Charging station metering performance detection method based on damping recursive least square

Abstract

The invention relates to the technical field of a Transform-CNN deep learning model and a defined memory recursive damping least square method in charging pile metrological verification, and discloses a damping recursive least square-based charging station metrological performance detection method. According to the charging station metering performance detection method based on damping recursive least square, in the operation process of charging pile equipment, metering misalignment may be caused by factors such as equipment aging or man-made interference, and a traditional method needs to depend on manual regular one-by-one maintenance to evaluate the metering performance; a large amount of manual resource investment is needed for overhauling charging pile equipment one by one, so that a charging pile error estimation model is established, model parameters, namely, charging gun errors, are subjected to parameter identification based on a least square method, whether the errors exceed a verification threshold value or not is judged, the charging piles with metering misaccuracy are detected, remote estimation is achieved, intelligent charging operation and maintenance are promoted, and the charging efficiency is improved. And the development of the electric automobile industry is actively promoted.

Description

Technical Field

[0001] The present invention relates to the technical field of Transformer-CNN deep learning models and limited memory recursive damped least squares method in the metrological verification of charging piles, and specifically to a method for detecting the metrological performance of a charging station based on damped recursive least squares. Background Art

[0002] With the booming development of the electric vehicle industry, it has become crucial to build a complete charging facility and reduce the charging operation and maintenance costs. During the operation of charging pile equipment, metrological inaccuracy may occur due to factors such as equipment aging or human interference. To ensure accurate metering and high performance of charging pile equipment and avoid direct economic losses to users or power grid operating companies caused by excessive metrological deviation of charging piles, currently, manual regular inspection one by one is usually relied on to evaluate the metrological performance. However, inspecting charging pile equipment one by one requires a large amount of manual input, including training professional technicians, dispatching maintenance teams, coordinating work arrangements, etc. At the same time, with the further increase in the number of charging piles and the further expansion of the laying range, the operation and maintenance costs to be borne will also continue to rise. Meanwhile, the inefficient manual verification method will surely hinder the further development of the new energy vehicle industry.

[0003] The present invention proposes an online detection method for the inaccuracy of electric vehicle charging piles. By establishing an electric vehicle charging station-pile model and using the recursive damped least squares method to realize the error estimation of the charging gun meter and the pile meter, remote inaccuracy detection is achieved, promoting intelligent charging operation and maintenance, and will actively promote the development of the electric vehicle industry. Summary of the Invention

[0004] (1) Technical Problem to be Solved

[0005] In view of the deficiencies of the prior art, the present invention provides a method for detecting the metrological performance of a charging station based on damped recursive least squares.

[0006] (2) Technical Solution

[0007] To achieve the above object, the present invention provides the following technical solution: A method for detecting the metrological performance of a charging station based on damped recursive least squares, comprising the following steps:

[0008] (1) Obtain asset coding information data such as charging stations, charging piles, and charging guns, and establish a metrological topology model of an AC / DC hybrid type charging station remote metrological performance detection model.

[0009] (2) Obtain the electricity metering data for remote calculation;

[0010] (3) Data preprocessing, perform data preprocessing operations such as data anomaly detection and screening, and data missing value filling on the electricity metering data in (2).

[0011] (4) Establish a mathematical model for the metering error of AC charging piles;

[0012] (5) Establish a mathematical model for the metering error of DC charging piles;

[0013] (6) Establish a mathematical model for the metering error of charging piles with a mixture of AC and DC types;

[0014] (7) Establish a model for estimating line losses in a charging station based on an artificial neural network;

[0015] (8) Solve the error of the charging gun using the generalized damped recursive least squares method;

[0016] (9) Determine whether the error of each charging gun exceeds the metering inaccuracy threshold, thereby verifying the charging piles with poor metering performance.

[0017] Preferably, in step (1), device information data is obtained, and these information data mainly include: the total meter number in the electric vehicle charging station, the charging pile number in the station, the charging pile type, the charging gun number, and the number topological relationship. The total station meter, charging piles, and charging guns are associated through asset codes to construct a metering topological model.

[0018] Preferably, in step (2), the charging data of the total station meter and the charging gun in the electric vehicle charging station is read. The data specifically includes: the current, voltage, and active power measurement values of the total station meter in the electric vehicle charging station, the current, voltage, and active power measurement values of the electric vehicle charging gun, and the conversion efficiency of the rectifier module of the DC charger.

[0019] Preferably, in step (3), data preprocessing operations such as outlier screening and missing value filling are performed on the data read in step (2). For a certain input data X Input ={x 1 , x 2 ,…, x n}, the specific implementation method is as follows:

[0020] (a) Use the 3σ method to detect and screen outliers in the data.

[0021] Mathematical formula:

[0022] Assume the data set is X = {x 1 , x 2 ,..., x n}, where μ is the mean and σ is the standard deviation.

[0023] 1. Calculate the mean (μ):

[0024]

[0025] 2. Calculate the standard deviation (σ):

[0026]

[0027] 3. 3σ interval: The range of three times the standard deviation of the normal distribution is:

[0028] [μ - 3σ, μ + 3σ]

[0029] 4. Outlier judgment: If a data point x i meets one of the following conditions:

[0030] x i <μ - 3σ or x i >μ + 3σ

[0031] then x i is considered an outlier.

[0032] (b) Use the method of polynomial interpolation to complement and repair the missing values of the input data.

[0033] 2. Repair of single-point missing

[0034] For time-series data X Input ={x 1 , x 2 , …, x n}, assuming that the i-th data point xi is missing, we can estimate this point by polynomial interpolation. Assume that we use an m-degree polynomial p(x) to approximate this set of data, and the form of the interpolation polynomial is:

[0035] P(x) = a 0 + a 1 x + a 2 x 2 +... + a m x m

[0036] The polynomial coefficients a 0 , a 1 ,..., a m can be solved by the values of the known data points, that is, determined by the following system of equations:

[0037] P(x 1 ) = y 1 ,

[0038] P(x 2 ) = y 2 ,

[0039] P(x n ) = y n ,

[0040] where y1 , y 2 , …, y n are known data points.

[0041] The repaired missing point x i can be calculated by substituting x i into the polynomial p(x):

[0042] x i = P(x i )

[0043] 3. Repair of consecutive point losses

[0044] For consecutive missing data points, assuming that the i-th to j-th (where i < j) data points are missing, we can still repair these missing points through polynomial interpolation. We can use the p(x) interpolation polynomial to calculate the estimated values of the missing points:

[0045] x k = P(x k ), where i ≤ k ≤ j

[0046] This method requires sufficient known data points at both ends of the missing points to ensure the accuracy of interpolation.

[0047] Preferably, the specific process of the AC charging pile error calculation model established in step (4) is as follows:

[0048] (a) For a charging station with m charging piles and n charging guns in the station, define the error φ i of the total power meter in the station and the measurement points of the charging guns as the ratio of the actual value of the electricity consumption to the indicated value of the meter, and the true error ξ i as the ratio of the difference between the true value and the indicated value to the indicated value:

[0049]

[0050] In the formula:

[0051] i —— charging gun number;

[0052] x i (t) —— the measured indicated value of the charging gun side at time t;

[0053] x i_true (t) —— the actual value.

[0054] When |ξ i | exceeds the normal measurement error threshold ξ th , that is, |ξ i | > ξ th , it is considered that the metering of this device is inaccurate. ξ th is set according to the accuracy class requirements of the electricity meter in the charging pile.

[0055] b) Establish an estimation model for the metering error of the AC charging pile, and its formula is as follows:

[0056]

[0057] Where:

[0058] y(t) - The indication value of the total watt-hour meter;

[0059] L(t) - The dynamic loss caused by the line impedance;

[0060] L con - Fixed loss;

[0061] v o (t) is the measurement noise.

[0062] During the measurement period, use matrices to represent the relationship between all input data of each variable in the calculation unit:

[0063] Y(t) = X(t)Φ x +L(t)+L con +V o (t) (4)

[0064] Where:

[0065] Y(t) - The column vector of y(t), that is, Y(t) = [y(1), y(2), …, y(t)] T ;

[0066] X(t) - A matrix of size t×n, where the element x i,j represents the positive active power of the charging gun numbered j at the i-th moment;

[0067] L(t) - The column vector of L(t); L con is the column vector of the constant L con ;

[0068] V o (t) - The column vector of v o (t).

[0069] Preferably, for the DC charging pile error calculation model established in step (5), the model distinguishes between split-type DC chargers and integrated DC chargers, and the specific process is as follows:

[0070] The split - type machine conducts AC - DC conversion by a charger with a rectification module and delivers direct current to the DC charger; the integrated machine has an AC - DC conversion module inside itself. Its incoming line is not connected to the charger but directly connected to the AC branch in the charging station. Therefore, split - type machines connected to the same charger have the same conversion efficiency, while integrated machines have their own independent conversion efficiencies.

[0071] Since the electric energy metering point is on the output side of the direct current, energy loss during the conversion of alternating current to direct current, that is, the AC - DC conversion efficiency, is considered when establishing the electric energy conservation equation for DC - type charging piles:

[0072]

[0073] Its formula is as follows:

[0074]

[0075] In the formula:

[0076] n I —— The number of charging piles of the integrated - machine type;

[0077] n S —— The number of split - type charging piles;

[0078] n j —— The number of charging machines connected to the charging machine j.

[0079] η i —— The conversion efficiency of the AC - DC conversion module of the integrated charger.

[0080] η j —— The conversion efficiency of the AC - DC conversion module of the split - type charger.

[0081] Preferably, the error calculation model for the mixture of AC / DC - type charging piles in the station established in step (6) is as follows:

[0082] The error calculation models for AC and DC charging piles in the AC - DC hybrid charging station are used to obtain the error calculation model for the hybrid charging station:

[0083]

[0084] In the formula:

[0085] n AC —— The number of AC charging piles in the station;

[0086] n DC-I —— The number of integrated DC chargers;

[0087] n DC-S——Number of split - type DC chargers.

[0088] η i (t) — Conversion efficiency of the AC - DC conversion module of the integrated charger.

[0089] η j (t) — Conversion efficiency of the AC - DC conversion module of the split - type charger.

[0090] Preferably, in step (7), an on - line loss estimation model of the charging station based on an artificial neural network is established. Since the dynamic on - line loss value cannot be directly obtained, in step (7), a filtering model based on an artificial neural network is designed by using the total electric energy meter measurement data Y(t) and the charging gun measurement data Φ(t) to approximately convert the on - line loss. Let XF(t) be the filter input matrix, ΦF be the filter coefficient matrix, and F(t) be the filter output matrix. The relationship between these three variables forms the filter, which is expressed as:

[0091] F(t) = X F (t)Φ F

[0092] The specific method is as follows:

[0093] (a) Calculate the observed on - line loss value

[0094]

[0095] In the formula:

[0096] y(t) — Reading of the total electric energy meter;

[0097] x i (t) — Measurement reading on the charging gun side at time t;

[0098] (b) Cluster the observed on - line loss data.

[0099] The clustering result is represented in the form of a training membership matrix X train (t) and a training cluster center column vector C train . Take X train (t) as the training target and input it into the training block, and average C train to initialize XF.

[0100] (c) Neural network training

[0101] In the training block, Atrain(t) is the target set and Y(t) is the input set. The target set and the input set form a sample set. Then the sample set is randomly divided into training samples, validation samples, and test samples, and the recommended ratio of the three is 14 / 3 / 3. Further use the segmented sample set to train the neural network.

[0102] Finally, the input of the trained neural network is Y(t), and the output membership matrix is X NN (t), and this matrix is assigned to X F (t).

[0103] Preferably, in step (8), the generalized damped recursive least squares method is used to solve the models established in steps (4), (5), and (6). The specific process is as follows:

[0104] (a) Augmented data matrix construction

[0105] Combine the data input matrix X(t) and the neural network output membership matrix X F (t) to construct the augmented system total input matrix X G (t):

[0106] X G (t) = [X(t)|X F (t)] (14)

[0107] Construct the generalized system parameter matrix Φ G_true as:

[0108]

[0109] (b) Based on the AC and DC charging pile error models, establish an identification system model:

[0110] Then (3) can be written as:

[0111] Y(t) = X G (t)Φ Gtrue +V G (t)+V o (t) (16)

[0112] Equation (16) is the basic equation for generalized smart meter error estimation. According to the derivation process of damping, the recursive least squares

[15] , the estimation term is replaced by It is easy to obtain the objective function GDRLSJ G (t) as:

[0113]

[0114] μ(t) = μ 0 ×mean(a G (t) T a G (t)) (19)

[0115] When the recursive algorithm is synchronized with the measurement time, Φ G_etm (t) = [Φ etm(t) T , Φ F_etm (t) T T is Φ G_true the estimate at the t-th recursion; Φ etm (t) and Φ F_etm (t) are the estimates of Φ true and Φ F , W(t) is the weight matrix; ρ is the weight factor; when ρ < 1, the weights of historical data decay exponentially, and the algorithm has the property of fading memory. μ(t) is the damping penalty coefficient; μ 0 is the damping coefficient. The purpose of μ(t) is to penalize the objective function when the difference between the t-th and (t - 1)-th estimates becomes large, thereby restricting the range of change of the estimate. a G (t) = [a G (t), a NN (t)] is the t-th row vector of X G (t); a NN (t) is the t-th row vector of X NN (t).

[0116] To minimize J G (t), we take the first-order partial derivative of J G (t) and set it equal to 0. Then we can get:

[0117]

[0118] where I is the identity matrix, P G (t) is the covariance matrix:

[0119] P G (t) = [μ(t)I - ρμ(t)I + ρP G (t - 1) -1 + a G (t) T a G (t)] -1 (21)

[0120] (c) To obtain the error coefficient of a certain charging gun sub-meter at time k, according to Equation (21), the final algorithm obtains the error matrix of the electricity meter at time k for each element in, that is, the error coefficient matrix ε k of the charging gun sub-meter is obtained. The error coefficient ε i，k of the i-th charging gun sub-meter at time k corresponds to the i-th element among them. The algorithm ends when the predetermined number of iterations is reached

[0121] ​In step (7), it is judged according to the measurement errors of each charging gun calculated in step (6): if it exceeds the out-of-accuracy threshold, it is detected that the charging pile where the charging gun is located is a measurement-out-of-accuracy charging pile; otherwise, it is a measurement-normal charging pile.

[0122] (III) Beneficial effects

[0123] Compared with the prior art, the present invention provides a method for detecting the metering performance of a charging station based on damped recursive least squares, having the following beneficial effects:

[0124] 1. For the method for detecting the metering performance of a charging station based on damped recursive least squares, during the operation of the charging pile equipment, the metering may be inaccurate due to factors such as equipment aging or human interference. Traditional methods rely on manual periodic inspections one by one to evaluate the metering performance, and a large amount of human resources are required to inspect the charging pile equipment one by one. Therefore, the present invention establishes a charging pile error estimation model, identifies the model parameters, that is, the charging gun error, based on the least squares method, and judges whether the error exceeds the verification threshold, so as to detect the charging pile with inaccurate metering, realize remote estimation, promote intelligent charging operation and maintenance, and will actively promote the development of the electric vehicle industry. Description of the drawings

[0125] Figure 1 is a flowchart of the method for detecting the metering performance of a charging station based on damped recursive least squares of the present invention;

[0126] Figure 2 is a diagram of the calculation result of the charging gun error of the present invention. Detailed implementation manners

[0127] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0128] Please refer to Figure 1-2 , the present invention provides a technical solution: a method for detecting the metering performance of a charging station based on damped recursive least squares, by establishing a metering error model for AC and DC charging piles, as well as an AC-DC conversion efficiency fitting model for DC charging piles, and using the limited memory recursive damped least squares method to solve the model, realizing the remote verification of the out-of-accuracy charging piles. The overall calculation process is as Figure 1 shown.

[0129] Step 1: Obtain the total meter data of the charging station, the measurement data of the charging guns, and the operating data of the charging piles, specifically including: current, voltage, power factor, rated AC-DC conversion efficiency, operating ambient temperature, etc.;

[0130] Step 2: Preprocess and standardize the operating data of the charging piles in 1);

[0131] Step 3: Establish a mathematical model for the metering error of AC charging piles;

[0132] Step 4: Establish a mathematical model for the metering error of DC charging piles;

[0133] Step 5: Perform deep learning training based on the Transformer-CNN model to achieve the fitting of AC-DC conversion efficiency;

[0134] Step 6: Solve the error of the model charging gun by using the recursive damped least squares method;

[0135] Step 7: Determine whether the error of each charging gun exceeds the metering inaccuracy threshold, so as to calibrate the inaccurate charging piles.

[0136] Among them, in step 1), each parameter can be expressed as:

[0137] Voltage value U(t), current value I(t), power factor Ambient temperature T(t) and constant rated conversion efficiency η N .

[0138] Among them, in step 2), the extracted feature data is standardized to eliminate the influence of units and value ranges. There are n samples in total, and the calculation steps are as follows:

[0139] Organize the charging pile data in step 1) into a standardized form. The specific steps include:

[0140] Calculate the average value of the feature data and variance s ij :

[0141]

[0142] Among them, in step 3), the process of establishing the error calculation model of the AC charging pile is as follows:

[0143] a) Obtain the measured value of the charging pile: y true is the true power flowing through the total meter of the charging station, b ti represents the true power flowing through charging gun i, N represents the number of charging guns in the station, e 1 represents the line loss, e 0 represents the fixed loss, and N represents the number of charging guns in the station.

[0144] b) Based on the mathematical relationship of electric energy: "Power supply of the charging station" = "Power consumption of each charging pile" + "Power loss", the following power conservation equation can be established using the metering point data of each charging pile in the charging station:

[0145]

[0146] Among them, the metering error of the device is defined as the percentage value of the difference between the measured value and the true value of the power consumption divided by the measured value. Then the total meter metering error a y and the metering error ai of the sub-meter i can be expressed as:

[0147]

[0148]

[0149] c) Define b i to represent the sub-meter measurement value. Substituting the metering error into the power conservation equation (2) gives the charging station system equation for metering error identification:

[0150]

[0151] Among them, the specific process of the AC charging pile error calculation model established in step 4) is as follows:

[0152] a) Obtain the measured value of the charging pile: a y represents the total meter metering error, y represents the total meter measurement value, a i represents the sub-meter metering error, b i represents the sub-meter measurement value, N represents the number of charging guns in the station, η represents the conversion efficiency of the AC-DC conversion module, e 1 represents the line loss, e 0 represents the fixed loss.

[0153] b) Based on the mathematical relationship of electric energy: "Power supply of the charging station" = "Power consumption of each charging pile" / "Conversion efficiency of the AC-DC module" + "Fixed loss" + "Line loss", the following power conservation equation can be established using the metering point data of each charging pile in the charging station:

[0154]

[0155] Among them, step 5) is a deep learning model for fitting the AC-DC conversion efficiency of DC charging piles. The specific training and deployment steps are as follows:

[0156] a) Use the preprocessed charging pile operation data to construct a training data set and a test data set. The input matrix is:

[0157]

[0158] b) Construct a Transformer-CNN deep learning structure, and the calculation process of each layer is as follows:

[0159] F 1 = f Transformer (X Input (t)) (8)

[0160] F i+1 = f Transformer (F i ), i = 1, 2 (9)

[0161] F MFB1 = f MFB (F 3 ) (10)

[0162] F MFB2 = f MFB (F MFB1 ) (11)

[0163] F 4 ,F 5 = Conv(Up(F 1 ,F 2 )) (12)

[0164] F 6 = Conv(Up(Concat(F 3 ,F MFB1 ,F MFB2 ))) (13)

[0165] X output (t) = Up(Linear(Concat(F 4 ,F 5 ,F 6 ))) (14)

[0166] In the formula: f Transformer and f MFB respectively represent the feature maps extracted by learning through the Transformer and MFB modules, Concat represents feature concatenation, and Up represents the upsampling operation.

[0167] c) Input the real-time operation data into the trained model to obtain the fitting value of the real-time conversion efficiency.

[0168] Among them, in step 6), the recursive damped least squares method is used to solve the error of the model charging gun, and the specific solution process is as follows:

[0169] a) Define the total meter reading matrix A of the charging station and the line loss matrix L as the inputs of the identification system, the total meter reading matrix Y of the model as the input, and the error coefficient X matrix of the charging gun meters as the output of the identification parameters of the system. Establish an identification system model based on the AC and DC error models:

[0170] Y = AX + L + V 0 (15)

[0171] where V 0 is the measurement noise.

[0172] b) For the calculation iteration at time k, let the observed value z k be the difference between the total meter reading y k at time k of the charging station and the estimated line loss L k at time k. Let a k be the row vector of the sub - meter measurement reading matrix A k , representing the reading vectors of all sub - meters, and x k be the errors of all sub - meters. Then we have:

[0173] z k = a k x k + v o (16)

[0174] Establish the cost function J in the recursive damped least - squares algorithm w which is expressed as:

[0175]

[0176] where: is the model estimation parameter, W is the weight coefficient, and U k is the damping term.

[0177] Introduce the forgetting factor ρ to optimize the least - squares to weighted least - squares. The weight coefficient W k is:

[0178]

[0179] Add the damping term U k to the cost function in Equation (17), where μ k is the damping coefficient at the current time, and it is expanded as:

[0180]

[0181] μ k = μ × mean(a k T a k ) (20)

[0182] Calculate the damping gain through Equation (19), increase the limit on the change of the input value, and when the current input changes greatly compared to the previous input, for the cost function J w The added damping term U k limits the change range of the estimate.

[0183] Simplify to obtain the recurrence formula for the parameters:

[0184]

[0185] In the formula: P k is the covariance matrix of the recursive least squares method, and the covariance matrix is updated as follows:

[0186] P k =(μ k I - ρυμ k I + ρυP k-1 -1 +a k T a k ) -1 (22)

[0187] In the formula: I is the identity matrix with the same dimension as P k dimension.

[0188] c) In order to obtain the error coefficient of a certain charging gun sub-meter at time k, according to Equation (21), the final algorithm obtains the error matrix of the electricity meter at time k for each element in, that is, the error coefficient matrix ε of the charging gun sub-meter is obtained k , and the error coefficient ε of the i-th charging gun sub-meter at time k i，k corresponds to the i-th element among them. When the predetermined number of iterations is reached, the algorithm ends.

[0189] Among them, in step 7), it is judged according to the measurement errors of each charging gun obtained in step 6): if it exceeds the out-of-tolerance threshold, it is detected that the charging pile where the charging gun is located is a metering out-of-tolerance charging pile; otherwise, it is a metering normal charging pile.

[0190] In order to verify the effectiveness of the method proposed in the present invention, charging pile data in a real scenario is used for verification. The data includes a total of 50 charging gun data under 8 charging stations, the data acquisition frequency is once every 15 minutes, and the data length is 400 days. The data includes the charging station - charging pile - charging gun topology structure, the measured values of the main meter and the charging gun electricity consumption, the voltage value, the current value, the power factor, the environmental temperature and other operating parameters during the charging process, as well as the rated AC / DC conversion efficiency of each charging pile at the factory. After on-site verification, the charging guns numbered 10#, 20#, 30#, 40#, and 50# are metering out of tolerance. The verification method of the present invention can accurately detect out-of-tolerance charging piles, and its accuracy rate and F1 score both reach 100%.

[0191] In summary, the present invention provides a remote detection method for the metering inaccuracy of charging piles based on the limited memory recursive damping least squares method, providing a new approach for the intelligent operation and maintenance of charging piles.

[0192] The dimensions and shapes of all components in this structure are not specifically limited here and need to be produced according to the actual situation.

[0193] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A charging station metering performance detection method based on damped recursive least squares, comprising the following steps: (1) Obtain asset coding information data such as charging stations, charging piles, and charging guns, and establish a metering topology model for the remote metering performance detection model of AC / DC hybrid charging stations. (2) Obtaining electricity metering data for remote computing; (3) Data preprocessing: performing data anomaly detection and screening, data missing value filling and other data preprocessing operations on the electricity metering data in (2). (4) Establish a mathematical model for AC charging pile measurement errors; (5) Establish a mathematical model for DC charging pile measurement errors; (6) Establish a mathematical model for the metering error of charging piles with mixed AC and DC types; (7) Establish a line loss estimation model in charging stations based on artificial neural networks; (8) The generalized damped recursive least squares method is used to solve the charging gun error; (9) Determine whether the error of each charging gun exceeds the metering inaccuracy threshold, thereby calibrating charging piles with poor metering performance.

2. According to claim 1, a charging station metering performance detection method based on damped recursive least squares is characterized by: Wherein, step (1) obtains equipment information data, which mainly includes: the station master meter number of the electric vehicle charging station, the station charging pile number, the charging pile type, the charging gun number and the number topological relationship. The charging station master meter, charging pile and charging gun are associated through asset coding to construct a metering topology model.

3. The method for detecting the metering performance of a charging station based on damped recursive least squares according to claim 1 is characterized in that: Wherein step (2) reads the charging data of the total meter in the electric vehicle charging station and the charging gun, and the data specifically includes: the current, voltage, and active power metering values ​​of the total meter in the electric vehicle charging station, the current, voltage, and active power metering values ​​of the electric vehicle charging gun, and the conversion efficiency of the DC charger rectifier module.

4. The method for detecting the metering performance of a charging station based on damped recursive least squares according to claim 1 is characterized in that: In step (3), the data read in step (2) is subjected to data preprocessing operations such as outlier screening and missing value filling. For a certain input data X Input ={x1,x2,…,x n }, the specific implementation method is as follows: (a) The 3σ method is used to screen outliers in the data. Mathematical formula: Assume that the data set is X = {x1, x2, ..., x n }, where μ is the mean and σ is the standard deviation.

1. Calculate the mean (μ):

2. Calculate the standard deviation (σ): 3.3σ interval: The range of 3 times the standard deviation of the normal distribution is: [μ-3σ,μ+3σ] 4. Outlier judgment: If a data point x i One of the following conditions is met: x i <μ-3σ or x i >μ+3σ Then x i is considered an outlier. (b) Use polynomial interpolation method to complete the missing values ​​of input data.

2. Repair of single point loss For time series data X Input ={x1,x2,...,x n }, assuming that the i-th data point x i If it is missing, we can estimate this point through polynomial interpolation. Assume that we use an m-degree polynomial p(x) to approximate this set of data. The form of the topological polynomial is: P(x)=a0+a1x+a2x 2 +…+a m x m Polynomial coefficients a0, a1, ..., a m It can be solved by the values ​​of known data points, that is, determined by the following system of equations: P(x1)=y1, P(x2)=y2, P(x n )=y n , where y1,y2,...,y n are known data points. The missing point x after repair i By changing x i Substitute the polynomial p(x) to calculate: x i =P(x i ) 3. Repair of missing continuous points For consecutive missing data points, assuming that the i-th to j-th (i<j in the set) data points are missing, we can still repair these missing points through polynomial interpolation. We can use the p(x) broadcast polynomial to calculate the estimated value of the missing points: x k =P(x k ), where i≤k≤j This method requires that there are enough known data points on both ends of the missing point to ensure the accuracy of the interpolation.

5. The method for detecting charging station metering performance based on damped recursive least squares according to claim 1 is characterized in that: The AC charging pile error calculation model established in step (4) is specifically as follows: (a) For a charging station with m charging piles and n charging guns, define the error φ between the total energy meter and the charging gun side measurement point in the station. i , is the ratio of the actual power consumption to the meter indication, the true error ξ i The ratio of the difference between the true value and the indicated value to the indicated value: Where: i——charging gun number; x i (t)——the metering indication on the charging gun side at time t; x i_true (t)——actual value. When |ξ i |Exceeding the normal measurement error threshold ξ th When |ξ i |>ξ th , it is believed that the device is inaccurate in measurement. th It is set according to the accuracy level requirement of the electric energy meter in the charging pile. b) Establish an AC pile measurement error estimation model, the formula is as follows: Where: y(t)——indication of total electric energy meter; L(t)——dynamic loss caused by line impedance; L con - fixed losses; v o (t) is the measurement noise. During the measurement period, a matrix is ​​used to represent the relationship between all input data of each variable in the calculation unit: Y(t)=X(t)Φ x +L(t)+L con +V o (t) (4) Where: Y(t)——y(t) is the derivative, and Y(t) is [y(1),y(2),…,y(t)] T ; X(t)——a matrix of size t×n, where the elements x i,j Indicates the forward active power of charging gun with serial number j at time i; L(t)——column vector of L(t); L con is a constant L con Column vector of ; V o (t)——v o (t) is a column vector.

6. The method for detecting charging station metering performance based on damped recursive least squares according to claim 1 is characterized in that: The DC charging pile error calculation model established in step (5) distinguishes between split-type DC chargers and integrated DC chargers. The specific process is as follows: The split-type machine uses a charger with a rectifier module to convert AC to DC and transmits DC power to the DC charger; the integrated machine has an AC to DC conversion module inside itself, and its incoming line is not connected to the charger, but directly connected to the AC branch in the charging station. Therefore, the split-type machines connected to the same charger have the same conversion efficiency, while the integrated machines have their own independent conversion efficiency. Since the energy metering point is located on the output side of DC power, the energy loss during the conversion of AC power to DC power is considered when establishing the energy conservation equation of the DC type charging pile, that is, the AC-DC conversion efficiency: The formula is as follows: Where: n I ——The number of charging piles of the integrated type; n S ——Number of split-type charging piles; n j ——The number of chargers connected to charger j. η i ——Conversion efficiency of the AC-DC conversion module of the integrated charger. η j ——Conversion efficiency of the AC-DC conversion module of the split charger.

7. A charging station metering performance detection method based on damped recursive least squares according to claim 1, characterized in that: The error calculation model for the mixed AC / DC charging piles in the station established in step (6) is specifically as follows: The error calculation model of the AC and DC hybrid charging stations is obtained by the error calculation model of the AC and DC hybrid charging stations: Where: n AC ——Number of AC charging piles in the station; n DC-I ——Number of integrated DC chargers; n DC-S ——Number of split-type DC chargers. η i (t)——Conversion efficiency of the AC-DC conversion module of the integrated charger. η j (t)——Conversion efficiency of the AC-DC conversion module of the split charger.

8. The method for detecting charging station metering performance based on damped recursive least squares according to claim 1 is characterized in that: Step (7) establishes a line loss estimation model for charging stations based on artificial neural networks. Since it is impossible to directly obtain dynamic line loss values, step (7) uses the total energy meter measurement data Y(t) and the charging gun measurement data Φ(t) to design a filter model based on artificial neural networks to approximate the line loss. Let XF(t) be the filter input matrix, ΦF be the filter coefficient matrix, and F(t) be the filter output matrix. These three variables constitute the filter, and the relationship between them is expressed as: F(t)=X F (t)Φ F The specific method is as follows: (a) Calculate the observed line loss value Where: y(t)——indication of total electric energy meter; x i (t)——the metering indication on the charging gun side at time t; (b) Clustering of observed line loss data. Clustering results are used to train the membership matrix X train (t) and the training cluster center column vector C train In the form of train (t) is input into the training block as the training target, and C train Averaging is performed to initialize XF. (c) Neural network training In the training block, Atrain(t) is the target set and Y(t) is the input set. The target set and the input set constitute the sample set. Then the sample set is randomly divided into training samples, validation samples and test samples. The ratio of the three is recommended to be 14 / 3 / 3. The divided sample set is further used to train the neural network. Finally, the trained neural network input is Y(t), and the output membership matrix X NN (t), the matrix is ​​assigned to X F (t).

9. The method for detecting charging station metering performance based on damped recursive least squares according to claim 1 is characterized in that: In step (8), the generalized damped recursive least squares method is used to solve the model established in step (4), step (5), and step (6). The specific process is as follows: (a) Augmented data matrix construction Combine the data input matrix X(t) and the neural network output membership matrix X F (t) Construct the total input matrix X of the augmented system G (t): X G (t)=[X(t)∣X F (t)] (14) Construct the generalized system parameter matrix Φ G_true for: (b) Establish an identification system model based on the AC and DC charging pile error models: Then (3) can be written as: Y(t)=X G (t)Φ Gtrue +V G (t)+V o (t) (16) Equation (16) is the basic equation for generalized smart meter error estimation. According to the recursive least squares derivation process of damping [15], the estimation term is replaced by X Getm (t), it is easy to obtain the objective function GDRLSJ G (t) is: μ(t)=μ0×mean(a G (t) T a G (t)) (19) When the recursive algorithm and the measurement time are synchronized, Φ G_etm (t)=[Φ etm (t) T ,Φ F_etm (t) T ] T is Φ G_true The estimate at the tth recursion; Φ etm (t) and Φ F_etm (t) is Φ true and Φ F Estimate of , W(t) is the weight matrix; ρ is the weight factor; When ρ < 1, the weight of historical data decays exponentially, and the algorithm has memory decay characteristics. μ(t) is the damping penalty coefficient; μ0 is the damping coefficient. The purpose of μ(t) is to penalize the objective function when the difference between the t-th and t-1-th estimates becomes larger, thereby limiting the range of variation of the estimate. G (t) = [a G (t),a NN (t)] is X G The t-th row vector of (t); a NN (t) is X NN The t-th row vector of (t). In order to minimize J G (t), we take J G (t) and set it equal to 0. Then we can get: Where I is the unit matrix, P G (t) is the covariance matrix: P G (t)=[μ(t)I-ρμ(t)I+ρP G (t-1) -1 +a G (t) T a G (t)] -1 (21) (c) In order to obtain the error coefficient of a charging gun sub-meter at time k, according to formula (21), the final algorithm obtains the error matrix of the meter at time k: Each element in the equation is the error coefficient matrix ε of the charging gun submeter. k , the error coefficient ε of the i-th charging gun submeter at time k i，k Corresponding to the i-th element. The algorithm ends when the predetermined number of iterations is reached. In step (7), a judgment is made based on the metering error of each charging gun calculated in step (6): if it exceeds the inaccuracy threshold, the charging pile where the charging gun is located is detected as a charging pile with inaccurate metering; otherwise, it is a charging pile with normal metering.

Images