intra-group scatter matrix, and the airflow field intra-group scatter matrix using the mean values, respectively;

[0181] An inter-group matrix determination module 204 is configured to determine, through the matrix calculation layer, a joint eigenvector of each sampling point based on the target temperature field eigenvector, the target electric field eigenvector, and the target airflow field characteristic corresponding to each sampling point, and determine an inter-group scatter matrix of the sampling points based on the mean of the joint eigenvector;

[0182] A compression module 205 is configured to combine the intra-group scatter matrix and the inter-group scatter matrix into a three-dimensional tensor including two channels through a dimensionality conversion layer of the first model, and compress high-dimensional feature data corresponding to the three-dimensional tensor into low-dimensional feature data;

[0183] The flame stability determination module 206 is configured to determine the flame stability value of the target device according to the low-dimensional feature data through the multi-layer perception layer of the first model.

[0184] Regarding the apparatus in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0185] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, which can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0186] Other embodiments of the present disclosure will readily occur to those skilled in the art after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only, with the true scope and spirit of the present disclosure being indicated by the following claims.

[0187] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes can be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A method for determining flame stability, characterized in that: The method comprises: Collecting physical field data related to the combustion process in the combustion chamber of the target device, the physical field data including temperature field data, electric field data, and airflow field data in the combustion chamber of the target device; the temperature field data, the electric field data, and the airflow field data are respectively the temperature value of each sampling point in the combustion chamber at each sampling time, the electric field intensity values in the x, y, and z directions, and the airflow velocity values in the x, y, and z directions; the target device is a device using plasma heating; Extracting the characteristic vectors of the temperature field data, the electric field data, and the airflow field data respectively through the feature extraction layer of the first model to obtain a target temperature field characteristic vector, a target electric field characteristic vector, and a target airflow field characteristic vector; Calculating the mean of each of the target temperature field eigenvectors, the target electric field eigenvectors, and the target airflow field eigenvectors through the matrix calculation layer of the first model, and determining the temperature field group scatter matrix, the electric field group scatter matrix, and the airflow field group scatter matrix using the mean values, respectively; Determining, through the matrix calculation layer, a joint eigenvector of each sampling point according to the target temperature field eigenvector, the target electric field eigenvector, and the target airflow field characteristic corresponding to each sampling point, and determining an inter-group scatter matrix of the sampling points according to the mean of the joint eigenvector; Combining the intra-group scatter matrix and the inter-group scatter matrix into a three-dimensional tensor including two channels through a dimensionality conversion layer of the first model, and compressing high-dimensional feature data corresponding to the three-dimensional tensor into low-dimensional feature data; The flame stability value of the target device is determined according to the low-dimensional feature data through the multi-layer perception layer of the first model.

2. The method according to claim 1, characterized in that The extracting the characteristic vectors of the temperature field data, the electric field data, and the airflow field data respectively includes: Applying three-dimensional convolution to the temperature values to extract spatial correlation features of the temperature values in a local area, and using a time series model to extract hidden features of the spatial correlation features at the current sampling moment, and concatenating the hidden features at each sampling moment to obtain a target temperature field feature vector; Performing a weighted summation on the electric field intensity values in the x, y, and z directions to obtain a fused electric field eigenvalue; inputting the electric field eigenvalue into a multi-channel convolutional network to obtain a convolved feature tensor; flattening the feature tensor into a global vector along the spatial dimension to obtain a target electric field eigenvector; The airflow velocity values in the x, y, and z directions are weighted and summed to obtain a fused airflow field eigenvalue; the airflow field eigenvalue is fused with the target electric field eigenvector to obtain a joint eigenvalue; the joint eigenvalue is input into the multi-channel convolutional network, and the high-order features of the joint features are extracted to obtain a joint high-order eigenvalue; the joint high-order eigenvalue is flattened into a global vector to obtain a target airflow field eigenvector.

3. The method according to claim 1, characterized in that The respectively calculating the mean of each of the target temperature field characteristic vectors, the target electric field characteristic vectors, and the target airflow field characteristic vectors, and respectively using the mean to determine the temperature field group dispersion matrix, the electric field group dispersion matrix, and the airflow field group dispersion matrix, comprises: Calculating the mean of each of the target temperature field characteristic vectors, the target electric field characteristic vectors, and the target airflow field characteristic vectors respectively to obtain a temperature field mean vector, an electric field mean vector, and an airflow field mean vector; respectively calculating the difference vectors between each target temperature field characteristic vector and the temperature field mean vector, and forming a temperature field group scatter matrix from each difference vector; respectively calculating the difference vectors between each target electric field characteristic vector and the electric field mean vector, and forming an electric field group scatter matrix from each difference vector; The difference vectors between each target airflow field characteristic vector and the airflow field mean vector are calculated respectively, and the airflow field group scatter matrix is composed of each difference vector.

4. The method according to claim 1, wherein Determining a joint eigenvector of each sampling point according to the target temperature field eigenvector, the target electric field eigenvector, and the target airflow field characteristic corresponding to each sampling point, and determining an inter-group scatter matrix of the sampling points according to a mean of the joint eigenvector, includes: splicing the target temperature field feature vector, the target electric field feature vector, and the target airflow field feature vector corresponding to each sampling point to obtain a joint feature vector for each sampling point; Calculating the mean of the joint eigenvectors of each of the sampling points to obtain a total mean vector; The difference between the temperature field mean vector and the total mean vector, the difference between the electric field mean vector and the total mean vector, and the difference between the airflow field mean vector and the total mean vector are calculated respectively to obtain the inter-group scatter matrix of the sampling points.

5. The method according to claim 1, wherein The step of merging the intra-group scatter matrix and the inter-group scatter matrix into a three-dimensional tensor including two channels, and compressing high-dimensional feature data corresponding to the three-dimensional tensor into low-dimensional feature data, comprises: Merging the intra-group scatter matrix and the inter-group scatter matrix into a three-dimensional tensor with two channels through a channel expansion operation; Performing a first convolution operation on the three-dimensional tensor to learn a local feature relationship between the inter-group scatter matrix and the intra-group scatter matrix to obtain a first feature map; Performing a second-layer convolution operation on the first feature map to learn high-order features of the first feature map to obtain a second-layer feature map; A global average pooling operation and a full connection operation are performed on the second-layer feature map to compress the high-dimensional features of the second-layer feature map into low-dimensional features.

6. The method according to claim 1, characterized in that The loss function of the first model during training is: Among them, L contrastive Represents the loss value, Y pred (i) and Y pred (j) represents the prediction results of the first model at time i and time j, respectively, ∥Y pred (i)-Y pred (j)∥ represents Y pred (i) and Y pred (j) The Euclidean distance, m represents the preset boundary value.

7. The method according to claim 1, characterized in that After determining the flame stability of the device using plasma heating according to the low-dimensional feature data, the method further includes: If the flame stability value is less than a preset safety threshold, the voltage value and the current value are adjusted.

8. A flame stability determination device, characterized in that: The device comprises: a data acquisition module for acquiring physical field data related to the combustion process in the combustion chamber of the target device, the physical field data including temperature field data, electric field data, and airflow field data in the combustion chamber of the target device; the temperature field data, the electric field data, and the airflow field data are respectively the temperature value of each sampling point in the combustion chamber at each sampling moment, the electric field intensity values in the x, y, and z directions, and the airflow velocity values in the x, y, and z directions; the target device is a device using plasma heating; a feature extraction module, configured to extract the feature vectors of the temperature field data, the electric field data, and the airflow field data respectively through the feature extraction layer of the first model, and obtain a target temperature field feature vector, a target electric field feature vector, and a target airflow field feature vector; an intra-group matrix determination module, configured to calculate the mean of each of the target temperature field eigenvectors, the target electric field eigenvectors, and the target airflow field eigenvectors through the matrix calculation layer of the first model, and to determine the temperature field intra-group scatter matrix, the electric field intra-group scatter matrix, and the airflow field intra-group scatter matrix using the mean values; an inter-group matrix determination module, configured to determine, through the matrix calculation layer, a joint eigenvector of each sampling point according to the target temperature field eigenvector, the target electric field eigenvector, and the target airflow field characteristic corresponding to each sampling point, and determine an inter-group scatter matrix of the sampling points according to the mean of the joint eigenvector; a compression module, configured to combine the intra-group scatter matrix and the inter-group scatter matrix into a three-dimensional tensor including two channels through a dimensionality conversion layer of the first model, and compress high-dimensional feature data corresponding to the three-dimensional tensor into low-dimensional feature data; A flame stability determination module is used to determine the flame stability value of the target device according to the low-dimensional feature data through the multi-layer perception layer of the first model.

9. A device using plasma heating, characterized in that include: Controller; a memory for storing instructions executable by the controller; The controller is configured to execute the instructions to implement the flame stability determination method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that When the instructions in the computer-readable storage medium are executed by a controller of a device using plasma heating, the device using plasma heating is enabled to perform the flame stability determination method according to any one of claims 1 to 7.

Images