Electric vehicle charging pile load prediction method and device

Abstract

The present invention relates to a method and device for predicting charging load of electric vehicles taking into account demand response. The method comprises the following steps: dividing the area where electric vehicle charging piles are located into residential areas, work areas and public areas according to the incentive factors on the demand side and the charging behavior of electric vehicles; analyzing the characteristics of the prediction day, extracting the load data, demand response signal data and meteorological and traffic data of the corresponding area, standardizing each index, and selecting similar days by using grey correlation; electric vehicles in residential areas and work areas are mainly affected by price incentives, and are predicted by using a support vector machine; electric vehicles in public areas perform wavelet decomposition on the charging pile load at different levels according to the predictability of a demand response plan, and obtain a seasonal basic load part and a main load part with demand response as the main load, and respectively predict the two decomposed load parts by using a time series model and a grey Markov model, and the two parts of the prediction results are superimposed to obtain the total load of the public area.

Description

Technical Field

[0001] The present invention relates to the field of electric vehicle load prediction, and in particular to a method and a device for predicting the load of an electric vehicle charging pile. Background Art

[0002] As energy depletion, global warming and ecological crisis become more serious, people begin to pay attention to the use of new energy and clean energy. Energy-saving and clean electric vehicles have become an inevitable choice for people's future travel, and the number of electric vehicle charging stations in cities is also increasing. The emergence of generalized demand response resources such as electric vehicles and distributed power sources on the user side has brought more flexible and diverse ways to respond to demand, enhanced load transfer capabilities, and a wider range of transfer time.

[0003] At the same time, the gradual opening of the electricity sales market has provided conditions and environment for various generalized demand response resources to participate in the electricity market. Users with generalized demand response resources will actively respond to market incentive signals based on the output characteristics of different resources to maximize resource utilization. The two-way interaction between users and distribution networks has changed the load characteristics and change laws of active distribution systems, increased the difficulty of short-term load forecasting in operation scheduling and the electricity market, and reduced the forecast accuracy.

[0004] At present, the mainstream methods of electric vehicle load forecasting can be divided into two categories: one is a mathematical model based on statistics, such as regression analysis and time series method; the other is a method based on intelligent algorithms, such as neural networks, support vector machines and extreme learning machines, etc. Statistical algorithms have high requirements for historical data and cannot accurately take into account the impact of external factors such as industrial upgrading, economic growth or climate factors; and some intelligent algorithms have the characteristics of being too dependent on initial values, having local optimal solutions in the training process, and slow convergence speed. Summary of the invention

[0005] The purpose of the present invention is to overcome the deficiencies in the prior art and to provide a method and device for predicting the load of an electric vehicle charging pile, which can improve the accuracy of predicting the load of an electric vehicle charging pile.

[0006] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0007] In a first aspect, the present invention provides a method for predicting the load of an electric vehicle charging pile, comprising the following steps:

[0008] Get the area where the electric vehicle charging piles are located;

[0009] Divide the area where electric vehicle charging piles are located into different zones based on demand-side incentives and electric vehicle charging behavior;

[0010] The loads of electric vehicle charging piles in different zones are predicted separately to obtain the load of each zone; the loads of each zone are superimposed to obtain the total load.

[0011] Further, the method of dividing the area where the electric vehicle charging piles are located into different zones based on the demand-side incentive factors and the charging behavior of the electric vehicle includes: dividing the area where the electric vehicle charging piles are located into residential areas, work areas and public areas based on the demand-side incentive factors and the characteristics of the charging behavior of the electric vehicle;

[0012] The charging behavior of electric vehicles in residential and work areas is characterized by regular electricity use and strong periodicity; the demand-side incentives in residential and work areas include price incentives;

[0013] The charging behavior of electric vehicles in public areas is characterized by disordered charging behavior and poor regularity; the incentives on the demand side of public areas include participation;

[0014] The load forecast of electric vehicle charging piles in different zones includes:

[0015] Predict the load of electric vehicle charging stations in residential and work areas;

[0016] Forecast the load on electric vehicle charging stations in public areas.

[0017] Furthermore, the load of electric vehicle charging piles in residential areas and work areas is predicted, including:

[0018] Step 1: Analyze the characteristics of the forecast day and extract the influencing factors corresponding to the load of electric vehicle charging piles in residential areas and work areas;

[0019] Step 2: Obtain the data of historical days and select similar historical days based on the influencing factors of the forecast day and historical days;

[0020] Step 3: Take the 24-hour historical electricity price data, historical load data, and meteorological factors of similar historical days as input, use the SVR model to analyze and predict the load of electric vehicle charging piles in residential areas and work areas, and output the 24-hour load data of electric vehicle charging piles in residential areas and work areas on the predicted day.

[0021] Furthermore, the historical data includes the 24-hour historical electricity price data, historical load data, meteorological factors of the day, type of the day, and hourly electricity price data of the day; the meteorological factors of the day include temperature of the day, wind speed and cloud cover of the day, and precipitation of the day;

[0022] Methods for selecting similar historical days based on the influencing factors of the forecast day and the historical day include:

[0023] Construct a dataset of influencing factors of the forecast day and a dataset of influencing factors of the historical day;

[0024] According to the dataset of influencing factors of the forecast day and the dataset of influencing factors of the historical day, the correlation between the forecast day and the historical day is calculated;

[0025] According to the correlation between historical days and predicted days, the correlation is sorted, and the first N days are selected as historical similar days, where N is the number of similar days selected.

[0026] Furthermore, the influencing factors include: day type, day hourly electricity price data, day temperature, day wind speed and cloud cover;

[0027] The prediction day influencing factor dataset is

[0028] X0={x0(1),x0(2),...,x0(n)} (1)

[0029] Where x0(1) is the day of the week, coded from 1 to 7; x0(1) is whether it is a weekend or holiday, coded as 1 if yes, otherwise coded as 0; x0(3) is the weather, divided into sunny, cloudy, rainy and snowy days, coded as 1, 2, 3 respectively; x0(4) is the highest temperature of the day; x0(5) is the lowest temperature of the day; x0(6), x0(7), ..., x0(29) are the electricity prices of the 24 time periods in a day, and n is the number of influencing factors;

[0030] The historical daily influencing factor dataset is

[0031] X i ={x i (1), x i (2), ..., x i (n)} (2)

[0032] In the formula, x i (1) is the day of the week, coded from 1 to 7; x i (2) Whether it is a weekend or holiday, if yes, it is coded as 1, otherwise it is coded as 0; x i (3) is the weather, which is divided into sunny, cloudy, rainy and snowy days, and the codes are 1, 2, and 3 respectively; x i (4) is the maximum temperature of the day; x i (5) is the lowest temperature of the day; x i (6), x i (7), ..., x i (29) is the electricity price in each of the 24 time periods in a day, and n is the number of influencing factors.

[0033] Furthermore, according to the dataset of influencing factors of the forecast day and the dataset of influencing factors of the historical day, the correlation between the forecast day and the historical day is calculated, including:

[0034] Substitute each influencing factor into the correlation coefficient calculation formula to obtain the correlation coefficient between the historical date and the predicted date of each influencing factor; the correlation coefficient calculation formula is:

[0035]

[0036] Where: |x0(k)-x i (k)| is the kth point x0 and x i The absolute value difference of the influencing factors, namely the Hamming distance, indicates the influence of the replacement between different influencing factors on the correlation degree. The larger the Hamming distance, the stronger the correlation degree. P is the resolution coefficient, 0<P<1;

[0037] Solve the cumulative average of the correlation coefficients between the historical days and the predicted days of each influencing factor, that is, the correlation between the historical days and the predicted days. The formula is as follows

[0038]

[0039] Where: r i is the influence of the i-th factor on the sample, if r i If it is ≥0.7, it is called a strong correlation factor and has a positive effect.

[0040] Furthermore, based on historical similar days, the SVR model is used to analyze and predict the load of electric vehicle charging piles in residential areas and work areas, including:

[0041] Use the trained SVR model to predict the corresponding part of the load on the forecast day and obtain the load data for 24 hours on the forecast day;

[0042] The training method of the SVR model includes:

[0043] The 24-hour historical electricity price data of similar historical days, the temperature, wind speed, cloud cover, precipitation, and the 24-hour historical load data of the day are used as the input data of SVR, and the 24-hour load data of the predicted day is used as the output data of SVR to train the regression model.

[0044] Furthermore, the load of electric vehicle charging piles in public areas is predicted, including:

[0045] Step 1: Analyze the characteristics of the forecast day and extract the influencing factors corresponding to the load of the electric vehicle charging piles in the public area;

[0046] Step 2: Obtain the data of historical days and select similar historical days based on the influencing factors of the forecast day and historical days;

[0047] Step 3: Based on the data of similar historical days, predict and calculate the load of electric vehicle charging piles in public areas.

[0048] Furthermore, based on the data of similar historical days, the load of electric vehicle charging piles in public areas is calculated, including:

[0049] Step 1: Based on the load P(t) of similar historical days and the predictable demand response signal P DR(ture) (t), obtain the basic load P with strong regularity B (t);

[0050] P B (t) = P(t) - P DR(ture) (t)

[0051] Step 2: Perform wavelet decomposition on the base load with strong regularity to obtain the historical daily seasonal base load and the demand response dominant load on similar historical days;

[0052] Step 3: Establish a training mathematical model based on the seasonal basic load of the historical day; predict the seasonal basic load of the forecast day based on the mathematical model to obtain the seasonal basic load of the forecast day;

[0053] Step 4: grey prediction is performed based on the dominant load of the demand response on the similar historical days and Markov correction is performed to obtain the dominant load of the demand response on the forecast day;

[0054] Step 5: Superimpose the seasonal base load and the demand response-dominated load on the forecast day to obtain the total load in the public area.

[0055] Furthermore, the seasonal load with strong regularity is decomposed by wavelet, including:

[0056] The wavelet sequence ψ that depends on the parameters (a, b) is generated by the wavelet basis function after the translation factor a and the scaling factor b. a,b (t), scaling function

[0057] Using wavelet analysis to decompose and reconstruct, the j-th layer coefficient of wavelet transform By filter coefficient The decomposition formula for obtaining the j-1th layer coefficient is

[0058]

[0059] make Then c j-1 and d j-1 can be seen as c j The low-frequency signal and detail signal, that is, the low-frequency coefficient and the high-frequency coefficient;

[0060] Recursively apply the wavelet decomposition algorithm, and we get the signal c j The low frequency signal c M And the detail signal d at different resolutions M ,d M+1 ,d j-1 , called c M , d M , d M+1 , …, d j-1 c j Wavelet transform of

[0061] The filter coefficients {h n ,n∈z},{g n ,n∈z}, and the j-1th layer coefficients are used to reconstruct the jth layer wavelet coefficients:

[0062]

[0063] Recursively apply the wavelet reconstruction algorithm to obtain the wavelet transform restored signal;

[0064] The basic load with strong regularity on similar historical days is decomposed into waveforms, and the appropriate wavelet basis function is selected to calculate P A (t) Perform 3-layer wavelet transform to extract low-frequency information c 1 and three high frequency information d 1 ,d 2 ,d 3 , the high-frequency information will be reconstructed; among them, c 1 Part is seasonal base load part P b (t), the demand response signal and high-frequency information reconstruction part is the demand response main part load.

[0065] Furthermore, a mathematical model is established according to the historical daily seasonal base load; and the historical daily seasonal base load of the forecast day is forecasted according to the mathematical model to obtain the forecast daily seasonal base load, including:

[0066] Perform time series forecasting on the historical daily seasonal base load, perform stationarity detection on the observation value sequence, and perform white noise test and model identification on the data after it is stable, and determine the model parameters of the mathematical model;

[0067] The mathematical model is used to predict and error analysis is performed to obtain the seasonal base load prediction results for the public area of ​​the residential area on the prediction day.

[0068] Furthermore, the method of performing grey prediction and Markov correction on the load of the dominant part according to the demand response of the historical similar day includes:

[0069] Establish a grey prediction model for the relationship between the main load of the demand response subject on similar days, the demand response plan on similar days, and various meteorological and traffic factors, and obtain the multivariate grey prediction value of the prediction day based on the grey equation of similar days;

[0070] Aiming at the error terms existing in the main load of demand response, the multivariate grey prediction value of the forecast day is Markov corrected, and finally the main load of demand response on the forecast day is obtained.

[0071] In a second aspect, the present invention provides a device for predicting the load of an electric vehicle charging pile, the device comprising:

[0072] Acquisition module: used to obtain the area where the electric vehicle charging pile is located;

[0073] Partitioning module: used to divide the area where electric vehicle charging piles are located into different partitions based on demand-side incentives and electric vehicle charging behavior;

[0074] Prediction module: used to predict the load of electric vehicle charging piles in different partitions respectively, obtain the load of each area; superimpose the load of each area to obtain the total load.

[0075] In a third aspect, the present invention further provides an electric vehicle charging pile load prediction device, including a processor and a storage medium;

[0076] The storage medium is used to store instructions;

[0077] The processor is used to operate according to the instructions to execute the steps of the method described in the first aspect.

[0078] Compared with the prior art, the present invention has the following beneficial effects:

[0079] 1. The present invention takes into account the user demand level and is more in line with the actual application scenario; the load of electric vehicle charging piles in different partitions is predicted respectively, and specific scenarios are analyzed to improve the prediction accuracy. According to the incentive factors on the demand side and the charging behavior of electric vehicles, the area where the electric vehicle charging piles are located is divided into different partitions. Compared with traditional predictions, it takes into account the user demand level and is more in line with the actual application scenario;

[0080] 2. The selection of historical data in the present invention plays the most direct role in prediction. The use of grey correlation to screen out similar days not only avoids the instability of artificial screening, but also improves the scientific nature of screening;

[0081] 3. The present invention uses wavelet changes to decompose and reconstruct seasonal loads, and the error result is significantly lower than that of directly predicting using traditional methods;

[0082] 4. The present invention utilizes grey Markov multi-factor prediction. The grey correlation coefficient takes into account the time characteristics of the independent variable and multiple external variables such as meteorological factors, but does not consider the feedback of the prediction error on the result. Markov is a post-random prediction equation. Therefore, when the state variable and the time variable are discrete data, it will have a better regulating effect on the error generated in the intermediate accumulation process. The Markov fuzzy matrix is ​​used to correct the error generated by the grey prediction to improve the accuracy of the load part prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] Figure 1 A method for zoning electric vehicle charging stations considering demand response incentives

[0084] Figure 2 Load forecasting process for electric vehicle charging stations in residential and work areas

[0085] Figure 3 Load forecasting process for electric vehicle charging stations in public areas

[0086] Figure 4 is the wavelet decomposition result of the load to be processed

[0087] Figure 5 This is the prediction result of electric vehicles using the method proposed in this paper. DETAILED DESCRIPTION

[0088] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0089] Embodiment 1:

[0090] This embodiment provides a method for predicting the composite of electric vehicle charging piles, including the following steps:

[0091] Step 1. A method for zoning EV charging stations considering demand response incentives:

[0092] Considering that the main incentives for demand response and the charging behaviors of electric vehicles in different regions are different, electric vehicle charging piles are divided into residential areas, work areas and public areas. The characteristics of electric vehicle charging piles in residential areas and work areas are regular electricity consumption, strong periodicity, and simple influencing factors; there are a large number of private charging piles in residential areas, and the participation in demand response is not high; the response of residential areas and work areas to demand response is mainly reflected in price incentives. The characteristics of electric vehicle charging piles in public areas are disordered charging behavior, poor regularity, and many influencing factors; the participation of charging piles in public areas in demand response is high. In view of the different main incentives for electric vehicles to respond to demand in different scenarios, load forecasts are conducted for electric vehicle charging piles in residential areas, work areas, and public areas.

[0093] Step 2: Forecast the load of electric vehicle charging piles in residential areas and work areas, including:

[0094] Step 2.1: Analyze the forecast day type and extract historical hourly electricity price data, historical daily temperature, historical daily wind speed, cloud cover and other meteorological factors:

[0095] Assume that the forecast date is

[0096] X0={x0(1),x0(2),...,x0(n)} (1)

[0097] Where x0(1) is the day of the week, coded from 1 to 7; x0(1) is whether it is a weekend or holiday, coded as 1 if yes, otherwise coded as 0; x0(3) is the weather, divided into sunny, cloudy, rainy and snowy days, coded as 1, 2, 3 respectively; x0(4) is the highest temperature of the day; x0(5) is the lowest temperature of the day; x0(6), x0(7), ..., x0(29) are the electricity prices of the 24 time periods in a day, and n is the number of influencing factors;

[0098] The historical daily influencing factor dataset is

[0099] X i ={x i (1), x i (2), ..., x i (n)} (2)

[0100] In the formula, x i (1) is the day of the week, coded from 1 to 7; x i (2) Whether it is a weekend or holiday, if yes, it is coded as 1, otherwise it is coded as 0; x i (3) is the weather, which is divided into sunny, cloudy, rainy and snowy days, and the codes are 1, 2, and 3 respectively; x i (4) is the maximum temperature of the day; x i (5) is the lowest temperature of the day; xi (6), x i (7), ..., x i (29) is the electricity price in each of the 24 time periods in a day, and n is the number of influencing factors.

[0101] Step 2.2: Calculate the correlation between the predicted day and the historical day, standardize each indicator, sort the historical day and the predicted day according to the correlation, and select N similar historical days:

[0102] The correlation coefficient calculation formula is:

[0103]

[0104] Where: |x0(k)-x i (k)| is the kth point x0 and x i The absolute value difference is also called the Hamming distance, which indicates the influence of the replacement between different influencing factors on the correlation. The larger the Hamming distance, the stronger the correlation. P is the resolution coefficient, 0<P<1. For influencing factors with inconsistent units, they are normalized.

[0105] The correlation coefficient only represents the relationship between the forecast day and the historical day at a certain moment. Since the factors affecting the demand-side response will change at different times, the cumulative average value of the correlation coefficient is calculated as the correlation between samples as follows:

[0106]

[0107] Where: r i is the influence of the i-th factor on the sample. Combined with practical applications, if r i If ≥0.7, it is called a strong correlation factor, which has a positive effect. The experiment selects N days with similar historical backgrounds.

[0108] Step 2.3: Use the SVR model to analyze and predict the load of electric vehicle charging piles in residential areas and work areas:

[0109] The historical electricity price data of 7 similar days for 24 hours, daily meteorological factors, and the historical load data of similar days for 24 hours are used as the input and output data of SVR respectively. The regression model is trained and the trained model is used to predict the corresponding part of the load on the prediction day to obtain the load data of the prediction day for 24 hours.

[0110] Step 3: Forecast the load of electric vehicle charging piles in public areas, including:

[0111] Step 3.1: Analyze the day type of the forecast day, extract the corresponding historical load data, traffic factors, meteorological factors and demand response capabilities, calculate the correlation between the forecast day and the historical day, standardize each indicator, sort the historical day and the forecast day according to the correlation, and select N historical similar days X1, X2, X3, ..., X N .

[0112] Step 3.2: According to the demand response signal P DR The predictability of (t) leads to the basic load P with strong periodic regularity. B (t). The electric vehicle charging pile load P(t) in the public area is composed of the base load P0(t) when there is no response and the actual response load of the aggregation agent P DR(ture) (t) consists of two parts, but the information of these two parts cannot be directly obtained. The actual response load P DR(ture) (t) is P DR It is the error term r(t) of the demand response plan of the aggregation agent at time t caused by the information obtained in advance, meteorological factors, and randomness of user behavior. At the same time, the load when there is no response has a strong regularity in the basic load of the same season, and P0(t) can be divided into seasonal basic load P b (t) and the daily load fluctuation term R(t) caused by different daily meteorological factors, user behavior randomness, etc. The decomposition expression of the electric vehicle charging pile load P(t) in the public area based on the independence of seasonal base load and the predictability of demand response is:

[0113] P(t)=P b (t)+P DR (t)+r(t)+R(t) (5)

[0114] According to the demand response signal P DR The predictability of (t) leads to the basic load P with strong periodic regularity. B (t) Expression

[0115] P B (t) = P b (t)+r(t)+R(t) (6)

[0116] Where P(t) is the load of the electric vehicle charging pile, P b (t) Seasonal base load, P B (t) is the basic load with strong periodic regularity, P DR (t) demand response signal, r(t) is the demand response signal P DR (t) and actual response load P DR(ture) (t) is the error term at time t, and R(t) is the load fluctuation term on different days caused by different daily meteorological factors, user behavior randomness, etc.

[0117] Step 3.3: Decompose the waveform of the basic load part of N similar days and select the appropriate wavelet basis function. The basic idea of ​​wavelet transform is to use a set of functions to continuously approximate the signal itself, and the wavelet basis function is transformed by the translation factor a and the scaling factor b to generate a wavelet sequence ψ that depends on the parameters (a, b). a,b (t), scaling function Using wavelet analysis to decompose and reconstruct, the j-th layer coefficient of wavelet transform By filter coefficient Obtain the decomposition of the j-1th layer coefficients, and B (t) Perform j-layer wavelet transform and take the similar part after decomposition as the seasonal basic load part P b (t).

[0118] Step 3.4: Establish a mathematical model for the seasonal base load of the historical days obtained, and predict this part of the load on the forecast day. The seasonal base load is basically not affected by short-term meteorological factors, demand response signals, etc. The loads on different days have the same trend of change, but are also affected by some random processes. It is more appropriate to use a time series model. The exponential smoothing model is based on the description of trend and seasonal data. The purpose of the ARIMA model is to describe the correlation between data. Considering the seasonality of the data set, it has a more ideal effect on the prediction of seasonal base load. Use the time series model to analyze and predict the seasonal base load.

[0119] First, it is necessary to perform a stationary test on the observation sequence. If it is not stationary, then perform a difference operation on it until the data after difference is stationary. After the data is stable, perform a white noise test on it. White noise refers to a random stationary sequence with zero mean and constant variance. If it is a stationary non-white noise sequence, calculate the ACF (autocorrelation coefficient) and PACF (partial autocorrelation coefficient), perform ARMA and other model identification, determine the model parameters for the identified model, and finally apply prediction and perform error analysis.

[0120] Step 3.5: Establish a forecasting model for the relationship between charging pile load, demand response plan, and various meteorological and traffic factors, and forecast the load dominated by demand response resources on the forecast day. Electric vehicle demand response has only developed rapidly in recent years, and data is limited. The gray model is easy to operate and requires less data. It is suitable for making predictions when data is insufficient. Therefore, the gray system forecasting model is used to make a preliminary forecast of the load dominated by demand response resources.

[0121] Define a set of raw data

[0122] X=[X1,X2,X3,...,Y] T (7)

[0123] X1={x (1) (1),x (1) (2),...,x (1) (n)} (8)

[0124] Where X1, X2, ..., X N is the influencing factor of the charging pile on similar days, and Y is the charging pile variable on historical days.

[0125] Generate X1 cumulative data set X (1) The element x in (1) '(k), define x (1) The adjacent mean of the sequence z (1) , define the first-order one-variable differential equation GM(1,1) in whitened form, generate B and constant term vector by averaging the accumulated generated data, solve the grey parameter by least square method, substitute the grey parameter, and reduce the above result to get the free feasible solution of GM(1,1) grey differential equation, which can be expressed as

[0126]

[0127] Repeat the above steps to get similar days X1, X2, ..., X N The gray equation.

[0128] The multivariate grey prediction equation for the forecast day can be expressed as

[0129] y(k)=b0+b1x (1) (k)+b2 x (2) (k)+...+b n x (n) (k) (10)

[0130] In the formula, y(k) is the predicted value of the predicted object at time k, x (1) (k),x (2) (k),x (3) (k) is the predicted value of the influencing factor at time k, ε'(k+1)=D×(ε1,ε2,...,ε n ) T It is the gray action amount.

[0131] The grey prediction model takes into account the time characteristics of the independent variables and the grey correlation coefficients of multiple external variables such as meteorological factors, but does not consider the feedback of the prediction error on the results. Markov is a post-random prediction equation, so when the state variables and time variables are discrete data, it will have a better regulating effect on the errors generated in the intermediate accumulation process. Therefore, the Markov fuzzy matrix is ​​used to correct the errors generated by the grey prediction and improve the accuracy of the prediction of the load dominated by demand response resources.

[0132] The correction matrix D can be defined as D = S * A, S is the membership matrix, A is the transfer matrix, and the relative error ε(k) is defined. The corrected error value ε'(k+1) is

[0133] ε'(k+1)=D×(ε1,ε2,...,ε n ) T (11)

[0134] In the formula, ε'(k+1) is the corrected error value, D is the correction matrix, and ε(k) is the relative error term between the predicted value and the true value.

[0135] The revised forecast value is

[0136]

[0137] In the formula, y'(k+1) is the corrected predicted value, y(k+1) is the predicted value, and ε'(k+1) is the corrected error value.

[0138] The seasonal base load on the forecast day and the partial load dominated by generalized demand response resources are superimposed to obtain the total load in the public area.

[0139] In simple terms, the process of selecting similar days is the same, and only the final N similar days are needed. This paves the way for later data selection. When historical load, weather and other data are needed for later prediction, the load, weather and other data of similar days are selected, and prediction is made using similar days to improve accuracy.

[0140] Based on the data of similar historical days, the load of electric vehicle charging piles in public areas is predicted and calculated:

[0141] Electric vehicle load (historical load is known) - demand response signal (predictable) = seasonal load Seasonal load = seasonal base load + error term (seasonal base load obtained by wavelet decomposition)

[0142] Demand response signal + error term = demand response dominant load (grey prediction + Markov correction error term)

[0143] P(t)=P0(t)+P DR(ture) (t) (13)

[0144] P0(t)=P b (t)+r1(t) (14)

[0145] P DR(ture) (t) = P DR (t)+r2(t) (15)

[0146] P(t)=P b(t)+r1(t)+P DR (t)+r2(t)=P DR (t)+P b (t)+R(t) (16)

[0147] Where P(t) is the load of the electric vehicle charging pile, P0(t) is the basic load when there is no response, and P DR(ture) (t) is the actual response load, P b (t) seasonal base load, r1(t) is the load fluctuation item on different days caused by different daily meteorological factors, user behavior randomness, etc., P DR (t) is the demand response signal, r2(t) is the demand response signal P DR (t) and actual response load P DR(ture) (t) is the error term at time t, and R(t) is the load error term dominated by demand response.

[0148] The load of electric vehicle charging pile is composed of the base load P0(t) when there is no response and the actual demand response P DR(ture) (t), as shown in formula (13). However, in practical applications, there are error terms, such as formula (14) (15).

[0149] Based on the known electric vehicle historical load P(t) in formula (16) and the demand response signal P that can be predicted in advance, DR (t), and obtain the historical similar day seasonal load P b (t)+R(t), where R(t), r1(t), and r2(t) all represent error terms.

[0150] The seasonal base load P is obtained by wavelet decomposition. b (t), according to formula (16), we can get the demand response leading part load P DR (t)+R(t). For historical similar days P b (t) Perform time series forecasting and obtain the predicted day P b (t); for historical similar days P DR (t)+R(t) is used for grey prediction and Markov correction is used to obtain the predicted day P DR (t)+R(t); the two parts are superimposed to obtain the predicted daily total load P(t) of the public area.

[0151] This embodiment provides a specific implementation process of a method for predicting the load of electric vehicle charging piles taking into account demand response. The experiment predicts the load of electric vehicle charging piles in San Francisco in the first half of 2017.

[0152] Figure 1 A method for dividing areas of electric vehicle charging stations taking into account demand response incentive factors is provided for illustration in an embodiment of the present invention.

[0153] Considering that the main incentives for demand response and the charging behaviors of electric vehicles in different regions are different, electric vehicle charging piles are divided into residential areas, work areas and public areas. The characteristics of electric vehicle charging piles in residential areas and work areas are regular electricity consumption, strong periodicity, and simple influencing factors; there are a large number of private charging piles in residential areas, and the participation in demand response is not high; the response of residential areas and work areas to demand response is mainly reflected in price incentives. The characteristics of electric vehicle charging piles in public areas are disordered charging behavior, poor regularity, and many influencing factors; the participation of charging piles in public areas in demand response is high. In view of the different main incentives for electric vehicles to respond to demand in different scenarios, load forecasts are conducted for electric vehicle charging piles in residential areas, work areas, and public areas.

[0154] Figure 2 The load forecasting flow chart of residential and work areas in order to take into account the demand response division area. The following is a forecast of the load of electric vehicle charging piles in residential and work areas in San Francisco in the first half of 2017. The specific implementation steps include:

[0155] Step 1: Select June 16, 2017 as the forecast day, and analyze meteorological factors such as forecast day type, historical hourly electricity price data, historical daily temperature, historical daily wind speed and cloud cover:

[0156] Assume that the forecast date is

[0157] X0={x0(1),x0(2),...,x0(n)} (17)

[0158] History Day

[0159] X i ={x i (1), x i (2), ..., x i (n)} (18)

[0160] Where: k is the serial number, 1,...,i; x k (1) is the day of the week, coded from 1 to 7; x k (2) Whether it is a weekend or holiday, if yes, it is coded as 1, otherwise it is coded as 0; x k (3) is the weather, which is divided into sunny, cloudy, rainy and snowy days, and the codes are 1, 2, and 3 respectively; x k (4) is the maximum temperature of the day; x k (5) is the lowest temperature of the day; x i (6), x i (7), ..., x i (29) is the electricity price for each of the 24 time periods in a day.

[0161] Step 2: Since meteorological factors, electricity price factors, date codes, etc. are indicators of different units, they need to be normalized. Calculate the correlation between the forecast day and the historical day, sort the historical days according to the correlation between the forecast day, and select N similar historical days:

[0162] The correlation coefficient calculation formula is:

[0163]

[0164] Where: |x0(k)-x i (k)| is the kth point x0 and x i The absolute value difference is also called the Hamming distance, which indicates the influence of the replacement between different influencing factors on the correlation. The larger the Hamming distance, the stronger the correlation. P is the resolution coefficient, 0<P<1.

[0165] The correlation coefficient only represents the relationship between the forecast day and the historical day at a certain moment. Since the factors affecting the demand-side response will change at different times, the cumulative average value of the correlation coefficient is calculated as the correlation between samples as follows:

[0166]

[0167] Where: r i To investigate the impact of the ith historical day on the predicted day, the experiment selected 7 similar historical days. After screening, the similar days were April 7, April 14, April 21, April 28, May 12, May 19, and May 26.

[0168] Step 3: Use the SVR model to analyze and predict the load of electric vehicle charging piles in residential areas and work areas. Use the historical electricity price data of 7 similar days for 24 hours, daily meteorological factors, and the historical load data of 24 hours on similar days as the input and output data of SVR, train the regression model, and use the trained model to predict the corresponding part of the load on the forecast day to obtain the load data of 24 hours on the forecast day.

[0169] Figure 3 The flow chart of public area load forecasting for considering demand response division area includes the following specific implementation steps:

[0170] Step 1: Select June 16, 2017 as the forecast day, analyze the day type of the forecast day, extract the corresponding historical load data, traffic factors, meteorological factors and demand response capabilities, calculate the correlation between the forecast day and the historical day, standardize each indicator, sort the historical day and the forecast day according to the correlation, and select 7 historical similar days X1, X2, X3, ..., X7. Among them, the traffic factor is expressed by the Davidson model

[0171] T(q)=T0[1+j*(q / (cq))] (21)

[0172] Where: T(q) is the travel time on the road section when the flow rate is q, T0 is the travel time of the vehicle on the road section when the flow rate is zero, c is the actual traffic capacity of the road section, and j is the curvature parameter of the model.

[0173] The experiment selected 7 similar historical days. After screening, the similar days were April 7, April 14, April 21, April 28, May 12, May 19, and May 26.

[0174] Step 2: The basic idea of ​​wavelet transform is to use a set of functions to continuously approximate the signal itself, and the wavelet basis function is transformed by the translation factor a and the scaling factor b to generate a wavelet sequence ψ that depends on the parameters (a, b). a,b (t), scaling function Using wavelet analysis to decompose and reconstruct, the j-th layer coefficient of wavelet transform By filter coefficient The decomposition formula for obtaining the j-1th layer coefficient is

[0175]

[0176] make Then c j-1 and d j-1 can be seen as c j The low-frequency signal and detail signal of the signal c are the low-frequency coefficients and high-frequency coefficients. Recursively applying the wavelet decomposition algorithm, we can get the signal c j The low frequency signal c M And the detail signal d at different resolutions M ,d M+1 ,d j-1 , called c M , d M , d M+1 , …, d j-1 c j Wavelet transform of .

[0177] The filter coefficients {h n ,n∈z},{g n ,n∈z}, and the j-1th layer coefficients are used to reconstruct the jth layer wavelet coefficients:

[0178]

[0179] Recursively apply the wavelet transform to recover the signal using the wavelet reconstruction algorithm.

[0180] Decompose the waveform of 7 similar days, select the appropriate wavelet basis function, and A (t) Perform 3-layer wavelet transform to extract low-frequency information c 1 and three high frequency information d 1 ,d 2 ,d 3 , the high-frequency information will be reconstructed. Among them, c 1 Part is seasonal base load part P b (t).

[0181] Step 3: Establish a mathematical model for the seasonal base load of historical days, and predict this part of the load on the forecast day. The seasonal base load is basically not affected by short-term meteorological factors, demand response signals, etc. The loads on different days have the same trend of change, but are also affected by some random processes. It is more appropriate to use a time series model. The exponential smoothing model is based on the description of trend and seasonal data. The purpose of the ARIMA model is to describe the correlation between data. Considering the seasonality of the data set, it has a more ideal effect on the prediction of seasonal base load. Use the time series model to analyze and predict the seasonal base load.

[0182] First, it is necessary to perform a stationary test on the observation sequence. If it is not stationary, then perform a difference operation on it until the data after difference is stationary. After the data is stable, perform a white noise test on it. White noise refers to a random stationary sequence with zero mean and constant variance. If it is a stationary non-white noise sequence, calculate the ACF (autocorrelation coefficient) and PACF (partial autocorrelation coefficient), perform ARMA and other model identification, determine the model parameters for the identified model, and finally apply prediction and perform error analysis.

[0183] Step 4: Establish a forecasting model for the relationship between charging pile load, demand response plan, and various meteorological and traffic factors, and forecast the load dominated by demand response resources on the forecast day. Electric vehicle demand response has only developed rapidly in recent years, and data is limited. The gray model is easy to operate, requires less data, and is suitable for making predictions when data is insufficient. Therefore, the gray system forecasting model is used to make a preliminary forecast of the load dominated by demand response resources.

[0184] Define a set of raw data

[0185] X=[X1,X2,X3,...,Y] T (twenty four)

[0186] X1={x (1) (1),x (1) (2),...,x (1) (n)} (25)

[0187] Where X1, X2, ..., X N is the influencing factor of the charging pile on similar days, and Y is the charging pile variable on historical days.

[0188] Generate X1 cumulative data set X (1) The element x in (1) '(k)

[0189]

[0190] Let z (1) For x (1) The nearest mean of the generated sequence

[0191] Z (1) =(z (1) (2),z (1) (3),...,z (1) (n)) (27)

[0192] z (1) (k) = 0.5x (1) '(k)+0.5x (1) '(k-1) (28)

[0193] Define the gray equation GM(N,1)

[0194] x (1) '(k)+az (1) (k) = b (29)

[0195] Among them, a is the development coefficient, b is the grey action;

[0196] Accumulate and generate data mean to generate B and constant term vector

[0197]

[0198] set up The least squares estimated parameter sequence of the grey differential equation satisfies

[0199]

[0200] The whitening equation model is obtained

[0201] x (1) (k)+az (1) (k) = b (32)

[0202] The free feasible solution of the grey differential equation can be expressed as

[0203]

[0204] Repeat the above steps to obtain the grey equations of other influencing factors such as X1, X2, X3;

[0205] The multivariate grey prediction equation for the forecast day can be expressed as

[0206] y(k)=b0+b1x (1) (k)+b2x (2) (k)+...+b n x (n) (k) (34)

[0207] Among them, y(k) is the predicted value of the predicted object at time k, x (1) (k),x (2) (k),x (3) (k) is the predicted value of the influencing factor at time k, b i It is the gray action amount.

[0208] The grey prediction model takes into account the time characteristics of the independent variables and the grey correlation coefficients of multiple external variables such as meteorological factors, but does not consider the feedback of the prediction error on the results. Markov is a post-random prediction equation, so when the state variables and time variables are discrete data, it will have a better regulating effect on the errors generated in the intermediate accumulation process. Therefore, the Markov fuzzy matrix is ​​used to correct the errors generated by the grey prediction and improve the accuracy of the prediction of the load dominated by demand response resources.

[0209] The correction matrix D can be defined as D = S * A, S is the membership matrix, A is the transfer matrix, and the relative error is defined as

[0210] ε'(k+1)=D×(ε1,ε2,...,ε n ) T (35)

[0211] The revised forecast value is

[0212]

[0213] The seasonal base load on the forecast day and the partial load dominated by generalized demand response resources are superimposed to obtain the total load in the public area.

[0214] Embodiment 2:

[0215] This embodiment provides a device for predicting the load of an electric vehicle charging pile, characterized in that the device comprises:

[0216] Acquisition module: used to obtain the area where the electric vehicle charging pile is located;

[0217] Partitioning module: used to divide the area where electric vehicle charging piles are located into different partitions based on demand-side incentives and electric vehicle charging behavior;

[0218] Prediction module: used to predict the load of electric vehicle charging piles in different partitions respectively, obtain the load of each area; superimpose the load of each area to obtain the total load.

[0219] The device of this embodiment can execute the method described in the first embodiment.

[0220] Embodiment three:

[0221] The embodiment of the present invention also provides a device for predicting load of an electric vehicle charging pile, including a processor and a storage medium;

[0222] The storage medium is used to store instructions;

[0223] The processor is used to operate according to the instructions to execute the steps of the method described in embodiment 1.

[0224] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0225] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0226] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0227] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0228] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for predicting the load of an electric vehicle charging pile, characterized in that: The following steps are involved: Get the area where the electric vehicle charging piles are located; Divide the area where electric vehicle charging piles are located into different zones based on demand-side incentives and electric vehicle charging behavior; Predict the load of electric vehicle charging piles in different zones respectively to obtain the load of each zone; Superimpose the loads of each area to obtain the total load; The method of dividing the area where the electric vehicle charging piles are located into different zones according to the demand-side incentive factors and the charging behavior of the electric vehicle includes: dividing the area where the electric vehicle charging piles are located into residential areas, work areas and public areas according to the demand-side incentive factors and the characteristics of the charging behavior of the electric vehicle; The charging behavior of electric vehicles in residential and work areas is characterized by regular electricity use and strong periodicity; the demand-side incentives in residential and work areas include price incentives; The charging behavior of electric vehicles in public areas is characterized by disordered charging behavior and poor regularity; the incentives on the demand side of public areas include participation; The load forecast of electric vehicle charging piles in different zones includes: Predict the load of electric vehicle charging stations in residential and work areas; Predict the load of electric vehicle charging stations in public areas; The load of electric vehicle charging piles in residential and work areas is predicted, including: Analyze the forecast day characteristics and extract the influencing factors corresponding to the load of electric vehicle charging piles in residential areas and work areas; Obtain the data of historical days and select similar historical days based on the influencing factors of the predicted day and historical days; The 24-hour historical electricity price data, historical load data, and meteorological factors of the day are used as input, and the SVR model is used to analyze and predict the load of electric vehicle charging piles in residential areas and work areas, and the load data of the electric vehicle charging piles in residential areas and work areas on the predicted day of 24 hours are output; The historical data includes the 24-hour historical electricity price data, historical load data, meteorological factors of the day, type of the day, and hourly electricity price data of the day; the meteorological factors of the day include temperature of the day, wind speed and cloud cover of the day, and precipitation of the day; Methods for selecting similar historical days based on the influencing factors of the forecast day and the historical day include: Construct a dataset of influencing factors of the forecast day and a dataset of influencing factors of the historical day; According to the dataset of influencing factors of the forecast day and the dataset of influencing factors of the historical day, the correlation between the forecast day and the historical day is calculated; Sort the correlations between historical days and predicted days, and select the previous N days as historical similar days, where N is the number of similar days selected. The influencing factors include: day type, day hourly electricity price data, day temperature, day wind speed and cloud cover; The prediction day influencing factor dataset is X0={x0(1),x0(2),...,x0(n)} (1) Where x0(1) is the day of the week, coded from 1 to 7; x0(1) is whether it is a weekend or holiday, coded as 1 if yes, otherwise coded as 0; x0(3) is the weather, divided into sunny, cloudy, rainy and snowy days, coded as 1, 2, 3 respectively; x0(4) is the highest temperature of the day; x0(5) is the lowest temperature of the day; x0(6), x0(7), ..., x0(29) are the electricity prices of the 24 time periods in a day, and n is the number of influencing factors; The historical daily influencing factor dataset is X i ={x i (1),x i (2),...,x i (n)} (2) In the formula, x i (1) is the day of the week, coded from 1 to 7; x i (2) Whether it is a weekend or holiday, if yes, it is coded as 1, otherwise it is coded as 0; x i (3) is the weather, which is divided into sunny, cloudy, rainy and snowy days, and the codes are 1, 2, and 3 respectively; x i (4) is the maximum temperature of the day; x i (5) is the lowest temperature of the day; x i (6), x i (7), ..., x i (29) is the electricity price in each of the 24 time periods of a day, and n is the number of influencing factors; According to the dataset of influencing factors of the forecast day and the dataset of influencing factors of the historical day, the correlation between the forecast day and the historical day is calculated, including: Substitute each influencing factor into the correlation coefficient calculation formula to obtain the correlation coefficient between the historical date and the predicted date of each influencing factor; the correlation coefficient calculation formula is: Where: |x0(k)-x i (k)| is the kth point x0 and x i The absolute value difference of the influencing factors, namely the Hamming distance, indicates the influence of the replacement between different influencing factors on the correlation degree. The larger the Hamming distance, the stronger the correlation degree. P is the resolution coefficient, 0<P<1; Solve the cumulative average of the correlation coefficients between the historical days and the predicted days of each influencing factor, that is, the correlation between the historical days and the predicted days. The formula is as follows Where: r i is the influence of the i-th factor on the sample, if r i If it is ≥0.7, it is called a strong correlation factor and has a positive effect.

2. The electric vehicle charging pile load prediction method according to claim 1, characterized in that: Based on historical similar days, the SVR model is used to analyze and predict the load of electric vehicle charging piles in residential areas and work areas, including: Use the trained SVR model to predict the corresponding part of the load on the forecast day and obtain the load data for 24 hours on the forecast day; The training method of the SVR model includes: The 24-hour historical electricity price data of similar historical days, the temperature, wind speed, cloud cover, precipitation, and the 24-hour historical load data of the day are used as the input data of SVR, and the 24-hour load data of the predicted day is used as the output data of SVR to train the regression model.

3. The electric vehicle charging pile load prediction method according to claim 1, characterized in that: Based on the data of similar historical days, the load of electric vehicle charging piles in public areas is calculated, including: Based on the load P(t) of similar historical days and the predictable demand response signal P DR(ture) (t), obtain the basic load P with strong regularity B (t); P B (t)=P(t)-P DR(ture) (t) Perform wavelet decomposition on the base load with strong regularity to obtain the historical daily seasonal base load and the dominant load of demand response on similar historical days; Establishing a training mathematical model based on the seasonal basic load of the historical day; predicting the seasonal basic load of the forecast day based on the mathematical model to obtain the seasonal basic load of the forecast day; Perform grey prediction based on the dominant load of the demand response on the similar historical days and perform Markov correction on the error to obtain the dominant load of the demand response on the forecast day; The seasonal base load and the demand response-dominated load on the forecast day are superimposed to obtain the total load in the public area.

4. The electric vehicle charging pile load prediction method according to claim 1, characterized in that: The wavelet decomposition of the basic load with strong regularity includes: The wavelet sequence ψ that depends on the parameters (a, b) is generated by the wavelet basis function after the translation factor a and the scaling factor b. a,b (t), scaling function Using wavelet analysis to decompose and reconstruct, the j-th layer coefficient of wavelet transform By filter coefficient The decomposition formula for obtaining the j-1th layer coefficient is make Then c j-1 and d j-1 Respectively as c j The low-frequency signal and detail signal, that is, the low-frequency coefficient and the high-frequency coefficient; Recursively apply the wavelet decomposition algorithm, and we get the signal c j The low frequency signal c M And the detail signal d at different resolutions M ,d M+1 ,d j-1 , called c M , d M , d M+1 , …, d j-1 c j Wavelet transform of The filter coefficients {h n ,n∈z},{g n ,n∈z}, and the j-1th layer coefficients are used to reconstruct the jth layer wavelet coefficients: Recursively apply the wavelet reconstruction algorithm to obtain the wavelet transform restored signal; The basic load with strong regularity on similar historical days is decomposed into waveforms, and the wavelet basis function is selected to perform three-layer wavelet transform on the basic load with strong regularity to extract low-frequency information c 1 and three high frequency information d 1 ,d 2 ,d 3 , the high-frequency information will be reconstructed; among them, c 1 Part is the seasonal base load part Pb(t), and the part reconstructed by the demand response signal and high-frequency information is the demand response main part load; A mathematical model is established according to the historical daily seasonal base load; and the historical daily seasonal base load of the forecast day is forecasted according to the mathematical model to obtain the forecast daily seasonal base load, including: Perform time series forecasting on the historical daily seasonal base load, perform stationarity detection on the observation value sequence, and perform white noise test and model identification on the data after it is stable, and determine the model parameters of the mathematical model; Apply the mathematical model to predict and perform error analysis to obtain seasonal base load prediction results for the public area of ​​the residential area on the prediction day; The method of performing grey prediction and Markov correction on the dominant part load according to the demand response of the historical similar day includes: Establish a grey prediction model for the relationship between the main load of the demand response subject on similar days, the demand response plan on similar days, and various meteorological and traffic factors, and obtain the multivariate grey prediction value of the prediction day based on the grey equation of similar days; According to the error terms existing in the main load of demand response, the multivariate grey prediction value of the forecast day is Markov corrected to obtain the main load of demand response on the forecast day.

5. An electric vehicle charging pile load prediction device for executing the electric vehicle charging pile load prediction method according to any one of claims 1 to 4, characterized in that: The device comprises: Acquisition module: used to obtain the area where the electric vehicle charging pile is located; Partitioning module: used to divide the area where electric vehicle charging piles are located into different partitions based on demand-side incentives and electric vehicle charging behavior; Prediction module: used to predict the load of electric vehicle charging piles in different partitions respectively, obtain the load of each area; superimpose the load of each area to obtain the total load.

Images