A measurement error correction method for electric vehicle DC high-power charging pile

Abstract

The present invention provides a method for correcting the metering error of a DC high-power charging pile for an electric vehicle, which belongs to the technical field of electrical variable measurement, and comprises: first, collecting historical charging data of the same model of charging piles, including input parameters (grid voltage and current) and output parameters (DC voltage and current). Then, using fast Fourier transform to analyze the grid voltage and DC output voltage waveforms, and extracting the standard DC coefficient matrix and the fluctuating DC coefficient matrix. Next, establishing a training data set, including charging parameters, coefficient matrix and power error. Using embedded multi-branch neural network training to obtain a metering error model. Finally, for the charging pile to be tested, using the error model to correct the power metering. The present invention can effectively analyze the power quality during the charging process, identify the metering error, and solve the technical problem of large errors in the metering of DC high-power charging piles for electric vehicles in the prior art.

Description

Technical Field

[0001] The invention belongs to the technical field of electric variable measurement, and in particular relates to a method for correcting measurement errors of a DC high-power charging pile for an electric vehicle. Background Art

[0002] With the continuous development of electric vehicle technology and the continuous improvement of social awareness of energy conservation and environmental protection, electric vehicles are playing an increasingly important role in urban transportation. As a representative of clean energy vehicles, electric vehicles can not only significantly reduce carbon emissions, but also reduce noise pollution, playing an important role in urban environmental protection. The construction of charging infrastructure is the key to promoting the popularization of electric vehicles.

[0003] At present, electric vehicle charging piles are mainly divided into two types: AC slow charging and DC fast charging. AC slow charging is suitable for occasions such as homes or parking lots, and the charging power is relatively low, generally between a few kilowatts and 20,000 watts; while DC fast charging is suitable for public places such as gas stations and shopping malls, and the power can reach tens of kilowatts, greatly shortening the charging time and meeting the needs of electric vehicle owners. With the continuous increase in the number of electric vehicles, DC high-power charging piles will surely become the mainstream of charging infrastructure in the future.

[0004] However, there are some technical challenges in the actual application of DC high-power charging piles. The first is the accuracy of charging metering. Due to the influence of factors such as grid voltage fluctuations and DC voltage pulsation during the charging process, there is a certain error in the charging power. This error not only affects the user's charging experience, but may also cause economic disputes between electric vehicle companies and charging operators. The second is the safety of the battery during the charging process. Under high-power charging, the battery will generate a lot of heat. If the heat is not managed well, it is likely to cause safety hazards. The third is the problem of charging efficiency. During high-power charging, the decrease in grid power factor and the power consumption of switching devices will reduce charging efficiency and shorten battery life.

[0005] In summary, the existing technology has a technical problem of large errors in the measurement of DC high-power charging piles for electric vehicles. Summary of the invention

[0006] In view of this, the present invention provides a method for correcting measurement errors of a DC high-power charging pile for an electric vehicle, which can solve the technical problem of large measurement errors in the prior art for DC high-power charging piles for electric vehicles.

[0007] The present invention is achieved in that:

[0008] The present invention provides a method for correcting measurement errors of a DC high-power charging pile for an electric vehicle, which comprises the following steps:

[0009] S10, obtaining charging parameters collected by precise sensors for multiple groups of historical charging processes of multiple DC high-power charging piles of the same model of electric vehicles, including input parameters and output parameters; wherein the input parameters are the electric energy parameters input from the power grid when the charging piles are connected to the power grid, including the instantaneous voltage of the power grid and the instantaneous current of the power grid; the output parameters include the instantaneous DC voltage, the instantaneous DC current, and the instantaneous electricity meter;

[0010] S20, according to the preset sliding time window, establish a grid instantaneous voltage curve according to the grid instantaneous voltage, and perform a fast Fourier transform to obtain a set of split sinusoidal curves of each time window of the grid instantaneous voltage curve, recorded as a grid voltage sinusoidal curve set;

[0011] S30, according to the preset sliding time window, the fluctuating part of the DC instantaneous voltage is screened according to the accumulated situation of the DC instantaneous voltage, a fluctuating voltage curve is established, and a fast Fourier transform is performed to obtain a set of split sinusoidal curves of each time window of the fluctuating voltage curve, which is recorded as a fluctuating voltage sinusoidal curve set;

[0012] S40, for each sliding time window, according to the grid voltage sinusoidal set and the fluctuating voltage sinusoidal set, calculate the standard DC coefficient matrix and the fluctuating DC coefficient matrix of the DC power output by the charging pile; the standard DC coefficient matrix is ​​used to calculate the ideal DC output; the fluctuating DC coefficient matrix is ​​used to analyze the fluctuation of the DC output;

[0013] S50, for each sliding time window, using an integral method, the instantaneous voltage and instantaneous current are used to calculate the power, obtain the theoretical power, and calculate the power error according to the corresponding instantaneous power measurement;

[0014] S60, establishing a training data set, including the charging parameters of each sliding time window corresponding to the charging parameters of multiple groups of historical charging processes of multiple electric vehicle DC high-power charging piles, as well as the standard DC coefficient matrix, the fluctuating DC coefficient matrix and the power error; wherein the training input is the charging parameter, the training output is the power error, and the intermediate data of the training is the standard DC coefficient matrix and the fluctuating DC coefficient matrix;

[0015] S70, using the training data set to train an embedded multi-branch neural network structure to obtain a measurement error model;

[0016] S80, according to the preset sliding time window, obtaining the input parameters and output parameters of the electric vehicle DC high-power charging pile to be error corrected, and using the measurement error model to obtain the power error;

[0017] S90: using the power error to correct the measurement of the electric vehicle DC high-power charging pile to be corrected.

[0018] On the basis of the above technical solution, the measurement error correction method of a DC high-power charging pile for electric vehicles of the present invention can also be improved as follows:

[0019] Among them, the embedded multi-branch neural network structure includes a standard DC coefficient matrix branch network, a fluctuating DC coefficient matrix branch network, a long-term data branch network, a single time window data branch network and a fusion branch network.

[0020] Furthermore, the input of the standard DC coefficient matrix branch network is the charging parameter, and the output is the corresponding standard DC coefficient matrix; the input of the fluctuating DC coefficient matrix branch network is the charging parameter, and the output is the corresponding fluctuating DC coefficient matrix; the input of the long-term data branch network is the standard DC coefficient matrix and the fluctuating DC coefficient matrix of all continuous time windows output by the standard DC coefficient matrix branch network and the fluctuating DC coefficient matrix branch network, and the output is the accumulated power error of the corresponding multiple time windows; the input of the single time window data branch network is the standard DC coefficient matrix and the fluctuating DC coefficient matrix of each continuous time window output by the standard DC coefficient matrix branch network and the fluctuating DC coefficient matrix branch network, and the output is the power error of the corresponding time window; the input of the fusion branch network is the accumulated power error output by the long-term data branch network and the power errors of each time window output by the single time window data branch network, and the output is the final power error correction value.

[0021] Furthermore, the standard DC coefficient matrix branch network adopts a fuzzy neural network structure including a fuzzy layer, and the specific structure is:

[0022] Input layer: receives charging parameters;

[0023] Fuzzy layer: fuzzification using Gaussian membership function;

[0024] Rule layer: using Mamdani fuzzy reasoning rules;

[0025] Deblurring layer: Deblurring is performed using the centroid method;

[0026] Output layer: Output standard DC coefficient matrix.

[0027] Furthermore, the fluctuating DC coefficient matrix branch network adopts an adaptive neuro-fuzzy inference system including a fuzzy layer, and the specific structure is:

[0028] Input layer: receives charging parameters;

[0029] Fuzzy layer: Use bell-shaped membership function;

[0030] Rule layer: adopt Takagi-Sugeno reasoning rules;

[0031] Normalization layer: normalize the output of the regular layer;

[0032] Deblurring layer: linear combination;

[0033] Output layer: Output fluctuation DC coefficient matrix.

[0034] The long-term data branch network has a specific structure:

[0035] Input layer: receiving the standard DC coefficient matrix and the fluctuating DC coefficient matrix of all continuous time windows;

[0036] LSTM layer: handles long-term temporal dependencies;

[0037] Fully connected layer: feature fusion;

[0038] Output layer: Outputs the accumulated power errors of multiple time windows.

[0039] The single time window data branch network has a specific structure:

[0040] Input layer: receives the standard DC coefficient matrix and the fluctuating DC coefficient matrix of each time window;

[0041] One-dimensional convolution layer: extract local features;

[0042] Pooling layer: dimensionality reduction;

[0043] Fully connected layer: feature fusion;

[0044] Output layer: Output the power error of the corresponding time window.

[0045] The specific structure of the fusion branch network is:

[0046] Input layer: the accumulated power error output by the long-term data branch network and the power error of each time window output by the single time window data branch network;

[0047] Attention mechanism layer: weights errors at different time scales;

[0048] Fully connected layer: feature fusion;

[0049] Output layer: Output the final power error correction value.

[0050] The preset sliding time window is 0.05s~1s.

[0051] According to the grid voltage sinusoidal set and the fluctuating voltage sinusoidal set, the formula for calculating the standard DC coefficient matrix and the fluctuating DC coefficient matrix of the DC power output by the charging pile is expressed as follows:

[0052] 1. Standard DC coefficient matrix calculation:

[0053] ;

[0054] In the formula, is the standard DC coefficient matrix; The grid voltage sinusoidal curve is the first The amplitude of the components; For the The angular frequency of the components; For the The phase of the components; is the number of components of the grid voltage sinusoidal set; is the transformation matrix.

[0055] Parameter acquisition method:

[0056] It is obtained by fast Fourier transforming the instantaneous voltage of the power grid. The specific steps are as follows:

[0057] Step 1: Perform discrete Fourier transform on the instantaneous voltage data of the power grid:

[0058] ;in, is the number of sampling points;

[0059] Step 2: Calculate the amplitude, angular frequency and phase:

[0060] ;

[0061] The transformation matrix obtained by the experiment is obtained by the following steps:

[0062] Step 1: Set different input voltage waveforms;

[0063] Step 2: Measure the corresponding output DC voltage;

[0064] Step 3: Use the least squares method to fit the transformation matrix.

[0065] 2. Fluctuation DC coefficient matrix calculation:

[0066] ;

[0067] In the formula, is the fluctuation DC coefficient matrix; The first sine curve of the fluctuating voltage is concentrated The amplitude of the components; For the The angular frequency of the components; For the The phase of the components; is the number of components of the fluctuating voltage sinusoidal set; is the compensation matrix; and are the coefficients and exponents of the complex terms, is the number of plural terms; Is an imaginary unit.

[0068] Parameter acquisition method:

[0069] It is obtained by fast Fourier transforming the fluctuation part of the DC instantaneous voltage. The specific steps are as follows:

[0070] Step 1: Extract the fluctuation part of the DC instantaneous voltage:

[0071] ;

[0072] in, is the DC instantaneous voltage, is the average value of the instantaneous DC voltage.

[0073] Step 2: Perform discrete Fourier transform on the fluctuation part:

[0074] ;

[0075] Step 3: Calculate the amplitude, angular frequency and phase:

[0076] ;

[0077] To obtain the compensation matrix, follow these steps:

[0078] Step 1: Analyze the fluctuation characteristics under different working conditions;

[0079] Step 2: Establish a relationship model between fluctuation characteristics and working conditions;

[0080] Step 3: Use a machine learning algorithm (such as a support vector machine) to train and obtain the compensation matrix.

[0081] and It is obtained by performing nonlinear fitting on the fluctuation characteristics. The specific steps are as follows:

[0082] Step 1: Extract the nonlinear components of the wave characteristics;

[0083] Step 2: Use the Levenberg-Marquardt algorithm for nonlinear fitting:

[0084] ;

[0085] in, is the nonlinear component of the wave characteristics, is the number of sampling points, is the number of plural items.

[0086] Compared with the prior art, the beneficial effects of the measurement error correction method for a DC high-power charging pile for electric vehicles provided by the present invention are as follows: First, the present invention makes full use of a large amount of historical charging data, and extracts two types of key parameters, the standard DC coefficient matrix and the fluctuating DC coefficient matrix, by conducting an in-depth analysis of the spectrum characteristics of the grid voltage and the DC voltage. These parameters can not only reflect the characteristics of voltage fluctuations during the charging process, but also provide strong support for subsequent measurement error modeling. Compared with the existing charging pile method that only relies on instantaneous voltage and current data for simple power calculation, the method of the present invention significantly improves the accuracy of measurement.

[0087] Secondly, the present invention adopts an embedded multi-branch neural network structure to fully explore the complex relationship between different time scales and features. The neural network includes multiple sub-networks such as a standard DC coefficient matrix branch, a fluctuating DC coefficient matrix branch, a long-term data branch, a single time window data branch, and a fusion branch. By training this composite network structure, a more accurate measurement error model can be obtained, which better solves the problem of low measurement accuracy in the prior art.

[0088] Finally, the method proposed in the present invention can not only accurately estimate the error of charging power, but also directly correct the metering results of the charging pile.

[0089] In general, the present invention makes full use of voltage fluctuation characteristic analysis and neural network modeling technology to solve the problem of low metering accuracy of DC high-power charging piles and solves the technical problem of large errors in the metering of DC high-power charging piles for electric vehicles in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] Figure 1 A flow chart of the method provided by the present invention. DETAILED DESCRIPTION

[0091] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0092] like Figure 1FIG. 1 is a flow chart of a method for correcting measurement errors of a DC high-power charging pile for an electric vehicle provided by the present invention. The method comprises the following steps:

[0093] S10, obtaining charging parameters collected by precise sensors for multiple groups of historical charging processes of multiple DC high-power charging piles of the same model of electric vehicles, including input parameters and output parameters; wherein the input parameters are the electric energy parameters input from the power grid when the charging piles are connected to the power grid, including the instantaneous voltage of the power grid and the instantaneous current of the power grid; the output parameters include the instantaneous DC voltage, the instantaneous DC current, and the instantaneous electricity meter;

[0094] S20, according to the preset sliding time window, establish the grid instantaneous voltage curve according to the grid instantaneous voltage, and perform fast Fourier transform to obtain the split sinusoidal curve set of each time window of the grid instantaneous voltage curve, recorded as the grid voltage sinusoidal curve set;

[0095] S30, according to the preset sliding time window, according to the cumulative situation of the DC instantaneous voltage, the fluctuating part of the DC instantaneous voltage is screened, a fluctuating voltage curve is established, and a fast Fourier transform is performed to obtain a set of split sinusoidal curves of each time window of the fluctuating voltage curve, which is recorded as a fluctuating voltage sinusoidal curve set;

[0096] S40. For each sliding time window, the standard DC coefficient matrix and the fluctuating DC coefficient matrix of the DC power output by the charging pile are calculated according to the grid voltage sinusoidal set and the fluctuating voltage sinusoidal set; the standard DC coefficient matrix is ​​used to calculate the ideal DC output; the fluctuating DC coefficient matrix is ​​used to analyze the fluctuation of the DC output;

[0097] S50, for each sliding time window, the instantaneous voltage and instantaneous current are calculated by integration to obtain the theoretical power, and the power error is calculated according to the corresponding instantaneous power measurement;

[0098] S60, establishing a training data set, including charging parameters of each sliding time window corresponding to multiple groups of charging parameters of historical charging processes of multiple electric vehicle DC high-power charging piles, as well as a standard DC coefficient matrix, a fluctuating DC coefficient matrix, and a power error; wherein the training input is the charging parameter, the training output is the power error, and the intermediate data of the training is the standard DC coefficient matrix and the fluctuating DC coefficient matrix;

[0099] S70, using the training data set to train an embedded multi-branch neural network structure to obtain a measurement error model;

[0100] S80, according to the preset sliding time window, obtaining the input parameters and output parameters of the electric vehicle DC high-power charging pile to be error corrected, and using the measurement error model to obtain the power error;

[0101] S90. Correct the measurement of the electric vehicle DC high-power charging pile for error correction by using the power error.

[0102] The specific implementation methods of the above steps are described in detail below:

[0103] The specific implementation method of step S10 is to first obtain multiple sets of historical charging process data of multiple DC high-power charging piles of electric vehicles of the same model. These data include charging parameters collected by precise sensors, where the input parameters are the electric energy parameters input from the power grid when the charging pile is connected to the power grid, including the instantaneous voltage and instantaneous current of the power grid; the output parameters include the instantaneous DC voltage, instantaneous DC current and instantaneous electricity measurement. By collecting a large number of historical charging data of charging piles of the same model, a more comprehensive sample can be obtained, laying the foundation for subsequent feature extraction and model training.

[0104] The specific implementation of step S20 is to first establish a grid instantaneous voltage curve using the acquired grid instantaneous voltage data according to a preset sliding time window. Then, the grid instantaneous voltage curve is subjected to a fast Fourier transform to obtain a grid voltage sinusoidal curve set for each time window. The purpose of this step is to analyze the Fourier spectrum characteristics of the grid voltage and provide a basis for the subsequent voltage fluctuation characteristic analysis.

[0105] The specific implementation method of step S30 is to use the acquired DC instantaneous voltage data to filter out the fluctuation part according to the preset sliding time window and establish a fluctuation voltage curve. Then, the fluctuation voltage curve is subjected to fast Fourier transformation to obtain a set of fluctuation voltage sinusoidal curves in each time window. The purpose of this step is to analyze the fluctuation characteristics of the DC output voltage and provide a basis for subsequent fluctuation analysis.

[0106] The specific implementation of step S40 is that for each sliding time window, the standard DC coefficient matrix and the fluctuating DC coefficient matrix of the DC power output by the charging pile are calculated according to the grid voltage sinusoidal set and the fluctuating voltage sinusoidal set obtained in steps S20 and S30. Among them, the standard DC coefficient matrix is ​​used to calculate the ideal DC output, and the fluctuating DC coefficient matrix is ​​used to analyze the fluctuation of the DC output. The calculation formula of the standard DC coefficient matrix is:

[0107] ;

[0108] in, is the standard DC coefficient matrix; The grid voltage sinusoidal curve is the first The amplitude of the components; For the The angular frequency of the components; For the The phase of the components; is the number of components of the grid voltage sinusoidal set; is the transformation matrix. It is obtained through experiments, and the specific steps are: 1) setting different input voltage waveforms; 2) measuring the corresponding output DC voltage; 3) using the least squares method to fit the conversion matrix.

[0109] The calculation formula of the fluctuation DC coefficient matrix is:

[0110] ;

[0111] in, is the fluctuation DC coefficient matrix; The first is the concentrated sine curve of the fluctuating voltage The amplitude of the components; For the The angular frequency of the components; For the The phase of the components; is the number of components of the fluctuating voltage sinusoidal set; is the compensation matrix; and are the coefficients and exponents of the complex terms, is the number of plural terms; is the imaginary unit. The compensation matrix It is obtained through training with machine learning algorithms (such as support vector machines). The specific steps are: 1) analyzing the fluctuation characteristics under different working conditions; 2) establishing a relationship model between the fluctuation characteristics and working conditions; 3) using machine learning algorithms to train to obtain the compensation matrix.

[0112] The specific implementation of step S50 is that for each sliding time window, the instantaneous voltage and instantaneous current are calculated by integration to obtain the theoretical power. Then the power error is calculated according to the corresponding instantaneous power measurement. The purpose of this step is to obtain the error between the actual power and the theoretical power, and provide reference data for the subsequent error correction model training.

[0113] The specific implementation of step S60 is to establish a training data set. The training data set includes the charging parameters of each sliding time window corresponding to the charging parameters of multiple groups of historical charging processes of multiple electric vehicle DC high-power charging piles, as well as the standard DC coefficient matrix and the fluctuating DC coefficient matrix calculated in step S40 and the power error calculated in step S50. Among them, the training input is the charging parameter, the training output is the power error, and the intermediate data of the training is the standard DC coefficient matrix and the fluctuating DC coefficient matrix. The purpose of this step is to establish the training data set required for the error correction model.

[0114] The specific implementation of step S70 is to use the training data set constructed in step S60 to train an embedded multi-branch neural network structure to obtain a measurement error model. The neural network structure includes the following five branch networks:

[0115] 1. Standard DC coefficient matrix branch network: the input is the charging parameter, and the output is the corresponding standard DC coefficient matrix. The branch network adopts a fuzzy neural network structure, including an input layer, a fuzzy layer, a rule layer, a defuzzification layer, and an output layer.

[0116] 2. Fluctuating DC coefficient matrix branch network: The input is the charging parameter, and the output is the corresponding fluctuating DC coefficient matrix. The branch network adopts the adaptive neuro-fuzzy inference system (ANFIS), which includes input layer, fuzzy layer, rule layer, normalization layer, defuzzification layer and output layer.

[0117] 3. Long-term data branch network: The input is the standard DC coefficient matrix and the fluctuating DC coefficient matrix of all continuous time windows, and the output is the cumulative power error of the corresponding multiple time windows. This branch network uses the LSTM layer to process long-term temporal dependencies and the fully connected layer for feature fusion.

[0118] 4. Single time window data branch network: The input is the standard DC coefficient matrix and the fluctuating DC coefficient matrix of each time window, and the output is the power error of the corresponding time window. This branch network uses a one-dimensional convolutional layer to extract local features, a pooling layer to reduce dimensionality, and a fully connected layer for feature fusion.

[0119] 5. Fusion branch network: The input is the cumulative power error output by the long-term data branch network and the power errors of each time window output by the single time window data branch network, and the output is the final power error correction value. This branch network uses the attention mechanism layer to weight the errors of different time scales, and the fully connected layer to perform feature fusion.

[0120] By training this embedded multi-branch neural network structure, a more accurate measurement error model can be obtained.

[0121] The specific implementation method of step S80 is to obtain the input parameters and output parameters of the electric vehicle DC high-power charging pile to be error corrected according to the preset sliding time window, and then calculate the power error using the measurement error model trained in step S70.

[0122] The specific implementation of step S90 is to use the power error calculated in step S80 to correct the measurement of the electric vehicle DC high-power charging pile to be corrected. Through this step, the error correction of the charging pile measurement can be achieved, and the accuracy of charging measurement can be improved.

[0123] In general, the present invention proposes a method for correcting the measurement error of a DC high-power charging pile for electric vehicles, which mainly includes the following key steps: 1) collecting a large amount of historical charging data and extracting the Fourier characteristics of the grid voltage and DC voltage; 2) calculating the standard DC coefficient matrix and the fluctuating DC coefficient matrix based on Fourier analysis; 3) establishing a measurement error model based on an embedded multi-branch neural network; 4) using the model to correct the measurement of the charging pile to be corrected. This method makes full use of the charging history data, and through in-depth voltage characteristic analysis and neural network models, effectively improves the measurement accuracy of the charging pile.

[0124] Specifically, the principle of the present invention is: by deeply analyzing the fluctuation characteristics of the grid voltage and DC voltage during the charging process, key parameters that can reflect the charging state are extracted, and a high-precision measurement error model is established based on these parameters, thereby achieving effective correction of the measurement results of the charging pile.

[0125] Specifically, first in step S10, the data of multiple DC high-power charging piles of electric vehicles of the same model in the historical charging process are collected, including parameters such as grid instantaneous voltage, grid instantaneous current, DC instantaneous voltage, DC instantaneous current, and instantaneous electric energy measurement. These data lay the foundation for subsequent feature extraction and model training.

[0126] Then, in steps S20 and S30, Fourier analysis is performed on the grid instantaneous voltage and the DC instantaneous voltage, respectively, and the sinusoidal curve set of the grid voltage and the fluctuating sinusoidal curve set of the DC voltage are extracted. These curve sets reflect the spectral characteristics of voltage fluctuations, laying the foundation for calculating the standard DC coefficient matrix and the fluctuating DC coefficient matrix.

[0127] In step S40, the standard DC coefficient matrix and the fluctuating DC coefficient matrix are calculated respectively according to the grid voltage sinusoidal set and the DC voltage fluctuation sinusoidal set. The standard DC coefficient matrix reflects the characteristics of the ideal DC output, while the fluctuating DC coefficient matrix characterizes the fluctuation characteristics of the actual DC output. As key parameters of the charging state, these two matrices provide necessary input features for subsequent error modeling.

[0128] In step S50, the theoretical power is calculated by integration and compared with the actual power obtained by instantaneous electric energy measurement to obtain the power error. This power error is used as the output label of the training sample and provides a reference for subsequent model training.

[0129] With the above-mentioned characteristic parameters and error labels, a training data set is constructed in step S60, which is ready for the model training in step S70. In step S70, an embedded multi-branch neural network structure is adopted, including a standard DC coefficient matrix branch, a fluctuating DC coefficient matrix branch, a long-term data branch, a single time window data branch, and a fusion branch. This composite network can fully explore the complex relationship between different time scales and features, thereby obtaining an accurate measurement error model.

[0130] Finally, in steps S80 and S90, the measurement error model trained in step S70 is used to estimate and correct the error of the new charging pile data. This method based on in-depth voltage feature analysis and complex neural network modeling can effectively improve the measurement accuracy of DC high-power charging piles and provide strong support for the high-quality development of electric vehicle charging services.

[0131] In general, the key to the method of the present invention is: 1) by analyzing the spectral characteristics of the grid voltage and the DC voltage, the key parameters that can reflect the charging state are extracted; 2) based on these parameters, an embedded multi-branch neural network structure is constructed to effectively model the metering error; 3) the trained model is used to estimate and correct the error of new charging data.

[0132] In order to better understand and implement the present invention, a specific embodiment 1 of the present invention is provided below, and the steps of this embodiment 1 are described in detail as follows:

[0133] First, the specific implementation method of step S10 is to obtain multiple sets of charging parameters of the historical charging process of multiple electric vehicle DC high-power charging piles of the same model. These charging parameters include input parameters and output parameters. The input parameters are the electric energy parameters input from the power grid when the charging pile is connected to the power grid, including the instantaneous voltage of the power grid. and grid instantaneous current The output parameters include DC instantaneous voltage , DC instantaneous current And instantaneous electricity metering By collecting a large amount of historical charging data of charging piles of the same model, a more comprehensive sample can be obtained, laying the foundation for subsequent feature extraction and model training.

[0134] Then, the specific implementation of step S20 is based on the preset sliding time window. , using the instantaneous voltage of the power grid The grid instantaneous voltage curve is established based on the data. By performing fast Fourier transform on the grid instantaneous voltage curve, the Fourier series representation of the grid voltage in each time window can be obtained:

[0135] ;

[0136] in, For the The amplitude of the component, For the The angular frequency of the component, For the The phase of the component, is the number of Fourier series components of the grid voltage. By performing Fourier analysis on the instantaneous grid voltage, the frequency spectrum characteristics of the grid voltage can be obtained, providing a basis for the subsequent analysis of voltage fluctuation characteristics.

[0137] The specific implementation method of step S30 is to also use the preset sliding time window , using the obtained DC instantaneous voltage Data, extract the fluctuation part of DC voltage :

[0138] ;

[0139] in, is the average value of the DC instantaneous voltage. Then for this fluctuating voltage curve Perform fast Fourier transform to obtain the Fourier series representation of the fluctuating voltage in each time window:

[0140] ;

[0141] in, For the The amplitude of the component, For the The angular frequency of the component, For the The phase of the component, is the number of Fourier series components of the fluctuating voltage. By performing Fourier analysis on the DC voltage fluctuation, the frequency spectrum characteristics of the DC output voltage can be obtained, providing a basis for subsequent fluctuation analysis.

[0142] The specific implementation method of step S40 is that for each sliding time window , according to the grid voltage sinusoidal set and the fluctuating voltage sinusoidal set obtained in steps S20 and S30, the standard DC coefficient matrix of the DC power output by the charging pile is calculated and the fluctuating DC coefficient matrix .

[0143] Standard DC Coefficient Matrix The calculation formula is:

[0144] ;

[0145] in, The grid voltage sinusoidal curve is the The amplitude of the component, For the The angular frequency of the component, For the The phase of the component, is the number of components of the grid voltage sinusoidal set, is the transformation matrix. The specific steps are: 1) setting different input voltage waveforms; 2) measuring the corresponding output DC voltage; 3) using the least squares method to fit the conversion matrix. Standard DC coefficient matrix Characteristic representing an ideal DC output.

[0146] Fluctuating DC coefficient matrix The calculation formula is:

[0147] ;

[0148] in, The first is the concentrated sine curve of the fluctuating voltage The amplitude of the component, For the The angular frequency of the component, For the The phase of the component, is the number of components of the fluctuating voltage sinusoidal set, is the compensation matrix, and are the coefficients and exponents of the complex terms, is the number of plural terms, is the imaginary unit. The compensation matrix It is obtained through training of machine learning algorithms (such as support vector machines). The specific steps are: 1) analyzing the fluctuation characteristics under different working conditions; 2) establishing a relationship model between the fluctuation characteristics and working conditions; 3) using machine learning algorithms to train and obtain the compensation matrix. Fluctuation DC coefficient matrix Indicates the fluctuation characteristics of DC output.

[0149] By calculating the standard DC coefficient matrix and the fluctuating DC coefficient matrix, the characteristics of the DC power output by the charging pile can be comprehensively analyzed, providing the necessary characteristic parameters for the subsequent error correction model training.

[0150] The specific implementation method of step S50 is that for each sliding time window , using the integral method to calculate the theoretical power :

[0151] ;

[0152] Then according to the corresponding instantaneous electricity measurement , calculate the power error :

[0153] ;

[0154] The purpose of this step is to obtain the difference between actual power and theoretical power and provide reference data for subsequent error correction model training.

[0155] The specific implementation of step S60 is to establish a training data set , including the charging parameters of each sliding time window corresponding to multiple groups of historical charging process charging parameters of multiple electric vehicle DC high-power charging piles, and the standard DC coefficient matrix calculated in step S40 , Fluctuation DC coefficient matrix and the power error calculated in step S50 . Specifically expressed as:

[0156] ;

[0157] in, For the The charging parameters of the samples, is the total number of samples. The purpose of this step is to establish the training data set required for the error correction model.

[0158] The specific implementation of step S70 is to use the training data set constructed in step S60 , train an embedded multi-branch neural network structure to obtain the measurement error model. The neural network structure includes the following 5 branch networks:

[0159] 1. Standard DC coefficient matrix branch network:

[0160] Input layer: receiving charging parameters ;

[0161] Fuzzy layer: fuzzification using Gaussian membership function;

[0162] Rule layer: using Mamdani reasoning rules;

[0163] Deblurring layer: Deblurring is performed using the centroid method;

[0164] Output layer: Output standard DC coefficient matrix ;

[0165] 2. Fluctuating DC coefficient matrix branch network:

[0166] Input layer: receiving charging parameters ;

[0167] Fuzzy layer: Use bell-shaped membership function;

[0168] Rule layer: adopt Takagi-Sugeno reasoning rules;

[0169] Normalization layer: normalize the output of the regular layer;

[0170] Deblurring layer: linear combination;

[0171] Output layer: Output fluctuation DC coefficient matrix ;

[0172] 3. Long-term data branch network:

[0173] Input layer: receives all continuous time windows and ;

[0174] LSTM layer: handles long-term temporal dependencies;

[0175] Fully connected layer: feature fusion;

[0176] Output layer: Output the accumulated power error of multiple time windows ;

[0177] 4. Single time window data branch network:

[0178] Input layer: receives each time window and ;

[0179] One-dimensional convolution layer: extract local features;

[0180] Pooling layer: dimensionality reduction;

[0181] Fully connected layer: feature fusion;

[0182] Output layer: Output the power error of the corresponding time window ;

[0183] 5. Fusion branch network:

[0184] Input Layer: and ;

[0185] Attention mechanism layer: weights errors at different time scales;

[0186] Fully connected layer: feature fusion;

[0187] Output layer: Output the final power error correction value ;

[0188] By training this embedded multi-branch neural network structure, a more accurate measurement error model can be obtained.

[0189] The specific implementation method of step S80 is to use a preset sliding time window , obtain the input parameters of the electric vehicle DC high-power charging pile to be corrected and output parameters , , Then, the power error is calculated using the measurement error model trained in step S70. .

[0190] The specific implementation of step S90 is to use the power error calculated in step S80 ,The measurement of the DC high-power charging pile of the electric vehicle to be corrected is corrected. Through this step, the error correction of the charging pile measurement can be achieved and the accuracy of the charging measurement can be improved.

[0191] In order to further better understand and implement the present invention, the following provides Example 2 of a specific application scenario of the present invention: an electric vehicle company has established a large DC fast charging station in City A, equipped with 30 DC high-power charging piles. In order to ensure the quality of charging services and improve the fairness of charging settlement, it is decided to use the measurement error correction method of the electric vehicle DC high-power charging pile proposed in the present invention to optimize and upgrade these charging piles.

[0192] First, the company's technical team collected historical charging data from 30 charging piles at the charging station over a two-month period, including the instantaneous voltage of the power grid. , grid instantaneous current , DC instantaneous voltage , DC instantaneous current And instantaneous energy measurement After data preprocessing and screening, a total of 1500 sets of valid charging sample data were obtained.

[0193] Next, the technical team performed Fourier analysis on these grid voltage data according to step S20. Specifically, the instantaneous grid voltage of each group of charging samples within a 200ms sliding time window was first analyzed. Perform a discrete Fourier transform:

[0194] ;

[0195] in, Indicates the number of sampling points in the window. Then extract the amplitude from the Fourier transform result , angular frequency and Phase :

[0196] ;

[0197] Through such processing, the grid voltage sinusoidal curve set within the 200ms window of each group of charging samples can be obtained. ,in The value is about 20, preferably 20.

[0198] Similarly, the technical team measured the DC instantaneous voltage in step S30. The data was analyzed for fluctuation characteristics. First, the fluctuation part of the DC voltage was extracted. :

[0199] ;

[0200] in, is the average value of the DC instantaneous voltage. Then Perform discrete Fourier transform to obtain the set of fluctuating voltage sinusoidal curves of each group of charging samples within a 200ms window. ,in The value is about 15, preferably 15.

[0201] In step S40, the technical team calculated the standard DC coefficient matrix based on the above grid voltage sinusoidal set and the fluctuating voltage sinusoidal set. and the fluctuating DC coefficient matrix . Take the group charging sample as an example, its standard DC coefficient matrix The calculation formula is:

[0202] ;

[0203] in, is the grid voltage sinusoidal curve set parameter of this group of samples in the 200ms window, is the transformation matrix. It is obtained through experimental testing. The specific steps are as follows: 1) set different input voltage waveforms; 2) measure the corresponding output DC voltage; 3) use the least squares method to fit .

[0204] Similarly, Fluctuating DC coefficient matrix of group charging samples The calculation formula is:

[0205] ;

[0206] in, is the set parameter of the fluctuating voltage sinusoidal curve of this group of samples in the 200ms window, is the compensation matrix, and are the coefficients and exponents of the complex terms. Compensation matrix It is obtained through training of machine learning algorithms (such as support vector machines). The specific steps are as follows: 1) Analyze the fluctuation characteristics under different working conditions; 2) Establish a relationship model between fluctuation characteristics and working conditions; 3) Use machine learning algorithms to train .

[0207] With the above characteristic parameters and ,The technical team further calculated the theoretical power of each group of charging samples in step S50 and power error :

[0208] ;

[0209] ;

[0210] in, ms is the sliding time window.

[0211] Through the above analysis, the company's technical team finally built a training set containing 1,500 sets of charging sample data. , each set of data includes charging parameters , and the corresponding standard DC coefficient matrix , Fluctuation DC coefficient matrix and power error .

[0212] In step S70, the technical team used an embedded multi-branch neural network structure to train the model. Specifically, the neural network includes 5 branches:

[0213] 1. Standard DC coefficient matrix branch network

[0214] Input layer: receiving charging parameters ;

[0215] Fuzzy layer: fuzzification using Gaussian membership function;

[0216] Rule layer: using Mamdani fuzzy reasoning rules;

[0217] Deblurring layer: Deblurring is performed using the centroid method;

[0218] Output layer: Output standard DC coefficient matrix ;

[0219] 2. Fluctuating DC coefficient matrix branch network

[0220] Input layer: receiving charging parameters ;

[0221] Fuzzy layer: Use bell-shaped membership function;

[0222] Rule layer: adopt Takagi-Sugeno reasoning rules;

[0223] Normalization layer: normalize the output of the regular layer;

[0224] Deblurring layer: linear combination;

[0225] Output layer: Output fluctuation DC coefficient matrix ;

[0226] 3. Long-term data branch network

[0227] Input layer: receives all continuous time windows and ;

[0228] LSTM layer: handles long-term temporal dependencies;

[0229] Fully connected layer: feature fusion;

[0230] Output layer: Output the accumulated power error of multiple time windows ;

[0231] 4. Single time window data branch network

[0232] Input layer: receives each time window and ;

[0233] One-dimensional convolution layer: extract local features;

[0234] Pooling layer: dimensionality reduction;

[0235] Fully connected layer: feature fusion;

[0236] Output layer: Output the power error of the corresponding time window ;

[0237] 5. Fusion Branch Network

[0238] Input Layer: and ;

[0239] Attention mechanism layer: weights errors at different time scales;

[0240] Fully connected layer: feature fusion;

[0241] Output layer: Output the final power error correction value ;

[0242] During the training process, the model optimization goal is to minimize the power error of each sample in the training set. and the final correction value Through iterative optimization, a more accurate measurement error correction model is finally obtained.

[0243] In steps S80 and S90, the company's technical team used the trained model to estimate and correct the error of the real-time charging data of 30 charging piles. For example, within the 200ms window, the charging parameters collected in real time are first , using the standard DC coefficient matrix branch network and the fluctuating DC coefficient matrix branch network to calculate and Then these feature parameters are input into the long-term data branch network and the single time window data branch network to obtain and Finally, the fusion branch network performs weighted fusion on the two error values ​​to give the final power error correction value. .

[0244] Through such error correction, the company found that before the implementation of the method of the present invention, the average measurement error of 30 charging piles was about ±3%, while after the implementation of the method of the present invention, the average measurement error was reduced to within ±1%. This not only greatly improves the fairness of charging settlement and meets the charging needs of users, but also provides effective guarantees for the settlement work between enterprises and charging station operators, avoiding disputes caused by measurement errors.

[0245] In order to further verify the effectiveness of the method of the present invention, the company's technical team also tested the performance of charging piles under different working conditions. The details are as follows:

[0246] 1. Grid voltage fluctuation conditions

[0247] In urban distribution networks, grid voltage often fluctuates to a certain extent due to load changes and other reasons. In order to simulate this situation, the technical team artificially set the grid voltage to fluctuate randomly within the range of ±5% and collected the corresponding charging data. Through the calculation of the error correction model, it was found that even in the case of large fluctuations in grid voltage, the metering error of the charging pile can be controlled within ±1%.

[0248] 2. High power charging conditions

[0249] For high-power DC charging piles, large voltage fluctuations and current ripples will be generated in high-power charging mode. In order to simulate this working condition, the technical team artificially adjusted the output power of the charging pile to vary between 50kW and 150kW. The results show that even in the high-power charging state, the method of the present invention can still effectively correct the metering error and ensure the quality of the charging service.

[0250] 3. Low temperature environment conditions

[0251] In the cold winter, the ambient temperature of the charging pile is low, which may affect the voltage fluctuation characteristics during the charging process. To this end, the technical team conducted a special test in an environment of about -10°C. The test results show that even in a low temperature environment, the method of the present invention can still accurately estimate the measurement error, meeting the application requirements under different environmental conditions.

[0252] Through the test verification of the above three typical working conditions, the company fully affirmed the excellent performance of the method of the present invention in improving the measurement accuracy of DC high-power charging piles. Compared with the existing simple power calculation method, the method of the present invention can deeply analyze the voltage fluctuation characteristics, establish an accurate error correction model, and maintain a high measurement accuracy under various complex working conditions, fully meeting the needs of electric vehicle charging services.

[0253] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for correcting measurement errors of a DC high-power charging pile for electric vehicles, characterized in that: The following steps are involved: S10, obtaining charging parameters collected by precise sensors for multiple groups of historical charging processes of multiple DC high-power charging piles of the same model of electric vehicles, including input parameters and output parameters; wherein the input parameters are the electric energy parameters input from the power grid when the charging piles are connected to the power grid, including the instantaneous voltage of the power grid and the instantaneous current of the power grid; the output parameters include the instantaneous DC voltage, the instantaneous DC current, and the instantaneous electricity meter; S20, according to the preset sliding time window, establish a grid instantaneous voltage curve according to the grid instantaneous voltage, and perform a fast Fourier transform to obtain a set of split sinusoidal curves of each time window of the grid instantaneous voltage curve, recorded as a grid voltage sinusoidal curve set; S30, according to the preset sliding time window, the fluctuating part of the DC instantaneous voltage is screened according to the accumulated situation of the DC instantaneous voltage, a fluctuating voltage curve is established, and a fast Fourier transform is performed to obtain a set of split sinusoidal curves of each time window of the fluctuating voltage curve, which is recorded as a fluctuating voltage sinusoidal curve set; S40, for each sliding time window, according to the grid voltage sinusoidal set and the fluctuating voltage sinusoidal set, calculate the standard DC coefficient matrix and the fluctuating DC coefficient matrix of the DC power output by the charging pile; the standard DC coefficient matrix is ​​used to calculate the ideal DC output; the fluctuating DC coefficient matrix is ​​used to analyze the fluctuation of the DC output; S50, for each sliding time window, using an integral method, the instantaneous voltage and instantaneous current are used to calculate the power, obtain the theoretical power, and calculate the power error according to the corresponding instantaneous power measurement; S60, establishing a training data set, including the charging parameters of each sliding time window corresponding to the charging parameters of multiple groups of historical charging processes of multiple electric vehicle DC high-power charging piles, as well as the standard DC coefficient matrix, the fluctuating DC coefficient matrix and the power error; wherein the training input is the charging parameter, the training output is the power error, and the intermediate data of the training is the standard DC coefficient matrix and the fluctuating DC coefficient matrix; S70, using the training data set to train an embedded multi-branch neural network structure to obtain a measurement error model; S80, according to the preset sliding time window, obtaining the input parameters and output parameters of the electric vehicle DC high-power charging pile to be error corrected, and using the measurement error model to obtain the power error; S90: using the power error to correct the measurement of the electric vehicle DC high-power charging pile to be corrected.

2. The method for correcting measurement errors of a DC high-power charging pile for electric vehicles according to claim 1, characterized in that: The embedded multi-branch neural network structure includes a standard DC coefficient matrix branch network, a fluctuating DC coefficient matrix branch network, a long-term data branch network, a single time window data branch network and a fusion branch network.

3. The measurement error correction method of a DC high-power charging pile for electric vehicles according to claim 2 is characterized in that: The input of the standard DC coefficient matrix branch network is the charging parameters, and the output is the corresponding standard DC coefficient matrix; the input of the fluctuating DC coefficient matrix branch network is the charging parameters, and the output is the corresponding fluctuating DC coefficient matrix; the input of the long-term data branch network is the standard DC coefficient matrix and the fluctuating DC coefficient matrix of all continuous time windows output by the standard DC coefficient matrix branch network and the fluctuating DC coefficient matrix branch network, and the output is the accumulated power errors of the corresponding multiple time windows; the input of the single time window data branch network is the standard DC coefficient matrix and the fluctuating DC coefficient matrix of each continuous time window output by the standard DC coefficient matrix branch network and the fluctuating DC coefficient matrix branch network, and the output is the power error of the corresponding time window; the input of the fusion branch network is the accumulated power error output by the long-term data branch network and the power errors of each time window output by the single time window data branch network, and the output is the final power error correction value.

4. The method for correcting measurement errors of a DC high-power charging pile for electric vehicles according to claim 3, characterized in that: The standard DC coefficient matrix branch network adopts a fuzzy neural network structure including a fuzzy layer, and the specific structure is: Input layer: receives charging parameters; Fuzzy layer: fuzzification using Gaussian membership function; Rule layer: using Mamdani fuzzy reasoning rules; Deblurring layer: Deblurring is performed using the centroid method; Output layer: Output standard DC coefficient matrix.

5. The method for correcting measurement errors of a DC high-power charging pile for electric vehicles according to claim 4, characterized in that: The fluctuating DC coefficient matrix branch network adopts an adaptive neuro-fuzzy inference system including a fuzzy layer, and the specific structure is: Input layer: receives charging parameters; Fuzzy layer: Use bell-shaped membership function; Rule layer: adopt Takagi-Sugeno reasoning rules; Normalization layer: normalize the output of the regular layer; Deblurring layer: linear combination; Output layer: Output fluctuation DC coefficient matrix.

6. The method for correcting measurement errors of a DC high-power charging pile for electric vehicles according to claim 1, characterized in that: The preset sliding time window is 0.05s~1s.

7. The method for correcting measurement errors of a DC high-power charging pile for electric vehicles according to claim 1, characterized in that: The calculation formula of the standard DC coefficient matrix is ​​as follows: ; In the formula, is the standard DC coefficient matrix; The grid voltage sinusoidal curve is the first The amplitude of the components; For the The angular frequency of the components; For the The phase of the components; is the number of components of the grid voltage sinusoidal set; is the transformation matrix.

8. The method for correcting measurement errors of a DC high-power charging pile for electric vehicles according to claim 7, characterized in that: The steps of obtaining the conversion matrix are as follows: Set different input voltage waveforms; Measure the corresponding output DC voltage; The transformation matrix was obtained using the least squares method.

9. A measurement error correction method for a DC high-power charging pile for electric vehicles according to claim 8, characterized in that: The calculation formula of the fluctuation DC coefficient matrix is ​​as follows: ; In the formula, is the fluctuation DC coefficient matrix; The first sine curve of the fluctuating voltage is concentrated The amplitude of the components; For the The angular frequency of the components; For the The phase of the components; is the number of components of the fluctuating voltage sinusoidal set; is the compensation matrix; and are the coefficients and exponents of the complex terms, is the number of plural terms; Is an imaginary unit.

10. A measurement error correction method for a DC high-power charging pile for electric vehicles according to claim 9, characterized in that: The compensation matrix is ​​obtained by the following steps: Analyze the fluctuation characteristics under different working conditions; Establish a model of the relationship between fluctuation characteristics and working conditions; The compensation matrix is ​​obtained by training a support vector machine.

Images