Electric vehicle load identification method and device based on double-layer classification

Abstract

The invention discloses an electric vehicle load identification method and device based on double-layer classification, and the method comprises the steps: carrying out the ADC sampling of a voltage and current signal of a main power supply, and obtaining a detected power signal: carrying out the feature extraction of the detected power signal obtained through sampling, and generating a feature vector; performing first-layer classification on the extracted feature vectors by using a pre-trained classifier, identifying a charging type corresponding to the charging process, and obtaining a first-layer classification result; performing time-frequency analysis on the measured power signal obtained by sampling, and extracting time-frequency domain features; and selecting a corresponding pre-trained deep network model according to the first-layer classification result, taking the extracted time-frequency domain features as the input of the selected deep network model to carry out second-layer classification, and identifying the state in the electric load process. The method has the advantages of being simple in implementation method, low in cost, high in recognition efficiency and precision, wide in application range and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of load identification, and in particular to a method and device for electric vehicle load identification based on double-layer classification. Background Art

[0002] Non-Intrusive Load Monitoring (NILM) can identify each electrical device and its working status by analyzing and processing the total load meter data. It can be applied to various fields such as building energy conservation, smart cities, and smart grids.

[0003] Electric vehicle charging identification is not only crucial for the balance of supply and demand in the power grid, but also helps to improve the stability, flexibility and economy of the power grid. The power grid needs to ensure the balance of supply and demand at all times to ensure that the power generation matches the power consumption, and the charging identification of electric vehicles can help better manage the load. For example, by mastering the user's charging time information, a more flexible load adjustment strategy can be implemented. For example, through price mechanisms, signal control and other means, users can be encouraged to choose to charge during the low power demand period (such as at night) to reduce the pressure on the power grid during peak hours. In some specific periods or specific areas, if a large number of electric vehicles are charged at high power at the same time, it may cause voltage fluctuations, line overloads or equipment damage in the local power grid. Therefore, through charging identification, the charging demand can be predicted in advance, which makes it easier to take appropriate measures to avoid overload.

[0004] For the load identification of electric vehicles, the prior art usually adopts data analysis methods or model identification methods, wherein the data analysis method is to analyze the electric vehicle power consumption data and judge the electric vehicle load identification based on certain parameter characteristics (such as power amplitude, etc.); the model identification method is to collect data during the charging and discharging process of the electric vehicle, extract the features, and then use a specific identification algorithm model to achieve identification. However, both the data analysis method and the model identification method are only applicable to identification in specific scenarios, and in practice there are many types of charging. For example, there are fast charging and slow charging according to the charging speed, and there are GB standards, IEC / ISO standards, etc. according to the charging standards. The power consumption data characteristics and applicable identification models under different charging types are different. The use of a unified parameter specific or identification model cannot ensure that the load status of electric vehicles can be accurately and stably identified in various environments. Summary of the invention

[0005] The technical problem to be solved by the present invention is: in view of the technical problems existing in the prior art, the present invention provides an electric vehicle load identification method and device based on double-layer classification, which has a simple implementation method, low cost, high identification efficiency and accuracy, and a wide range of applications.

[0006] In order to solve the above technical problems, the technical solution proposed by the present invention is:

[0007] A method for identifying electric vehicle load based on double-layer classification, comprising the following steps:

[0008] The voltage and current signals of the total power supply are sampled by ADC to obtain the measured power signal:

[0009] Extracting features of the sampled measured power signal to generate a feature vector;

[0010] Using a pre-trained classifier to perform a first-level classification on the extracted feature vector, identifying the charging type corresponding to the charging process, and obtaining a first-level classification result;

[0011] Perform time-frequency analysis on the sampled power signal to extract time-frequency domain features;

[0012] According to the first-layer classification result, a corresponding pre-trained deep network model is selected, and the extracted time-frequency domain features are used as input of the selected deep network model to perform a second-layer classification to identify the state of the electric load process.

[0013] Furthermore, the MiniRocket algorithm is used to extract time series features to generate feature vectors. The voltage signal and current signal in the sampled measured power signal are convolved with random convolution kernels to obtain convolution results, and the convolution results are subjected to maximum pooling to obtain current feature vectors and voltage feature vectors. The extracted current feature vectors and voltage feature vectors are concatenated to form the final feature vector.

[0014] Furthermore, in the first-level classification of the extracted feature vector using the pre-trained classifier, a Ridge classifier is used for classification. In the process of pre-training the Ridge classifier, the weight vector w is obtained by minimizing the following loss function:

[0015]

[0016] Where N is the number of samples, xi is the feature vector of the i-th sample, yi is the label of the i-th sample to indicate the charging type and charging standard, w is the weight vector of the classifier, and λ is the regularization parameter.

[0017] Furthermore, the sampled measured power signal is subjected to S-transform, and the time-frequency domain feature matrix is ​​extracted as the input of the deep network model for secondary classification. Each element in the matrix represents the signal strength or energy distribution at each time point and frequency point.

[0018] Furthermore, it also includes standardizing the time-frequency domain feature matrix obtained by S transformation to make the scales of various time-frequency features consistent.

[0019] Furthermore, the state of the electric load process includes any one of a charging state, a discharging state and a standby state.

[0020] Furthermore, the charging type includes a charging method type and a charging standard, and the charging method type includes a fast and slow type, a wireless charging type, and a battery replacement mode, among which the fast and slow types include slow charging of AC charging and fast charging of DC charging. The charging time of the slow charging exceeds a specified time threshold, and the charging speed of the fast charging exceeds a specified speed threshold. The battery replacement mode is to achieve fast charging through battery replacement, and the charging standards include GB standards, IEC / ISO standards, SAE standards, and CHAdeMO standards.

[0021] Furthermore, the corresponding pre-trained deep network model is selected according to the first-layer classification result, and the extracted time-frequency domain features are used as input of the selected deep network model to perform the second-layer classification, and the state of the electric load process is identified, including:

[0022] A training data set is formed using sample data containing current and voltage at a specified frequency. Each sample identifies the corresponding charging type. The training data set is input into a deep network model, and multiple rounds of iterative training are performed to obtain a trained deep network model.

[0023] Pre-train multiple deep network models corresponding to different charging types, and establish a mapping relationship between each deep network model and the first-layer classification results;

[0024] According to the classification results output by the first-layer classifier, the corresponding model is loaded from the stored multiple deep network models according to the mapping relationship, and the loaded model is used and the extracted time-frequency domain features are used as the input of the model for second-layer classification to obtain the recognition result of the load state in the electric load process.

[0025] An electric vehicle load identification device based on double-layer classification, comprising:

[0026] The sampling module is used to perform ADC sampling on the voltage and current signals of the main power supply to obtain the measured power signal:

[0027] A feature extraction module is used to extract features from the sampled measured power signal to generate a feature vector;

[0028] A first-layer classification module, used to perform first-layer classification on the extracted feature vector using a pre-trained classifier, identify the charging type corresponding to the charging process, and obtain a first-layer classification result;

[0029] The time-frequency analysis module is used to perform time-frequency analysis on the sampled measured power signal and extract the time-frequency domain features;

[0030] The second layer classification module is used to select the corresponding pre-trained deep network model according to the first layer classification result, and use the extracted time-frequency domain features as the input of the selected deep network model to perform second layer classification and identify the state of the electric load process.

[0031] An electronic device comprises a processor and a memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to perform the above method.

[0032] A computer-readable storage medium storing a computer program, wherein the computer program implements the above method when executed by a processor.

[0033] Compared with the prior art, the advantages of the present invention are: the present invention adopts a double-layer classification method, first samples the voltage and current signals of the power supply, extracts the feature vector and then performs the first-layer classification, preliminarily identifies the charging type to which the charging process belongs, and then uses the first-layer classification result to select the applicable deep network model, performs time-frequency analysis on the signal to obtain the time-frequency features, and then performs the second-layer classification based on the selected deep network model, so as to realize the fine identification of the load state of the electric vehicle. The double-layer classification method can be fully combined to quickly and accurately realize the load identification of electric vehicles in different scenarios, and can be flexibly applied to various electric vehicle charging scenarios, effectively improving the accuracy and reliability of load identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a schematic diagram of the implementation flow of the electric vehicle load identification method based on double-layer classification in this embodiment.

[0035] Figure 2 It is a waveform diagram of a power signal sampled in a specific application embodiment.

[0036] Figure 3 It is a detailed implementation flow diagram of electric vehicle load identification based on double-layer classification in a specific application embodiment of the present invention. DETAILED DESCRIPTION

[0037] The present invention is further described below in conjunction with the accompanying drawings and specific preferred embodiments, but the protection scope of the present invention is not limited thereby.

[0038] Taking into account the different characteristics of the load state of electric vehicles under different charging types, the applicable identification models are different. The use of a unified identification model cannot ensure accurate identification of the load state of electric vehicles in different scenarios. The present invention adopts a double-layer classification method to first sample the voltage and current signals of the power supply, extract the feature vector, and then perform the first-layer classification to preliminarily identify the charging type to which the charging process belongs. Then, the first-layer classification result is used to select the applicable deep network model, and the time-frequency characteristics obtained by time-frequency analysis of the signal are then subjected to the second-layer classification based on the selected deep network model to achieve fine identification of the load state of the electric vehicle. The double-layer classification method can be fully combined to quickly and accurately realize the load identification of electric vehicles in different scenarios, and can be flexibly applied to various electric vehicle charging scenarios to provide accurate load identification capabilities, thereby providing reliable data and decision-making basis for power safety and power regulation.

[0039] like Figure 1 As shown, the steps of the electric vehicle load identification method based on double-layer classification in this embodiment include:

[0040] Step S01: Perform ADC sampling on the voltage and current signals of the main power supply to obtain a measured power signal.

[0041] In this embodiment, in the process of ADC sampling based on the current and voltage of the total power supply, the sampling frequency can be configured to be no less than 256000Hz, and the sampling accuracy can be no less than 0.5%. In a specific application embodiment, the current and voltage image obtained by sampling at a sampling frequency of 256k is as follows: Figure 2 As shown, the blue line is the voltage signal and the yellow line is the current signal.

[0042] In a specific application embodiment, in a household application scenario, the total power source corresponds to the main power supply line at the rear end of the user's electric meter, and for an electric vehicle charging pile application scenario, the total power source corresponds to the starting end of the dedicated charging pile line drawn from the power distribution room. The acquisition of the total power source can be determined according to different application scenarios.

[0043] Step S02: extracting features from the sampled measured power signal to generate a feature vector.

[0044] In this embodiment, the MiniRocket algorithm can be used to extract time series features from the collected voltage and current data to generate feature vectors. The key to the MiniRocket algorithm is to use a variety of convolution kernels for feature extraction and to generate a variety of convolution feature maps, so as to capture the load change patterns of different types of equipment. The MiniRocket algorithm can capture different types of signal patterns by randomly generating multiple convolution kernels and performing convolution operations on each convolution kernel respectively. For example, different convolution kernel lengths, offsets, and weight initializations can help the algorithm capture the change patterns of different time scales in the power signal. Different modes of power signals (such as switches, load changes, etc.) can be quickly extracted through a variety of convolution kernels. By using the MiniRocket algorithm to extract feature vectors, this embodiment can more comprehensively extract the features of the power signal, thereby enhancing the accuracy and robustness of load identification.

[0045] In this embodiment, the MiniRocket algorithm is applied to the voltage and current data respectively, and the time series features are automatically extracted by pre-defining the parameters of the random convolution kernel (convolution length, offset) to generate a feature vector. Assuming that the voltage signal and the current signal are V(t) and I(t) respectively, the MiniRocket algorithm is used to perform a convolution operation on the two signals to obtain a set of convolution feature vectors, including a current feature vector and a voltage feature vector, and then the extracted current feature vector and the voltage feature vector are spliced ​​to form the final feature vector.

[0046] Specifically, for each signal x(t)∈{V(t),I(t)}, the convolution process is as follows:

[0047] Step S101. Initialization configuration

[0048] In this embodiment, the convolution kernel size, weight and bias in the MiniRocket algorithm are partially fixed and no longer completely random. For example, the kernel size is configured to 9, the weight is randomly selected from -1 and 2 at initialization, the bias depends on the length of the input time series, but there are fixed rules, and after the convolution operation, the 0.25, 0.5 and 0.75 quantiles are calculated from the output sequence as candidate values ​​for the bias.

[0049] Step S102: Convolve the voltage signal V(t) and the current signal I(t) using random convolution kernels respectively to obtain corresponding convolution results.

[0050] The calculation expression for convolution of voltage signal V(t) and current signal I(t) can be expressed as:

[0051]

[0052] Among them, y v(t) ,y I(t) Respectively represent the convolution results of voltage signal and current signal, w i V 、w i I They represent the weight values ​​of the voltage signal and the current signal respectively, and L represents the length of the convolution kernel.

[0053] Step S103: Perform maximum pooling on the convolution result to extract the most significant feature y max,V ,y max,I , and obtain the voltage and current eigenvectors respectively.

[0054] The calculation expressions for the maximum pooling of the convolution results of voltage and current are:

[0055] y max,V = max(y v(t) ) (3)

[0056] y max,I = max(y I(t) ) (4)

[0057] Step S104: merge the extracted voltage and current feature vectors to form a final feature vector f:

[0058] f = [y max,V ,y max,I ] (5)

[0059] Step S03: Use a pre-trained classifier to perform a first-level classification on the extracted feature vector, identify the charging type corresponding to the charging process, and obtain a first-level classification result.

[0060] The final feature vector f formed by concatenating the voltage and current feature vectors generated in step S02 is used as the input of the classifier for the first-level classification, and the charging type corresponding to the charging process is preliminarily identified. Considering that the diversity of electric vehicle charging is reflected in multiple aspects such as charging methods, standards, scenarios, and equipment, different charging types correspond to different load characteristics, and different applicable recognition models. This embodiment first uses the first-level classification to preliminarily classify the charging type corresponding to the charging process, and selects the deep network model required for the subsequent second-level classification based on the classification results to ensure the accuracy of recognition.

[0061] Different charging methods and charging standards are applicable to different identification models. In this embodiment, the charging type may specifically include the charging method type and the charging standard. The charging method type includes fast and slow types, wireless charging types, and battery replacement modes, among which the fast and slow types include slow charging of AC charging and fast charging of DC charging. The charging time of slow charging exceeds the specified time threshold, and the charging speed of fast charging exceeds the specified speed threshold. The battery replacement mode is to achieve fast charging through battery replacement. Charging standards include GB standards, IEC / ISO standards, SAE standards, and CHAdeMO standards, etc.

[0062] Specifically, charging types can be divided as follows:

[0063] (1) According to the charging method of electric vehicles, including:

[0064] Slow charging (AC charging): For example, the power range is 3.7kW to 22kW, which is suitable for home use and long-term parking scenarios, and the charging time is longer;

[0065] Fast charging (DC charging): For example, the power range is 50kW to 350kW, the charging speed is fast, and it is suitable for highway service areas and public places;

[0066] Wireless charging: Based on electromagnetic induction or magnetic resonance technology, it does not require physical contact and is suitable for smart charging scenarios;

[0067] Battery swap mode: fast "charging" through battery replacement, suitable for high-frequency usage scenarios such as shared cars and logistics vehicles.

[0068] (2) According to charging standards, including:

[0069] GB standard: includes both AC and DC charging interfaces;

[0070] IEC (International Electrotechnical Commission) / ISO (International Organization for Standardization) standards: mainly using Type 2 and CCS interfaces, with wide adaptability;

[0071] SAE (Society of Automotive Engineers) standard: Use Type 1 and CCS interfaces, support special interfaces such as Tesla;

[0072] CHAdeMO standard: DC fast charging standard.

[0073] Furthermore, it can be divided according to charging scenarios, for example, home scenarios (such as home charging piles and wall-mounted charging equipment), public scenarios (urban fast charging networks, serving short-term parking needs), highway scenarios (high-power charging piles, meeting long-distance driving power replenishment needs) and other special scenarios (such as dedicated charging facilities for heavy vehicles such as buses and trucks). It can also be divided according to the form of charging equipment, including wall-mounted, floor-standing, super charging piles and mobile charging equipment, etc., to further improve the accuracy of load identification.

[0074] In this embodiment, a Ridge classifier can be used for the first level of classification to identify the charging type of the charging process and achieve preliminary classification. The Ridge classifier is a linear classification model with L2 regularization, and its goal is to perform classification by minimizing the following loss function. Specifically, in the process of pre-training the Ridge classifier, the weight vector w is obtained by minimizing the following loss function:

[0075]

[0076] Where N is the number of samples, xi is the feature vector of the i-th sample, yi is the label of the i-th sample to indicate the charging type, w is the weight vector of the classifier, and λ is the regularization parameter used to control the complexity of the model to prevent overfitting.

[0077] After the Ridge classifier is trained, the newly input extended feature vector X is predicted by the trained Ridge classifier to obtain the classification result of the charging process. The classification result may include the charging mode type and the charging standard.

[0078] Step S04: Perform time-frequency analysis on the sampled measured power signal to extract time-frequency domain features.

[0079] This embodiment further performs time-frequency analysis on the sampled measured power signal to extract new time-frequency domain features. Specifically, the sampled measured power signal can be subjected to S transformation to extract the time-frequency domain feature matrix as the input of the deep network model for secondary classification, and each element in the matrix represents the signal strength or energy distribution at each time point and frequency point. S-Transform is a time-frequency analysis method that combines the advantages of short-time Fourier transform (STFT) and wavelet transform, and can provide high-precision time-frequency localization information. When an electric vehicle is charging, the current signal is usually non-stationary and has a complex time-varying frequency, but the signal may be affected by a variety of factors, such as equipment switching, load changes, seasonal changes, etc., which make the load signal present dynamic characteristics in frequency and time domain. Traditional time-frequency analysis methods (such as STFT) may not be able to capture the above characteristics well, and S-transform has significant advantages in processing non-stationary signals (for example, power consumption signals). This embodiment uses S transform to extract time-frequency features as model input. By adopting S transform, it can provide more detailed time-frequency localization information, thereby improving the accuracy of load decomposition and prediction, especially when facing complex power loads with time-varying frequencies. It can fully mine detailed time-frequency localization information, thereby effectively improving the accuracy of load identification.

[0080] Specifically, the S transformation can be performed according to the following formula:

[0081]

[0082] Where S(t,f) is the S transform of the signal x(t) at time t and frequency f, x(τ) is the input power signal, τ is the time variable, exp(-2πif(τ-t)) is the phase factor related to frequency f, exp(-2σ2(f)(τ-t)2) is the Gaussian window function, where σ(f) is the window width related to frequency. The width of the window function varies with frequency, with the low-frequency part being wider and the high-frequency part being narrower.

[0083] Through the above S-transformation, a two-dimensional matrix can be obtained, which represents the signal strength or energy distribution at each time point and frequency point. The time-frequency domain feature matrix can be used as the input of the deep learning model in the subsequent second-layer classification, which can help the model better identify and separate the load signal. Furthermore, it also includes standardizing the time-frequency domain feature matrix obtained by the S-transformation to make the scale of each time-frequency feature consistent, further ensuring the accuracy of recognition.

[0084] Step S05. Select a corresponding pre-trained deep network model according to the first-layer classification result, and use the extracted time-frequency domain features as the input of the selected deep network model to perform the second-layer classification to identify the state of the electric load process.

[0085] This embodiment pre-trains different deep network models for different charging types. For example, a training set is formed by collecting data under different charging methods and charging standards in advance, and then the deep network models are trained using the training sets to form deep network models suitable for different charging methods and charging standards. When performing the second-level classification, the corresponding pre-trained deep network model is selected according to the first-level classification result. For example, the corresponding deep network model is selected according to the charging method and charging standard obtained from the first-level classification result, and then the time-frequency features extracted in step S04 are used as model input for the second-level classification, which can realize the detailed classification of electric vehicle loads and refine the identification of specific load states during the electric vehicle load process, such as electric vehicle charging, discharging, standby, etc.

[0086] In a specific application embodiment, TimesNet can be used for the second layer of classification, and the time-frequency features can be used as the input of the TimesNet deep learning model to refine and identify the specific load characteristics of the charging process. TimesNet is a deep learning model for time series analysis. It is based on the observation of the multi-periodicity of time series, decomposing complex time changes into multiple intra-cycle and inter-cycle changes. By converting a one-dimensional time series into a two-dimensional tensor based on multiple cycles, and using the advantages of a 2D convolutional network to analyze time series data, it can be applied to tasks such as prediction, interpolation, classification, and anomaly detection to achieve accurate time series analysis. This embodiment combines TimesNet for the second layer of classification, which can refine and identify the specific load characteristics of the charging process as much as possible, obtain the specific load state of the electric vehicle during the load process, and ensure the accuracy and reliability of the recognition.

[0087] Specifically, the second-level classification can be performed according to the following steps:

[0088] Step S501. Model training

[0089] The training data set is composed of sample data of current and voltage at a specified frequency (such as 256kHz). Each sample identifies a specific charging type (normal charging, fast charging, discharging, standby, etc.). The training data set is input into TimesNet, and the data is divided according to the set batch size (128). Multiple rounds of iterative training are performed to obtain the trained TimesNet model. In each round of training, the prediction results of the model are calculated by forward propagation, and the loss value is calculated according to the loss function. Then, the optimizer is used for back propagation to update the parameters of the network until the preset stop condition is reached (such as the number of model training times reaches the preset threshold or the accuracy is no longer improved).

[0090] Step S502: Mapping relationship construction

[0091] According to step S401, multiple different TimesNet models are pre-trained, and a mapping relationship between each TimesNet model and the first-layer classification result (charging method type and corresponding charging standard, etc.) is established. For example, if the first-layer classification result identifies the "fast charging" type, it corresponds to the TimesNet model pre-trained for the fast charging situation. If it is the "discharge" type, it corresponds to the TimesNet model pre-trained under the corresponding discharge scenario.

[0092] Step S503. Loading based on rules

[0093] According to the classification results output by the first-layer classifier, the corresponding TimesNet model is dynamically loaded from the multiple stored pre-trained models into the memory according to the mapping relationship determined in step S502, and the loaded TimesNet model is used and the extracted time-frequency domain features are used as the input of the model to perform the second-layer refined classification, so as to obtain the recognition result of the load state in the electric load process.

[0094] In a specific application embodiment, Figure 3 As shown, in the process of realizing electric vehicle load identification by the above method, the present invention first performs ADC sampling based on the voltage and current signals of the total power supply, applies the MiniRocket algorithm to the voltage and current data respectively, automatically extracts time series features by pre-defining the parameters of the random convolution kernel (convolution length, offset), generates feature vectors, and inputs the generated voltage and current feature vectors into the Ridge classifier for the first layer classification to identify the charging method (such as fast charging) and charging standard corresponding to the charging process. If the specific charging type is not identified, it returns to data sampling for re-identification; if the specific charging type is identified, the S transform is further used to extract the time-frequency features as the input of the deep model, and the type of the deep model is identified according to the first layer classification results, and then TimesNet is used for the second layer classification to refine and identify the specific load characteristics of the charging process, and obtain the identification results of the load states such as charging, discharging, and standby of the electric vehicle. If the load state cannot be identified, it returns to data sampling for re-identification. Through the above method, the double-layer classification method can be fully combined to quickly and accurately realize the electric vehicle load identification in different scenarios.

[0095] The electric vehicle load identification device based on double-layer classification in this embodiment includes:

[0096] The sampling module is used to perform ADC sampling on the voltage and current signals of the main power supply to obtain the measured power signal:

[0097] A feature extraction module is used to extract features from the sampled measured power signal to generate a feature vector;

[0098] The first-layer classification module is used to perform the first-layer classification on the extracted feature vector using a pre-trained classifier, identify the charging type and the corresponding charging standard type corresponding to the charging process, and obtain the first-layer classification result;

[0099] The time-frequency analysis module is used to perform time-frequency analysis on the sampled measured power signal and extract the time-frequency domain features;

[0100] The second-layer classification module is used to select the corresponding pre-trained deep network model according to the first-layer classification results, and use the extracted time-frequency domain features as the input of the selected deep network model for second-layer classification to identify the state of the electric load process.

[0101] The electric vehicle load identification device based on double-layer classification in this embodiment corresponds one to one with the above-mentioned electric vehicle load identification method based on double-layer classification, and will not be described one by one here.

[0102] This embodiment further provides an electronic device, including a processor and a memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to perform the above method.

[0103] It is understandable that the above method of this embodiment can be executed by a single device, such as a computer or server, etc., and can also be applied to a distributed scenario and completed by multiple devices in cooperation with each other. In the case of a distributed scenario, one of the multiple devices can only execute one or more steps in the above method of this embodiment, and multiple devices interact to complete the above method. The processor can be implemented in the form of a general-purpose CPU, a microprocessor, an application-specific integrated circuit, or one or more integrated circuits, etc., for executing related programs to implement the above method of this embodiment. The memory can be implemented in the form of a read-only memory ROM, a random access memory RAM, a static storage device, and a dynamic storage device. The memory can store an operating system and other applications. When the above method of this embodiment is implemented by software or firmware, the relevant program code is stored in the memory and called and executed by the processor.

[0104] This embodiment further provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the above method is implemented.

[0105] Those skilled in the art will appreciate that the above-mentioned embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, may be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the functions in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including an instruction device, which implements the functions specified in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0106] The above is only a preferred embodiment of the present invention, and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment, it is not intended to limit the present invention. Therefore, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the scope of protection of the technical solution of the present invention.

Claims

1. A method for identifying electric vehicle load based on double-layer classification, characterized in that the steps include: The voltage and current signals of the total power supply are sampled by ADC to obtain the measured power signal: Extracting features of the sampled measured power signal to generate a feature vector; Using a pre-trained classifier to perform a first-level classification on the extracted feature vector, identifying the charging type corresponding to the charging process, and obtaining a first-level classification result; Perform time-frequency analysis on the sampled power signal to extract time-frequency domain features; According to the first-layer classification result, a corresponding pre-trained deep network model is selected, and the extracted time-frequency domain features are used as input of the selected deep network model to perform a second-layer classification to identify the state of the electric load process.

2. The electric vehicle load identification method based on double-layer classification according to claim 1 is characterized in that: The MiniRocket algorithm is used to extract time series features to generate feature vectors. The voltage signal and current signal in the sampled measured power signal are convolved with random convolution kernels to obtain convolution results, and the convolution results are subjected to maximum pooling to obtain current feature vectors and voltage feature vectors. The extracted current feature vector and voltage feature vector are concatenated to form the final feature vector.

3. The electric vehicle load identification method based on double-layer classification according to claim 1 is characterized in that: In the first-level classification of the extracted feature vector using the pre-trained classifier, the Ridge classifier is used for classification. In the process of pre-training the Ridge classifier, the weight vector w is obtained by minimizing the following loss function: Where N is the number of samples, xi is the feature vector of the i-th sample, yi is the label of the i-th sample to indicate the charging type, w is the weight vector of the classifier, and λ is the regularization parameter.

4. The electric vehicle load identification method based on double-layer classification according to claim 1 is characterized in that: The sampled measured power signal is subjected to S transformation, and the time-frequency domain feature matrix is ​​extracted as the input of the deep network model for secondary classification. Each element in the matrix represents the signal strength or energy distribution at each time point and frequency point.

5. The electric vehicle load identification method based on double-layer classification according to claim 1 is characterized in that: The state of the electric load process includes any one of a charging state, a discharging state and a standby state.

6. The electric vehicle load identification method based on double-layer classification according to any one of claims 1 to 5, characterized in that: The charging type includes a charging method type and a charging standard. The charging method type includes a fast and slow type, a wireless charging type, and a battery replacement mode, among which the fast and slow types include slow charging of AC charging and fast charging of DC charging. The charging time of the slow charging exceeds a specified time threshold, and the charging speed of the fast charging exceeds a specified speed threshold. The battery replacement mode is to achieve fast charging through battery replacement. The charging standards include GB standards, IEC / ISO standards, SAE standards, and CHAdeMO standards.

7. The electric vehicle load identification method based on double-layer classification according to any one of claims 1 to 5, characterized in that: The method of selecting a corresponding pre-trained deep network model according to the first-layer classification result, and using the extracted time-frequency domain features as input of the selected deep network model for second-layer classification, and identifying the state of the electric load process includes: A training data set is formed using sample data containing current and voltage at a specified frequency. Each sample identifies the corresponding charging type. The training data set is input into a deep network model, and multiple rounds of iterative training are performed to obtain a trained deep network model. Pre-train multiple deep network models corresponding to different charging types, and establish a mapping relationship between each deep network model and the first-layer classification results; According to the classification results output by the first-layer classifier, the corresponding model is loaded from the stored multiple deep network models according to the mapping relationship, and the loaded model is used and the extracted time-frequency domain features are used as the input of the model for second-layer classification to obtain the recognition result of the load state in the electric load process.

8. An electric vehicle load identification device based on double-layer classification, characterized in that: include: The sampling module is used to perform ADC sampling on the voltage and current signals of the main power supply to obtain the measured power signal: A feature extraction module is used to extract features from the sampled measured power signal to generate a feature vector; A first-layer classification module, used to perform first-layer classification on the extracted feature vector using a pre-trained classifier, identify the charging type corresponding to the charging process, and obtain a first-layer classification result; The time-frequency analysis module is used to perform time-frequency analysis on the sampled measured power signal and extract the time-frequency domain features; The second layer classification module is used to select the corresponding pre-trained deep network model according to the first layer classification result, and use the extracted time-frequency domain features as the input of the selected deep network model to perform second layer classification and identify the state of the electric load process.

9. An electronic device comprising a processor and a memory, wherein the memory is used to store a computer program, wherein: The processor is configured to execute the computer program to perform the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images