Power supply type identification method and device

Abstract

The invention discloses a power supply type identification method and device. The method comprises the following steps: acquiring current data, voltage data and power supply frequency data of a power supply circuit of target equipment; determining transient data and steady state data according to the current data and the voltage data; inputting the transient data and the steady-state data into a target model, extracting a transient feature vector by using an instantaneous branch of a double-branch attention fusion network of the target model, and extracting a steady-state feature vector by using a steady-state branch; performing feature alignment on the transient feature vector and the steady-state feature vector to obtain a query vector, a key vector and a value vector; and in the target model, performing power supply type identification according to the fusion feature vector to obtain a power supply type corresponding to the to-be-identified power supply. According to the scheme, multiple power supply types such as commercial power, mobile power supplies and new energy automobile batteries can be intelligently identified, the use requirements of different scenes are met, and the fault risk caused by power supply mismatching is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field, and in particular to a method and device for identifying a power source category. Background Art

[0002] Traditional plasma stoves have a single application scenario and a relatively simple power supply identification method, which cannot identify multiple power supply types, resulting in a lack of adaptability in startup control. This leads to poor user experience and insufficient system reliability in application scenarios such as outdoor barbecues. Summary of the Invention

[0003] The present invention aims to at least solve the technical problems existing in the prior art. To this end, the present invention provides a power source category identification method in a first aspect, the method comprising:

[0004] Acquiring current data, voltage data, and power frequency data of a power circuit of a target device from the time the target device is connected to a power source to be identified until stable combustion; the target device is a device that uses plasma ignition;

[0005] Determine transient data and steady-state data based on the current data and the voltage data; the transient data at least includes voltage time series data, current time series data, voltage change rate time series data, current change rate time series data, and harmonic amplitude time series data at each sampling moment; the steady-state data at least includes an initial voltage value, an initial current value, an AC power frequency, a steady-state voltage fluctuation rate, a steady-state power factor, and an ambient temperature compensation value;

[0006] Inputting the transient data and the steady-state data into a target model, extracting a transient feature vector of the transient data using a transient branch of a dual-branch attention fusion network of the target model, and extracting a steady-state feature vector of the steady-state data using a steady-state branch of the network;

[0007] Performing feature alignment on the transient feature vector and the steady-state feature vector to obtain a query vector, a key vector, and a value vector;

[0008] Performing feature fusion on the transient feature vector and the steady-state feature vector using the query vector, the key vector, and the value vector to obtain a fused feature vector;

[0009] In the target model, power source category identification is performed based on the fused feature vector to obtain the power source category corresponding to the power source to be identified.

[0010] Optionally, the instantaneous branch of the dual-branch attention fusion network is composed of a causal hole convolution layer, a gating mechanism layer, and a residual structure layer, and the instantaneous branch of the dual-branch attention fusion network of the target model is used to extract the transient feature vector of the transient data, including:

[0011] Inputting the transient data into the causal dilated convolutional layer, and performing a convolution operation on the transient data using a convolution kernel to obtain a first tensor;

[0012] Inputting the first tensor into the gating mechanism layer to determine the information flow passed to the next time step to obtain a second tensor;

[0013] Adding the transient feature data and the second tensor element by element at the residual structure layer to obtain a fused tensor;

[0014] A pooling operation is performed on the fused tensor to generate a compressed transient feature vector.

[0015] Optionally, the algorithm of the causal dilated convolutional layer is:

[0016]

[0017] Among them, y t represents the first tensor of the output, d represents the void rate, t represents the time position of the current output, k represents the convolution kernel weight index, w k represents the convolution kernel weight, x t-d·k Represents the input signal X t The element of row td·k of ;

[0018] The algorithm of the gating mechanism layer is:

[0019] Output=GatedConv(X)=tanh(W f *X)⊙σ(W g *X)

[0020] Among them, GatedConv(X) represents the second tensor of the output, is the input tensor, T represents the time step, F in represents the feature dimension, ⊙ is element-wise multiplication, W f , W g is the convolution weight, W f represents the convolution kernel that generates the activation signal A, W g Represents the convolution kernel that generates the gating signal G.

[0021] Optionally, the dimension of the transient feature vector is d*t, the dimension of the steady-state feature vector is d*s, and the feature alignment of the transient feature vector and the steady-state feature vector to obtain a query vector, a key vector, and a value vector includes:

[0022] Use dimension d k *d first projection matrix transforms the transient eigenvector into a matrix of dimension d k *t's query matrix;

[0023] Use dimension d k *d second projection matrix transforms the steady-state eigenvector into a matrix of dimension d k *s key matrix, and uses a dimension of d k *d's third projection matrix transforms the steady-state eigenvector into a matrix of dimension d k *The value matrix of s.

[0024] Optionally, the performing feature fusion on the transient feature vector and the steady-state feature vector using the query vector, the key vector, and the value vector to obtain a fused feature vector includes:

[0025] Determine a bilinear score according to the query vector and the key vector, and normalize the bilinear score into a probability distribution using a normalization function to obtain an attention weight matrix;

[0026] The product of the attention weight matrix and the value vector is determined to obtain a fused feature vector of the transient feature vector and the steady-state feature vector.

[0027] Optionally, the loss function used by the target model during training is a target loss function, which is a weighted sum of a center loss function and a cross entropy loss function.

[0028] The center loss function is:

[0029]

[0030] Among them, f i ,f j Respectively represent the i-th sample and the j-th sample in the fused feature vector, Indicates that it belongs to category y i The total number of samples, Power supply category y i The center vector of , m represents the batch size;

[0031] The cross entropy loss function is:

[0032]

[0033] Among them, N represents the total number of samples, w i represents the weight of the i-th sample, y i represents the true label of the i-th sample, represents the predicted probability of the i-th sample.

[0034] Optionally, the calculation method of the harmonic amplitude time series data is:

[0035] Performing a fast Fourier transform on the current time series data to obtain current spectrum data;

[0036] The harmonic amplitude of each period in the current spectrum data is extracted according to a sliding window to obtain harmonic amplitude time series data.

[0037] A second aspect of the present invention provides a power source category identification device, the device comprising:

[0038] a data acquisition module for acquiring current data, voltage data, and power frequency data of a power circuit of a target device from the time the target device is connected to a power source to be identified until stable combustion; the target device is a device that uses plasma ignition;

[0039] a data determination module, configured to determine transient data and steady-state data based on the current data and the voltage data; the transient data including at least voltage time series data, current time series data, voltage change rate time series data, current change rate time series data, and harmonic amplitude time series data at each sampling moment; and the steady-state data including at least initial voltage value, initial current value, AC power supply frequency, steady-state voltage fluctuation rate, steady-state power factor, and ambient temperature compensation value;

[0040] a feature vector extraction module, configured to input the transient data and the steady-state data into a target model, extract the transient feature vector of the transient data using the transient branch of the dual-branch attention fusion network of the target model, and extract the steady-state feature vector of the steady-state data using the steady-state branch of the network;

[0041] a feature alignment module, configured to perform feature alignment on the transient feature vector and the steady-state feature vector to obtain a query vector, a key vector, and a value vector;

[0042] a feature fusion module, configured to perform feature fusion on the transient feature vector and the steady-state feature vector using the query vector, the key vector, and the value vector to obtain a fused feature vector;

[0043] An identification module is used to identify the power category in the target model according to the fused feature vector to obtain the power category corresponding to the power supply to be identified.

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

[0045] processor;

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

[0047] The processor is configured to execute the instructions to implement the power category identification 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 processor of a device using plasma heating, the device using plasma heating can perform the power category identification 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, current data, voltage data, and power frequency data of the power circuit of the target device are obtained from the time the target device is connected to the power supply to be identified to the time the target device is in stable combustion; the target device is a device that uses plasma ignition; transient data and steady-state data are determined based on the current data and the voltage data; the transient data at least includes voltage time series data, current time series data, voltage change rate time series data, current change rate time series data, and harmonic amplitude time series data at each sampling moment; the steady-state data at least includes initial voltage value, initial current value, AC power frequency, steady-state voltage fluctuation rate, steady-state power factor, and ambient temperature compensation value; the transient data The steady-state data is input into the target model, and the transient feature vector of the transient data is extracted using the transient branch of the dual-branch attention fusion network of the target model, and the steady-state feature vector of the steady-state data is extracted using the steady-state branch of the network; the transient feature vector and the steady-state feature vector are feature aligned to obtain a query vector, a key vector, and a value vector; the transient feature vector and the steady-state feature vector are feature fused using the query vector, the key vector, and the value vector to obtain a fused feature vector; in the target model, the power source category is identified based on the fused feature vector to obtain the power source category corresponding to the power source to be identified. This solution can intelligently identify various power source types such as mains electricity, mobile power supplies, and new energy vehicle batteries to meet the needs of different scenarios; and the solution can operate efficiently and stably, significantly improving the user experience, while reducing the risk of failures caused by power mismatch and improving the reliability of the overall system performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a flowchart of a method for identifying a power source type according to an embodiment of the present invention;

[0052] Figure 2 This is a structural block diagram of a power category identification 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 a method for identifying a power source type provided by an embodiment of the present invention.

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

[0057] Step 101: Acquire current data, voltage data, and power frequency data of a power circuit of a target device from the time the target device is connected to a power source to be identified to the time the target device stabilizes combustion; the target device is a device that uses plasma ignition.

[0058] In the embodiments of the present invention, devices using plasma heating 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. The plasma heating principle involves high voltage at the end of the plasma breaking through air to generate an open flame, which then heats the device.

[0059] Equipment using plasma heating can be connected to various types of power sources such as mains electricity, mobile power supplies, and new energy vehicle batteries.

[0060] After the target device is connected to the power supply and ignited, it gradually enters a stable combustion state. A high-precision analog-to-digital converter (e.g., 16-bit or higher precision) is used to sample the voltage and current signals of the power circuit from the time the power supply is connected to the stable combustion process in real time, and to capture the dynamic changes of the power supply. The sampling frequency of the high-precision analog-to-digital converter is not less than 10kHz.

[0061] The current data, voltage data and power frequency data are obtained by sampling.

[0062] Step 102: Determine transient data and steady-state data based on the current data and the voltage data; the transient data at least includes voltage time series data, current time series data, voltage change rate time series data, current change rate time series data, and harmonic amplitude time series data at each sampling moment; the steady-state data at least includes initial voltage value, initial current value, AC power frequency, steady-state voltage fluctuation rate, steady-state power factor, and ambient temperature compensation value.

[0063] Transient data refers to data that changes over time, while steady-state data refers to data that remains constant.

[0064] The transient data at least includes voltage time series data, current time series data, voltage change rate time series data, current change rate time series data, and harmonic amplitude time series data at each sampling moment.

[0065] Among them, voltage time series data refers to the original voltage signal collected in real time by a high-precision analog-to-digital converter, which retains complete time dimension information. The method for obtaining voltage time series data V(t) is:

[0066]

[0067] Among them, V(t) represents the original voltage waveform, T represents the maximum sampling time, r s Indicates the sampling rate.

[0068] Current time series data refers to the original current signal collected in real time by a high-precision analog-to-digital converter, which retains complete time dimension information. The method for obtaining current time series data I(t) is:

[0069]

[0070] Where I(t) represents the original current waveform, T represents the maximum sampling time, and r s Indicates the sampling rate.

[0071] Voltage change rate time series data refers to the instantaneous change rate of the voltage waveform in the time dimension, which is obtained through differential calculation and is used to identify the dynamic response speed of voltage changes and the regulation ability of the power supply when the load changes suddenly. The method to obtain is:

[0072]

[0073] Where V(t) represents the original voltage waveform, T represents the maximum sampling time, and Δt represents the sampling interval.

[0074] Current change rate time series data refers to the instantaneous change rate of the current waveform in the time dimension, which is obtained through differential calculation and is used to quantify the dynamic response speed of current changes and reflect the transient output capability of the power supply. The method to obtain is:

[0075]

[0076] Where I(t) represents the original voltage waveform, T represents the maximum sampling time, and Δt represents the sampling interval.

[0077] As an optional embodiment, the calculation method of the harmonic amplitude time series data is:

[0078] Step 1011: Perform fast Fourier transform on the current time series data to obtain current spectrum data;

[0079] Step 1012: extract the harmonic amplitude of each period in the current spectrum data according to the sliding window to obtain harmonic amplitude time series data.

[0080] In step 1011-step 1012, the calculation method of the harmonic amplitude time series data Hk(t) is:

[0081] Hk(t)=FFT(I(tw:t),f(w))[k](k=1,3,5,7)

[0082] Where I(tw:t) represents the current signal segment from time tw to t, with length w, and f(w) represents the window function, such as the Hanning window. k represents the harmonic order, which is an integer multiple of the fundamental frequency.

[0083] The steady-state data at least includes an initial voltage value, an initial current value, an AC power frequency, a steady-state voltage fluctuation rate, a steady-state power factor, and an ambient temperature compensation value.

[0084] The initial voltage value refers to the stable voltage value after power is applied and the arc is not activated, reflecting the rated output voltage of the power supply. The initial current value refers to the stable current value after power is applied and the arc is not activated, reflecting the rated output current of the power supply.

[0085] AC power frequency refers to the frequency of the AC power supply. Mains electricity is AC, and mobile power banks and new energy vehicle batteries can also output AC.

[0086] Steady-state voltage fluctuation rate refers to the voltage fluctuation rate when reaching steady-state voltage.

[0087] The steady-state power factor reflects the power utilization efficiency of the power supply. The power factors of different power supplies may vary. The method to obtain the steady-state power factor PF is:

[0088]

[0089] Where P represents effective power; U rms , I rms Indicates the effective value of current and voltage.

[0090] Steady-state characteristics may be affected by ambient temperature. The ambient temperature compensation value is used to perform temperature compensation on the steady-state data to improve the robustness of the data. The ambient temperature compensation value includes the temperature-compensated voltage value Ucomp and the temperature-compensated current value Icomp, and the method for obtaining it is:

[0091] Ucomp=Umeas·[1+α U (Tref-Tenv)]

[0092] Icomp=Imeas·[1+α I (Tref-Tenv)]

[0093] Among them, U comp , I comp Respectively represent the voltage and current values after temperature compensation; U meas , I meas Respectively represent the actual measured original voltage value and original current value; α U ,α I Indicates voltage temperature coefficient and current temperature coefficient; T ref Indicates the reference temperature, T env Indicates the ambient temperature.

[0094] After obtaining the transient data and steady-state data, normalize them and scale the data values to [0,1]. The normalization formula is:

[0095]

[0096] Among them, x′ is the original eigenvalue, x min is the minimum characteristic value, x max is the maximum value of the feature.

[0097] Step 103: Input the transient data and the steady-state data into the target model, use the transient branch of the dual-branch attention fusion network of the target model to extract the transient feature vector of the transient data, and use the steady-state branch of the network to extract the steady-state feature vector of the steady-state data.

[0098] Traditional methods, when fusing transient and steady-state data, ignore the differences between their dynamic and static characteristics, leading to inadequate feature extraction. For example, they fail to distinguish between the temporal dynamics of transient data and the static nonlinear relationships of steady-state data. This invention designs two independent branches to process transient and steady-state data respectively, and then adaptively fuses them through an attention mechanism, optimizing both the dynamic temporal relationship and the static nonlinear relationship, improving the fusion effect.

[0099] Taking into account the essential differences between transient features and steady-state features, as well as the importance of fusing the information contained in these features, the present invention proposes a dual-branch attention fusion network to more effectively extract the respective feature information. The transient branch focuses on capturing the temporal dynamic relationship of transient data, and the steady-state branch focuses on extracting the nonlinear relationship of steady-state data. Then, through bilinear interaction, the importance of the two types of features is automatically identified and weightedly merged.

[0100] As an optional embodiment, the instantaneous branch of the dual-branch attention fusion network is composed of a causal hole convolution layer, a gating mechanism layer, and a residual structure layer. Step 103 includes:

[0101] Step 1031: Input the transient data into the causal dilated convolution layer, and perform a convolution operation on the transient data using a convolution kernel to obtain a first tensor;

[0102] Step 1032: Input the first tensor into the gating mechanism layer to determine the information flow to the next time step, and obtain a second tensor;

[0103] Step 1033: Add the transient feature data and the second tensor element by element at the residual structure layer to obtain a fused tensor;

[0104] Step 1034: Perform a pooling operation on the fused tensor to generate a compressed transient feature vector.

[0105] In steps 1031 to 1034, the instantaneous branch of the dual-branch attention fusion network consists of a causal hole convolution layer, a gate mechanism layer, and a residual structure layer. The instantaneous branch inputs transient data related to voltage and current, and the shape is Where T is the time step, F in The instantaneous branch is constructed using an improved temporal convolution module to capture the temporal dependencies of transient signals such as current and voltage.

[0106] Among them, the causal hole convolution layer ensures that at any time point t, the output depends only on the input at time point t and before, and does not depend on the input after t. Causal convolution can be achieved by appropriately "padding" the input data. Specifically, for a one-dimensional input sequence and a convolution kernel of size k, in order to achieve causal convolution, k-1 zeros can be padded at the beginning of the sequence, and then a standard convolution operation can be performed. In this way, the output of the convolution at any time point t will depend on the input at time point t and before.

[0107] Gating mechanism is a key technology used to control the flow of information in neural networks. Its main function is to decide which information should be retained, discarded or passed to the next time step by introducing different gating units.

[0108] In order to alleviate the gradient disappearance problem, residual connections are used to retain input information. The residual connection can be expressed as:

[0109] Foutput=X+GatedConv(X)

[0110] Perform a pooling operation on the output Foutput of the residual connection to generate a compressed transient feature vector F t ∈R dt .

[0111] Where T is the time step, F t ∈R at Represents compressed transient data.

[0112] The steady-state feature stream is processed using a multi-layer perceptron to learn the nonlinear relationship between steady-state features. The steady-state branch inputs static feature data, which has the shape of Among them, F s is the number of features. After being extracted by the multi-layer perceptron, the nonlinear features are output in the shape of F s ∈R ds .

[0113] As an optional embodiment, the algorithm of the causal dilated convolutional layer is:

[0114]

[0115] Among them, y t represents the first tensor of the output, d represents the void rate, t represents the time position of the current output, k represents the convolution kernel weight index, w k represents the convolution kernel weight, x t-d·k Represents the input signal X t The element of row td·k of ;

[0116] The algorithm of the gating mechanism layer is:

[0117] Output=GatedConv(X)=tanh(W f *X)⊙σ(W g *X)

[0118] Among them, GatedConv(X) represents the second tensor of the output, is the input tensor, T represents the time step, F in represents the feature dimension, ⊙ is element-wise multiplication, W f , W g is the convolution weight, W f represents the convolution kernel that generates the activation signal A, W g Represents the convolution kernel that generates the gating signal G. * denotes the causal convolution operation, tanh is the hyperbolic tangent function, and σ is the simmoid function.

[0119] Specifically, for the algorithm of the causal hole convolution layer, d represents the hole rate, which is used to expand the receptive field, t represents the time position of the current output, and all inputs must come from time steps t and earlier.

[0120] Step 104: perform feature alignment on the transient feature vector and the steady-state feature vector to obtain a query vector, a key vector, and a value vector.

[0121] In order to better learn the interactive relationship and complementary information between transient feature vectors and steady-state feature vectors, the present invention uses bilinear attention to capture the complex relationship between transient features and steady-state features, so as to more fully integrate the two different types of information, transient features and steady-state features, so that the network can better understand the overall characteristics of the power supply and make accurate judgments.

[0122] Since the dimensions of the transient eigenvector and the steady-state eigenvector are different and cannot interact directly, feature alignment is used to map both to a unified dimension d so that they are comparable in the same space.

[0123] As an optional embodiment, the dimension of the transient feature vector is d*t, and the dimension of the steady-state feature vector is d*s. Step 104 includes:

[0124] Step 1041: Use the first projection matrix with a dimension of d*d to transform the transient feature vector into a matrix with a dimension of d k *t's query matrix;

[0125] Step 1042: Use dimension d k *t's second projection matrix transforms the steady-state eigenvector into a matrix of dimension d k *s key vector, and use a dimension of dk *d's third projection matrix transforms the steady-state eigenvector into a matrix of dimension d k *The value vector of s.

[0126] In steps 1041-1042, the projection matrix W is used q The transient feature F t Convert to query vector Q, using projection matrix W k ,W v The steady-state characteristic F s Converted into key vector K and value vector V, the specific formula is as follows:

[0127] Q=W q F t , K=W k F s , V=W v F s (W q , W k , W v ∈R d*d )

[0128] Among them, F t ∈R dt represents the transient eigenvector, F s ∈R ds represents the steady-state eigenvector, W q ∈R d*d , W k ∈R d*d , W v ∈R d*d Represents a learnable projection matrix. Q∈R d is a query vector, which represents the “demand” of transient features and is used to match related steady-state features; K∈R d is a key vector, representing the “identity” of the steady-state feature, used to match related steady-state features; V∈R d It is a value vector, which represents the steady-state feature information actually involved in the fusion.

[0129] Step 105: Perform feature fusion on the transient feature vector and the steady-state feature vector using the query vector, the key vector, and the value vector to obtain a fused feature vector.

[0130] Specifically, bilinear attention is used to capture the complex relationship between transient feature vectors and steady-state feature vectors, fully integrating information of two different properties: transient feature vectors and steady-state feature vectors.

[0131] As an optional embodiment, step 105 includes:

[0132] Step 1051: Determine a bilinear score based on the query vector and the key vector, and use a normalization function to normalize the bilinear score into a probability distribution to obtain an attention weight matrix;

[0133] Step 1052: Determine the product of the attention weight matrix and the value vector to obtain a fused feature vector of the transient feature vector and the steady-state feature vector.

[0134] In step 1051-step 1052, the attention weight matrix is calculated as follows:

[0135]

[0136] Where U∈R d*d is a learnable parameter, It is a bilinear score obtained by Q and K calculated by feature alignment, which is used to measure the correlation strength between the transient feature Q and the steady-state feature K. represents the scaling factor, Softmax is the normalization function, which normalizes the bilinear score to a probability distribution to highlight important features, and A is the attention weight matrix, which represents the correlation weight between transient and steady-state features and determines the contribution of each feature during fusion.

[0137] The calculation method of feature fusion is:

[0138] F fusion =A·V

[0139] Among them, F fusion represents the fused feature vector, A is the attention weight matrix, which is calculated by bilinear interaction, and V represents the value vector, which is calculated by the feature alignment step.

[0140] Step 106: In the target model, power source category identification is performed based on the fused feature vector to obtain the power source category corresponding to the power source to be identified.

[0141] The target model identifies the power source category based on the fused feature vector and obtains the power source category corresponding to the power source to be identified. The power source category can be AC power, mobile power supply, new energy vehicle battery, etc.

[0142] As an optional embodiment, the loss function used by the target model during training is a target loss function, which is a weighted sum of a center loss function and a cross entropy loss function.

[0143] The center loss function is:

[0144]

[0145] Among them, f i ,fj Respectively represent the i-th sample and the j-th sample in the fused feature vector, Indicates that it belongs to category y i The total number of samples, c yi Power supply category y i The center vector of , m represents the batch size;

[0146] The cross entropy loss function is:

[0147]

[0148] Among them, N represents the total number of samples, w i represents the weight of the i-th sample, y i represents the true label of the i-th sample, represents the predicted probability of the i-th sample.

[0149] In the embodiment of the present invention, considering that there is a large overlap in the feature distributions of different power types, a center loss function is used to reduce the feature distance of similar samples to improve recognition accuracy.

[0150] However, some power categories may appear less frequently, resulting in an imbalanced power category. In this case, weighted cross entropy loss is used to give greater weight to rare categories.

[0151] The objective loss function is:

[0152] L=L WCE +λL center

[0153] Among them, λ represents the balance factor, which controls the contribution of the central loss to the total loss and prevents excessive feature aggregation from blurring the classification boundaries.

[0154] In summary, the power category identification method provided by the embodiment of the present invention obtains the current data, voltage data and power frequency data of the power circuit of the target device from the time when the target device is connected to the power supply to be identified to the time when the target device is stably burned; the target device is a device that uses plasma ignition; transient data and steady-state data are determined based on the current data and the voltage data; the transient data at least includes voltage time series data, current time series data, voltage change rate time series data, current change rate time series data, and harmonic amplitude time series data at each sampling moment; the steady-state data at least includes initial voltage value, initial current value, AC power frequency, steady-state voltage fluctuation rate, steady-state power factor and ambient temperature compensation value; The transient data and the steady-state data are input into the target model, and the transient feature vector of the transient data is extracted using the transient branch of the dual-branch attention fusion network of the target model, and the steady-state feature vector of the steady-state data is extracted using the steady-state branch of the network; the transient feature vector and the steady-state feature vector are feature aligned to obtain a query vector, a key vector, and a value vector; the transient feature vector and the steady-state feature vector are feature fused using the query vector, the key vector, and the value vector to obtain a fused feature vector; in the target model, the power source category is identified based on the fused feature vector to obtain the power source category corresponding to the power source to be identified. This solution can intelligently identify various power source types such as mains electricity, mobile power supplies, and new energy vehicle batteries to meet the needs of different scenarios; and the solution can operate efficiently and stably, significantly improving the user experience, while reducing the risk of failures caused by power mismatch and improving the reliability of the overall system performance.

[0155] Figure 2 This is a structural block diagram of a power category identification device provided by an embodiment of the present invention. Figure 2 As shown, the device 200 includes:

[0156] A data acquisition module 201 is configured to acquire current data, voltage data, and power frequency data of a power circuit of a target device from the time the target device is connected to a power source to be identified until stable combustion; the target device is a device that uses plasma ignition;

[0157] The data determination module 202 is configured to determine transient data and steady-state data based on the current data and the voltage data; the transient data includes at least voltage time series data, current time series data, voltage change rate time series data, current change rate time series data, and harmonic amplitude time series data at each sampling moment; and the steady-state data includes at least initial voltage value, initial current value, AC power frequency, steady-state voltage fluctuation rate, steady-state power factor, and ambient temperature compensation value.

[0158] A feature vector extraction module 203 is configured to input the transient data and the steady-state data into a target model, extract the transient feature vector of the transient data using the transient branch of the dual-branch attention fusion network of the target model, and extract the steady-state feature vector of the steady-state data using the steady-state branch of the network;

[0159] A feature alignment module 204 is configured to perform feature alignment on the transient feature vector and the steady-state feature vector to obtain a query vector, a key vector, and a value vector;

[0160] A feature fusion module 205 is configured to perform feature fusion on the transient feature vector and the steady-state feature vector using the query vector, the key vector, and the value vector to obtain a fused feature vector;

[0161] The identification module 206 is configured to identify the power category in the target model according to the fused feature vector to obtain the power category corresponding to the power to be identified.

[0162] 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.

[0163] 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).

[0164] 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.

[0165] 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 identifying a power source category, characterized in that: The method comprises: Acquiring current data, voltage data, and power frequency data of a power circuit of a target device from the time the target device is connected to a power source to be identified until stable combustion; the target device is a device that uses plasma ignition; Determine transient data and steady-state data based on the current data and the voltage data; the transient data at least includes voltage time series data, current time series data, voltage change rate time series data, current change rate time series data, and harmonic amplitude time series data at each sampling moment; the steady-state data at least includes an initial voltage value, an initial current value, an AC power frequency, a steady-state voltage fluctuation rate, a steady-state power factor, and an ambient temperature compensation value; Inputting the transient data and the steady-state data into a target model, extracting a transient feature vector of the transient data using a transient branch of a dual-branch attention fusion network of the target model, and extracting a steady-state feature vector of the steady-state data using a steady-state branch of the network; Performing feature alignment on the transient feature vector and the steady-state feature vector to obtain a query vector, a key vector, and a value vector; Performing feature fusion on the transient feature vector and the steady-state feature vector using the query vector, the key vector, and the value vector to obtain a fused feature vector; In the target model, power source category identification is performed based on the fused feature vector to obtain the power source category corresponding to the power source to be identified.

2. The method according to claim 1, characterized in that The instantaneous branch of the dual-branch attention fusion network is composed of a causal hole convolution layer, a gating mechanism layer, and a residual structure layer. The instantaneous branch of the dual-branch attention fusion network using the target model extracts the transient feature vector of the transient data, including: Inputting the transient data into the causal dilated convolutional layer, and performing a convolution operation on the transient data using a convolution kernel to obtain a first tensor; Inputting the first tensor into the gating mechanism layer to determine the information flow passed to the next time step to obtain a second tensor; Adding the transient feature data and the second tensor element by element at the residual structure layer to obtain a fused tensor; A pooling operation is performed on the fused tensor to generate a compressed transient feature vector.

3. The method according to claim 2, characterized in that The algorithm of the causal hole convolution layer is: Among them, y t represents the first tensor of the output, d represents the void rate, t represents the time position of the current output, k represents the convolution kernel weight index, w k represents the convolution kernel weight, x t-d·k Represents the input signal X t The element of row td·k of ; The algorithm of the gating mechanism layer is: Output=GatedConv(x)=tanh(W f *X)⊙σ(W g *X) Among them, GatedConv(X) represents the second tensor of the output, is the input tensor, T represents the time step, F in represents the feature dimension, ⊙ is element-wise multiplication, W f , W g is the convolution weight, W f represents the convolution kernel that generates the activation signal A, W g Represents the convolution kernel that generates the gating signal G.

4. The method according to claim 1, wherein The dimension of the transient feature vector is d*t, the dimension of the steady-state feature vector is d*s, and the feature alignment of the transient feature vector and the steady-state feature vector to obtain a query vector, a key vector, and a value vector includes: Use dimension d k *d first projection matrix transforms the transient eigenvector into a matrix of dimension d k *t's query matrix; Use dimension d k *d second projection matrix transforms the steady-state eigenvector into a matrix of dimension d k *s key matrix, and uses a dimension of d k *d's third projection matrix transforms the steady-state eigenvector into a matrix of dimension d k *The value matrix of s.

5. The method according to claim 4, characterized in that The step of performing feature fusion on the transient feature vector and the steady-state feature vector using the query vector, the key vector, and the value vector to obtain a fused feature vector includes: Determine a bilinear score according to the query vector and the key vector, and normalize the bilinear score into a probability distribution using a normalization function to obtain an attention weight matrix; The product of the attention weight matrix and the value vector is determined to obtain a fused feature vector of the transient feature vector and the steady-state feature vector.

6. The method according to claim 1, characterized in that The loss function used by the target model during training is the target loss function, which is the weighted sum of the center loss function and the cross entropy loss function. The center loss function is: Among them, f i ,f j Respectively represent the i-th sample and the j-th sample in the fused feature vector, Indicates that it belongs to category y i The total number of samples, Power supply category y i The center vector of , m represents the batch size; The cross entropy loss function is: Where N is the total number of samples, w i represents the weight of the i-th sample, y i represents the true label of the i-th sample, represents the predicted probability of the i-th sample.

7. The method according to claim 1, characterized in that The calculation method of the harmonic amplitude time series data is: Performing a fast Fourier transform on the current time series data to obtain current spectrum data; The harmonic amplitude of each period in the current spectrum data is extracted according to a sliding window to obtain harmonic amplitude time series data.

8. A power source category identification device, characterized in that: The device comprises: a data acquisition module for acquiring current data, voltage data, and power frequency data of a power circuit of a target device from the time the target device is connected to a power source to be identified until stable combustion; the target device is a device that uses plasma ignition; a data determination module, configured to determine transient data and steady-state data based on the current data and the voltage data; the transient data including at least voltage time series data, current time series data, voltage change rate time series data, current change rate time series data, and harmonic amplitude time series data at each sampling moment; and the steady-state data including at least initial voltage value, initial current value, AC power supply frequency, steady-state voltage fluctuation rate, steady-state power factor, and ambient temperature compensation value; a feature vector extraction module, configured to input the transient data and the steady-state data into a target model, extract the transient feature vector of the transient data using the transient branch of the dual-branch attention fusion network of the target model, and extract the steady-state feature vector of the steady-state data using the steady-state branch of the network; a feature alignment module, configured to perform feature alignment on the transient feature vector and the steady-state feature vector to obtain a query vector, a key vector, and a value vector; a feature fusion module, configured to perform feature fusion on the transient feature vector and the steady-state feature vector using the query vector, the key vector, and the value vector to obtain a fused feature vector; An identification module is used to identify the power category in the target model according to the fused feature vector to obtain the power category corresponding to the power supply to be identified.

9. A device using plasma heating, characterized in that include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the power category identification 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 processor of a device using plasma heating, the device using plasma heating is enabled to perform the power category identification method according to any one of claims 1 to 7.

Images