CN116796892A - Short-term electric vehicle charging load probability prediction method based on composite quantile regression - Google Patents

Short-term electric vehicle charging load probability prediction method based on composite quantile regression Download PDF

Info

Publication number
CN116796892A
CN116796892A CN202310674536.0A CN202310674536A CN116796892A CN 116796892 A CN116796892 A CN 116796892A CN 202310674536 A CN202310674536 A CN 202310674536A CN 116796892 A CN116796892 A CN 116796892A
Authority
CN
China
Prior art keywords
charging load
prediction
electric vehicle
vehicle charging
predicted
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202310674536.0A
Other languages
Chinese (zh)
Inventor
庞彬
李建平
于鹤洋
霍英宁
耿光超
江全元
陈奕
徐川子
向新宇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang University ZJU
Hangzhou Power Supply Co of State Grid Zhejiang Electric Power Co Ltd
Original Assignee
Zhejiang University ZJU
Hangzhou Power Supply Co of State Grid Zhejiang Electric Power Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Zhejiang University ZJU, Hangzhou Power Supply Co of State Grid Zhejiang Electric Power Co Ltd filed Critical Zhejiang University ZJU
Priority to CN202310674536.0A priority Critical patent/CN116796892A/en
Publication of CN116796892A publication Critical patent/CN116796892A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Business, Economics & Management (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Economics (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Marketing (AREA)
  • Pure & Applied Mathematics (AREA)
  • Tourism & Hospitality (AREA)
  • Computational Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Operations Research (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Algebra (AREA)
  • Probability & Statistics with Applications (AREA)
  • Databases & Information Systems (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Primary Health Care (AREA)
  • Water Supply & Treatment (AREA)
  • Public Health (AREA)
  • Quality & Reliability (AREA)

Abstract

The invention discloses a short-term electric vehicle charging load probability prediction method based on composite quantile regression. Firstly, 4 types of data including historical charging load, air temperature, weather type and day type of the electric automobile are obtained, and data are subjected to quantization and normalization pretreatment. Then, based on the set model structure and model input, and the specific loss function with penalty term, an LSTM neural network Composite Quantile Regression (CQRLSTM) prediction model is constructed. And training a prediction model by using historical data, inputting the data to be predicted into the model to obtain a series of quantile prediction results, and obtaining the upper limit and the lower limit of a prediction confidence interval from the model output results to obtain a prediction curve of the charging load of the electric automobile. And finally, generating a charging load probability density curve based on the output prediction result and a nuclear density estimation method. The obtained prediction curve and the prediction interval can be used as input quantity of an ordered charging strategy, and are beneficial to design and operation of the ordered charging strategy.

Description

Short-term electric vehicle charging load probability prediction method based on composite quantile regression
Technical Field
The invention relates to the field of electric vehicle charging load prediction, in particular to a short-term electric vehicle charging load probability prediction method based on composite quantile regression.
Background
With the continuous development of new energy technology, new energy industries such as electric vehicles, solar energy and the like will have larger and larger influence in daily life, and the electric vehicles have the characteristics of green, energy saving, environmental protection and high efficiency, and are important development in the future.
In recent years, the permeability of the electric automobile is rapidly improved, a series of problems of overrun of peak load of a transformer in a living area, insufficient capacity of the transformer and the like are caused, and the conventional living area distribution transformer cannot meet the requirement of large-scale unordered load access of the electric automobile. The ordered charging technology of the electric automobile is an effective method for solving the problems. The implementation and operation of most ordered charging strategies require the acquisition of predicted values of future electric vehicle charging loads.
The electric automobile charging load has high randomness and strong volatility, the point prediction of the load can not reflect all the characteristics of the load, the probability prediction can obtain the upper and lower limits of the confidence interval of the charging load, the probability distribution of the charging load can also be obtained, and the electric automobile charging load has larger reference value. Common neural network probability prediction works are based on Quantile Regression (QR) principles, where a prediction model can only output one quantile, while a Composite Quantile Regression (CQR) model can output multiple prediction quantiles simultaneously. At present, the electric automobile charging load probability prediction is less studied, especially in the aspect of neural network composite quantile regression.
Disclosure of Invention
The invention provides a short-term electric vehicle charging load probability prediction method based on composite quantile regression, which is based on electric vehicle charging load historical data, considers the influence of air temperature, weather type and day type, combines the composite quantile regression with an LSTM neural network, predicts a plurality of quantiles of future electric vehicle charging load, and obtains the upper limit and the lower limit of a confidence interval and probability distribution of the charging load.
The invention solves the technical problems by adopting the following technical scheme:
a short-term electric vehicle charging load probability prediction method based on composite quantile regression comprises the following steps:
s1, acquiring characteristic data related to electric vehicle charging load prediction, preprocessing the characteristic data, and constructing a training data set; the characteristic data comprise historical charging load, air temperature, weather type and day type of the electric automobile;
s2, building a load probability prediction model based on composite quantile regression and LSTM neural network, and inputting the model into an electric vehicle charging load sequence [ x ] in the past day 1 ,…,x n ]And air temperature, weather, and the like, matched to the time of the charge load sequenceType, day type sequence, where n is the number of sequence points; the load probability prediction model predicts n time points in the future simultaneously every time, each time point needs to predict k quantiles simultaneously, and the model output after inverse normalization is as follows:
wherein Q is yi (τ j |x i ) Input x predicted for model i Corresponding electric automobile charging load y i τ of (V) j Fractional number τ j For the j-th quantile, [ y ] 1 ,…,y n ]Representation and input x i The corresponding real value of the charging load of the electric automobile to be predicted, i=1, 2, …, n;
training the load probability prediction model by using the training data set with the minimum loss function as a target, and obtaining a final load probability prediction model;
s3, when actual prediction is carried out, charging load sequence [ x ] of electric automobile in day before date to be predicted 1 ,…,x n ]And the temperature, weather type and day type sequences matched with the time of the charging load sequence are input into the final load probability prediction model to obtain a fractional matrix Q of model output Y The method comprises the steps of carrying out a first treatment on the surface of the Then, for any ith future time point, the quantile matrix Q is selected according to the designated confidence interval Y Extracting confidence interval upper and lower limits of future charging load from ith row of the battery pack and using a quantile matrix Q τ Taking the 0.5 quantile in the ith row as a predicted value of the future charge load, thereby forming a predicted curve of the future charge load;
s4, according to the appointed prediction confidence, outputting a fractional number matrix Q based on the model Y K quantiles predicted at each future time point, and generating the electric vehicle charging load probability distribution at each future time point by using a nuclear density estimation method.
Preferably, when the feature data is constructed as a training sample, the weather type is quantized to a value between 0 and 1, and the day type is set to 0 and 1 according to the working day and the rest day, respectively.
Preferably, the model structure of the load probability prediction model in step S2 is as follows:
the charging load sequence is subjected to an input layer 1, a one-dimensional convolution layer, a pooling layer and a full connection layer to obtain intermediate data A1; the temperature, weather type and day type sequences are combined into a 3 Xn matrix, and the matrix is input into the layer 2 and the full connection layer together to obtain intermediate data A2; the intermediate data A1 and the intermediate data A2 are input into a splicing layer for splicing, then are output into a quantile prediction result Q through an output layer formed by cascading a full-connection layer and a shaping layer after passing through an LSTM layer, a Dropout layer and a full-connection layer Y 。
Preferably, the loss function used for training the load probability prediction model is as follows:
wherein L (τ) j ) Is equal to tau j A related function in the form of:
M(x i ) K quantiles predicted for the ith time pointRelated penalty terms for solving quantile crossover problem, penalty term M (x i ) The expression of (2) is:
where λ is a constant.
Preferably, the load probability prediction model predicts the charging load fraction of the electric vehicle for 24 hours in the future, n=96 time points in the future are predicted simultaneously at each time, the interval between adjacent time points is 15 minutes, and k=19 fractions are predicted simultaneously at each time point, namely tau is predicted simultaneously j ∈[0.05,…,0.95]And carrying out quantile prediction.
Preferably, in S4, the prediction confidence is determined from Q Y Extracting m quantiles from k quantiles predicted at each future time point, and generating probability distribution of electric vehicle charging load by using the m quantiles
Wherein y is l Is the first of the extracted m quantiles; h is the kernel bandwidth and K (·) is the kernel function used to construct the kernel density estimate.
Preferably, the kernel function K (·) is an Epanechnikov kernel function having the formula:
preferably, the kernel bandwidth h needs to search for an optimal value by adopting a grid search cross-validation method.
Preferably, in the step S3, the designated confidence interval is 90% confidence interval, and the ith future time point is determined from the quantile matrix Q Y Is extracted from the ith row of (2)And->Respectively as the upper and lower limits of the 90% confidence interval.
Preferably, in the step S4, the specified prediction confidence is 90% confidence, and the prediction confidence is determined from Q Y Extracting m=19 quantiles from k=19 quantiles predicted at each future point in time, from τ j ∈[0.05,…,0.95]Generating electric vehicle charging load probability distribution of each future time point; or the specified predictive confidence is 80% confidence, from Q Y Extracting m=17 quantiles from k=19 quantiles predicted at each future point in time, from τ j ∈[0.10,…,0.90]An electric vehicle charging load probability distribution for each future point in time is generated.
The invention provides a short-term electric vehicle charging load probability prediction method based on composite quantile regression, which is based on electric vehicle charging load historical data, considers the influence of air temperature, weather type and day type, uses a specific loss function training model, adds a punishment item into a loss function, solves the quantile crossing problem, outputs a series of prediction quantiles of future charging load, can obtain a charging load prediction curve, the upper and lower limits of a charging load confidence interval and the probability distribution of the charging load, has better prediction precision and prediction reliability, and the obtained prediction result can be used as the input quantity of an ordered charging strategy, thereby being beneficial to the design and operation of the ordered charging strategy.
Drawings
FIG. 1 is a schematic flow chart of a method in an embodiment of the invention;
FIG. 2 is a schematic diagram of a CQRLSTM model structure in an embodiment of the present invention;
FIG. 3 is a graph showing the effect of a prediction interval and a prediction curve according to an embodiment of the present invention;
FIG. 4 is a graph of the effect of a predictive probability density function provided by an embodiment of the invention;
Detailed Description
Embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. While the invention has been described in connection with certain embodiments thereof, it should be understood that the invention may be embodied in many different forms and should not be construed as limited to the embodiments set forth herein, but rather should be construed in view of a more thorough and complete understanding of the present invention. It should be understood that the drawings and embodiments of the invention are for illustration purposes only and are not intended to limit the scope of the present invention.
Referring to fig. 1, as a preferred embodiment of the present invention, a short-term electric vehicle charging load probability prediction method based on composite quantile regression is provided, and the inventive concept is as follows: firstly, 4 types of data including historical charging load, air temperature, weather type and day type of the electric automobile are obtained, and data are subjected to quantization and normalization pretreatment. Then, based on the set model structure and model input, and the specific loss function with penalty term, an LSTM neural network Composite Quantile Regression (CQRLSTM) prediction model is constructed. And training a prediction model by using historical data, inputting the data to be predicted into the model to obtain a series of quantile prediction results, and obtaining the upper limit and the lower limit of a prediction confidence interval from the model output results to obtain a prediction curve of the charging load of the electric automobile. And finally, generating a charging load probability density curve based on the output prediction result and a nuclear density estimation method. Specific steps of the above prediction method are described in detail below.
S1, acquiring characteristic data related to electric vehicle charging load prediction, and preprocessing the characteristic data to construct a training data set. The characteristic data include historical charging load, air temperature, weather type and day type of the electric automobile.
The historical charging load, the air temperature, the weather type and the day type of the electric automobile are all time sequence data, and the electric automobile can be segmented according to the required sample length. The training sample is constructed, the feature data with different dimensions can be normalized, and the influence of the dimension difference on the final prediction result is avoided. Since the weather type and the day type are not characteristic of continuous values, each class can be assigned in the range of 0 to 1 after the weather type is classified, and the day type is classified according to the working day and the rest day and is set to 0 and 1 respectively.
In the embodiment of the present invention, the weather types are classified into 15 types, each weather type is quantized to a value between 0 and 1, and the quantization method for the weather type is shown in table 1.
TABLE 1
Weather name Quantized value
Heavy Rain 1.0
Snow with big snow 0.9
Heavy rain 0.9
Middle snow 0.8
Middle rain 0.8
Thunder gust rain 0.7
Haze 0.6
Snow-making device 0.6
Rain shower 0.6
Snow with small size 0.5
Rain with small size 0.5
Mist spray 0.5
Yin type vagina 0.4
Clouds of people 0.3
Sunny days 0.2
S2, building a load probability prediction model based on composite quantile regression and an LSTM neural network (CQRLSTM), and inputting the model into an electric vehicle charging load sequence [ x ] in the past day 1 ,…,x n ]And a sequence of air temperature, weather type, day type matching the time of the charge load sequence, wherein n is a sequence point number; the load probability prediction model simultaneously predicts n future time points each time, each time point needs to simultaneously predict k quantiles of the charging load of the future electric automobile, and the model output after inverse normalization is as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,input x predicted for model i Corresponding electric automobile charging loady i τ of (V) j Fractional number τ j For the j-th quantile, [ y ] 1 ,…,y n ]Representation and input x i The corresponding real value of the charging load of the electric automobile to be predicted, i=1, 2, …, n.
In the embodiment of the present invention, in step S2, the load probability prediction model structure is shown in fig. 2, and the specific internal processing procedure is as follows:
the charging load sequence is subjected to an input layer 1, a one-dimensional convolution layer, a pooling layer and a full connection layer to obtain intermediate data A1; the temperature, weather type and day type sequences are combined into a 3 Xn matrix, and the matrix is input into the layer 2 and the full connection layer together to obtain intermediate data A2; the intermediate data A1 and the intermediate data A2 are input into a splicing layer for splicing, then are output into a quantile prediction result Q through an output layer formed by cascading a full-connection layer and a shaping layer after passing through an LSTM layer, a Dropout layer and a full-connection layer Y 。
The load probability prediction model needs to be trained in advance before being practically applied to reasoning, the training mode is similar to the traditional neural network, the minimum loss function can be used as a target, the load probability prediction model is trained by using the training data set, and a final load probability prediction model is obtained.
In the embodiment of the invention, the loss function adopted for training the load probability prediction model is as follows:
wherein L (τ) j ) Is equal to tau j A related function in the form of:
wherein, the liquid crystal display device comprises a liquid crystal display device,i.e. matrix Q output by load probability prediction model Y Corresponding value of (a) i.e. charging load y of electric automobile i τ of (V) j Quantile, which is related to the input x i And (5) correlation.
M(x i ) K quantiles predicted for the ith time pointA penalty term for solving the quantile crossover problem.
The penalty term M (x i ) The expression of (2) is:
where λ is a constant. In the embodiment of the invention, lambda takes a value of 0.01.
The training process of the load probability prediction model can be realized by referring to the prior art, generally, the preprocessed training data set in the step S1 can be divided into a training set and a testing set, the training set is utilized to train the load probability prediction model, the testing set is utilized to test the load probability prediction model, and the optimizer is utilized to optimize network parameters based on loss function values, so that the final load probability prediction model is obtained.
S3, when actual prediction is carried out, charging load sequence [ x ] of electric automobile in day before date to be predicted 1 ,…,x n ]And inputting the air temperature, weather type and day type sequences which are matched with the time of the charging load sequence into the final load probability prediction model obtained in S2 to obtain a quantile matrix Q of model output Y 。
Then for any ith future time point, the following steps are performed according to the specified confidence intervalFractional number matrix Q Y Extracting confidence interval upper and lower limits of future charging load from ith row of the battery pack and using a quantile matrix Q Y The 0.5 quantile in row i of (i) is taken as a predicted value of the future charge load, thereby forming a predicted curve of the future charge load.
The specified confidence interval and the lengths of the input sequence and the output sequence in the present invention may be specified according to actual prediction requirements. In the embodiment of the invention, the prediction target of the load probability prediction model is the charging load quantile of the electric automobile for 24 hours in the future, n=96 time points in the future are predicted simultaneously each time, the interval between adjacent time points is 15 minutes, and k=19 quantiles are predicted simultaneously each time point, namely tau is predicted simultaneously j ∈[0.05,…,0.95]And carrying out quantile prediction. At the same time, the assigned confidence interval may be set to be the 90% confidence interval, whereby any i-th future point in time is determined from the quantile matrix Q based on the 90% confidence interval Y Is extracted from the ith row of (2)And->Respectively as the upper and lower limits of the 90% confidence interval. The confidence interval is combined with the 0.5 quantile, so that the value corresponding to each time point in the prediction curve and the upper and lower limits of the confidence interval can be determined, and the prediction value and the prediction curve of the future charging load can be generated. The predicted value and the predicted curve generated by the embodiment of the invention can be used as the input quantity of the ordered charging strategy to support the design and operation of the ordered charging strategy.
S4, according to the appointed prediction confidence, outputting a fractional number matrix Q based on the model Y K quantiles predicted at each future time point, and generating the electric vehicle charging load probability distribution at each future time point by using a nuclear density estimation method.
In the step S4, the specified prediction confidence may be specified according to the actual prediction requirement. From the following componentsFractional matrix Q output from model Y There are n x k quantiles, each column corresponding to k quantiles at a point in time. Therefore, if the specified prediction confidence is different, the number of required bits is also different when the probability distribution of the electric vehicle charging load is generated by adopting the kernel density estimation method. According to the n multiplied by k quantile prediction results output by the model, m quantiles can be extracted from k quantiles predicted at each time point according to the requirements of prediction confidence intervals, and then the m quantiles are utilized to generate probability distribution of the charging load of the electric automobile based on a kernel density estimation method, wherein the probability distribution is expressed as follows:
wherein y is l Is the first of the extracted m quantiles; m is the total number of extracted predictive quantiles; k (·) is a kernel function, which in this example is chosen to construct a kernel density estimate having the general formula:
h is the core bandwidth, and has a great influence on the accuracy of non-parameter core density estimation, so that the optimal core bandwidth can be found by adopting a grid search-based cross-validation method. In the invention, the grid search cross-validation method belongs to the prior art, can be realized by self programming, and can also be realized by a plurality of existing tools, and the description is omitted. In an embodiment of the invention, the implementation is done using the relevant tools in the scikit-learn package.
In addition, as previously described, the number of quantiles m that need to be extracted for different prediction confidence intervals is different. In the embodiment of the present invention, if k=19, the number of extracted prediction quantiles m may be selected as 19 or 17, respectively, corresponding to the specified prediction confidence being 90% confidence and 80% confidence, respectively. When m=19, τ is used j ∈[0.05,…,0.95]All 19 predictive quantile results of (a) generate electric vehicle charging for each future point in timeThe probability distribution of the electric load, the coverage range of the generated probability density curve is 90% confidence interval; when m=17, τ is used j ∈[0.1,0.15,…,0.9]And generating electric vehicle charging load probability distribution of each future time point by 17 quantile prediction results, wherein the coverage range of the generated probability density curve is 80% confidence interval. The embodiment of the invention can generate probability density curves under two confidence degrees for subsequent application.
The electric vehicle charging load probability distribution described in S1 to S4 above, which is generated for each future point in time, is specifically applied to one example below to exhibit its technical effects.
Examples
In order to verify the effectiveness of the short-term electric vehicle charging load probability prediction method based on composite quantile regression, the method is developed and realized by using a python 3.8 programming language in the embodiment, and the test and verification of the embodiment are completed by using a PC (personal computer) provided with an Intel Core i5-8300H 2.3GHz CPU, an Nvidia GTX 1060 6G display card and a 32G memory.
The embodiment of the invention carries out technical verification based on the historical data of the charging load of the electric automobile in the residential area in a certain area of Hangzhou city. The data time used in this example was 2022, 7 months 1 day to 7 months 31 days, for a total of 31 days. The normalized data sample table is shown in table 2.
TABLE 2
Time Charging load Day type Air temperature Weather type
2022/7/15 22:45 0.83 0 0.61 0.23
2022/7/15 23:00 0.89 0 0.6 0.2
2022/7/15 23:15 0.92 0 0.6 0.2
2022/7/15 23:30 0.93 0 0.6 0.2
2022/7/15 23:45 0.94 0 0.6 0.2
2022/7/16 0:00 0.95 1 0.6 0.2
2022/7/16 0:15 0.95 1 0.59 0.2
2022/7/16 0:30 0.92 1 0.58 0.2
2022/7/16 0:45 0.89 1 0.56 0.2
2022/7/16 1:00 0.86 1 0.55 0.2
Data were used as training sets for the first 21 days and test sets for the last 9 days. The model is trained using a training set. After training is completed, testing is performed on the test set. According to the aforementioned step S3, taking the 0.5 quantile as the predicted value of the future charging load, the confidence interval is designated as the 90% confidence interval, and the corresponding predicted result on the 2 nd day is shown in fig. 3, so that the predicted curve has a good matching degree with the actual curve, the predicted interval completely covers the actual value, and the prediction reliability is high. Further, according to the aforementioned step S4, taking 00 in the prediction result of the 2 nd day: 00. 04: 00. 08: 00. 12: 00. 16: 00. 20:00 total 6 time points, the probability density curves of 90% and 80% confidence intervals generated by the model are shown in fig. 4, and the probability density curves obtained by prediction can be seen to have good effect. This example demonstrates the effectiveness of the proposed method of the present invention.
The foregoing is a preferred embodiment of the present invention and it should be noted that modifications and additions to those skilled in the art may be made without departing from the principles of the present invention, which modifications and additions are also considered to be within the scope of the present invention.

Claims (10)

1. A short-term electric vehicle charging load probability prediction method based on composite quantile regression is characterized by comprising the following steps:
s1, acquiring characteristic data related to electric vehicle charging load prediction, preprocessing the characteristic data, and constructing a training data set; the characteristic data comprise historical charging load, air temperature, weather type and day type of the electric automobile;
s2, building a load probability prediction model based on composite quantile regression and LSTM neural network, and inputting the model into an electric vehicle charging load sequence [ x ] in the past day 1 ,…,x n ]And a sequence of air temperature, weather type, day type matching the time of the charge load sequence, wherein n is a sequence point number; the load probability prediction model predicts n time points in the future simultaneously every time, each time point needs to predict k quantiles simultaneously, and the model output after inverse normalization is as follows:
wherein Q is yi (τ j |x i ) Input x predicted for model i Corresponding electric automobile charging load y i τ of (V) j Fractional number τ j For the j-th quantile, [ y ] 1 ,…,y n ]Representation and input x i The corresponding real value of the charging load of the electric automobile to be predicted, i=1, 2, …, n;
training the load probability prediction model by using the training data set with the minimum loss function as a target, and obtaining a final load probability prediction model;
s3, when actual prediction is carried out, charging load sequence [ x ] of electric automobile in day before date to be predicted 1 ,…,x n ]And the temperature, weather type and day type sequences matched with the time of the charging load sequence are input into the final load probability prediction model to obtain a fractional matrix Q of model output Y The method comprises the steps of carrying out a first treatment on the surface of the Then, for any ith future time point, the quantile matrix Q is selected according to the designated confidence interval Y Extracting confidence interval upper and lower limits of future charging load from ith row of the battery pack and using a quantile matrix Q Y Taking the 0.5 quantile in the ith row as a predicted value of the future charge load, thereby forming a predicted curve of the future charge load;
s4, according to the appointed prediction confidence, outputting a fractional number matrix Q based on the model Y K quantiles predicted at each future time point, and generating the electric vehicle charging load probability distribution at each future time point by using a nuclear density estimation method.
2. The short-term electric vehicle charging load probability prediction method based on the CQRLSTM as claimed in claim 1, wherein the method comprises the following steps of: when the characteristic data is constructed as a training sample, the weather type is required to be quantitatively processed to be a value between 0 and 1, and the day type is respectively set to be 0 and 1 according to the working day and the rest day.
3. The short-term electric vehicle charging load probability prediction method based on the CQRLSTM as claimed in claim 1, wherein the method comprises the following steps of: the model structure of the load probability prediction model in step S2 is as follows:
the charging load sequence is subjected to an input layer 1, a one-dimensional convolution layer, a pooling layer and a full connection layer to obtain intermediate data A1; the temperature, weather type and day type sequences are combined into a 3 Xn matrix, and the matrix is input into the layer 2 and the full connection layer together to obtain intermediate data A2; the intermediate data A1 and the intermediate data A2 are input into a splicing layer for splicing, and then are subjected to LSTM layer, dropout layer,After the full connection layer, the final output is a quantile prediction result Q through an output layer formed by cascading the full connection layer and the shaping layer Y 。
4. The short-term electric vehicle charging load probability prediction method based on the CQRLSTM as claimed in claim 1, wherein the method comprises the following steps of: the loss function adopted for training the load probability prediction model is as follows:
wherein L (τ) j ) Is equal to tau j A related function in the form of:
ρ τ (u)＝u(τ-I(u)),u＝y i -Q yi (τ j |x i )
M(x u ) K quantiles Q predicted for the ith time point yi (τ 1 |x i )～Q yi (τ k |x i ) Related penalty terms for solving quantile crossover problem, penalty term M (x i ) The expression of (2) is:
where λ is a constant.
5. The short-term electric vehicle charging load probability prediction method based on the CQRLSTM as claimed in claim 1, wherein the method comprises the following steps of: pre-prediction of the load probability prediction modelThe measurement target is the charge load fraction of the electric automobile for 24 hours in the future, n=96 time points in the future are predicted simultaneously each time, the interval between adjacent time points is 15 minutes, and k=19 fractions are predicted simultaneously each time point, namely tau is predicted simultaneously j ∈[0.05,…,0.95]And carrying out quantile prediction.
6. The short-term electric vehicle charging load probability prediction method based on the CQRLSTM as claimed in claim 1, wherein the method comprises the following steps of: in S4, according to the appointed prediction confidence, Q is selected from Y Extracting m quantiles from k quantiles predicted at each future time point, and generating probability distribution of electric vehicle charging load by using the m quantiles
Wherein y is l Is the first of the extracted m quantiles; h is the kernel bandwidth and K (·) is the kernel function used to construct the kernel density estimate.
7. The short-term electric vehicle charging load probability prediction method based on the CQRLSTM as claimed in claim 6, wherein the method comprises the following steps: the kernel function K (·) adopts an Epanechnikov kernel function, and has a general formula:
8. the short-term electric vehicle charging load probability prediction method based on the CQRLSTM as claimed in claim 6, wherein the method comprises the following steps: the core bandwidth h needs to search an optimal value by adopting a grid search cross-validation method.
9. As claimed inThe short-term electric vehicle charging load probability prediction method based on CQRLSTM, which is characterized in that: in S3, the designated confidence interval is 90% confidence interval, and the ith future time point is determined from the quantile matrix Q Y Extracts Q from line i of (1) yi (τ 1 |x i ) And Q yi (τ 19 |x i ) Respectively as the upper and lower limits of the 90% confidence interval.
10. The short-term electric vehicle charging load probability prediction method based on the CQRLSTM as claimed in claim 6, wherein the method comprises the following steps: in S4, the specified prediction confidence is 90% confidence, and Q is required Y Extracting m=19 quantiles from k=19 quantiles predicted at each future point in time, from τ j ∈[0.05,…,0.95]Generating electric vehicle charging load probability distribution of each future time point; or the specified predictive confidence is 80% confidence, from Q Y Extracting m=17 quantiles from k=19 quantiles predicted at each future point in time, from τ j ∈[0.10,…,0.90]An electric vehicle charging load probability distribution for each future point in time is generated.
CN202310674536.0A 2023-06-07 2023-06-07 Short-term electric vehicle charging load probability prediction method based on composite quantile regression Pending CN116796892A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310674536.0A CN116796892A (en) 2023-06-07 2023-06-07 Short-term electric vehicle charging load probability prediction method based on composite quantile regression

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310674536.0A CN116796892A (en) 2023-06-07 2023-06-07 Short-term electric vehicle charging load probability prediction method based on composite quantile regression

Publications (1)

Publication Number Publication Date
CN116796892A true CN116796892A (en) 2023-09-22

Family

ID=88049014

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310674536.0A Pending CN116796892A (en) 2023-06-07 2023-06-07 Short-term electric vehicle charging load probability prediction method based on composite quantile regression

Country Status (1)

Country Link
CN (1) CN116796892A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117172391A (en) * 2023-11-02 2023-12-05 广东电网有限责任公司 Line loss reasonable interval prediction method, device and medium based on multiple regression

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117172391A (en) * 2023-11-02 2023-12-05 广东电网有限责任公司 Line loss reasonable interval prediction method, device and medium based on multiple regression
CN117172391B (en) * 2023-11-02 2024-02-27 广东电网有限责任公司 Line loss reasonable interval prediction method, device and medium based on multiple regression

Similar Documents

Publication Publication Date Title
Li et al. Photovoltaic power forecasting with a hybrid deep learning approach
CN105069525B (en) Round-the-clock 96 Day Load Curve Forecastings and optimization update the system
Hu et al. A hybrid model based on CNN and Bi-LSTM for urban water demand prediction
Cheng et al. Probabilistic residential load forecasting based on micrometeorological data and customer consumption pattern
CN111915092B (en) Ultra-short-term wind power prediction method based on long-short-term memory neural network
CN111754037B (en) Long-term load hybrid prediction method for regional terminal integrated energy supply system
CN112465256A (en) Building power consumption prediction method and system based on Stacking model fusion
Lu et al. A hybrid model based on convolutional neural network and long short-term memory for short-term load forecasting
Wang et al. Short-term electricity price forecasting based on similarity day screening, two-layer decomposition technique and Bi-LSTM neural network
CN116796892A (en) Short-term electric vehicle charging load probability prediction method based on composite quantile regression
Ghasvarian Jahromi et al. A novel method for day-ahead solar power prediction based on hidden Markov model and cosine similarity
CN111242355A (en) Photovoltaic probability prediction method and system based on Bayesian neural network
CN113193551A (en) Short-term power load prediction method based on multi-factor and improved feature screening strategy
CN111815026A (en) Multi-energy system load prediction method based on feature clustering
CN108428055A (en) A kind of load characteristics clustering method considering load vertical characteristics
CN105678406A (en) Short-term load prediction method based on cloud model
CN114418174A (en) Electric vehicle charging load prediction method
CN115358437A (en) Power supply load prediction method based on convolutional neural network
CN115456287A (en) Long-and-short-term memory network-based multi-element load prediction method for comprehensive energy system
CN113344279B (en) Resident load prediction method based on LSTM-SAM model and pooling
CN115759389A (en) Day-ahead photovoltaic power prediction method based on weather type similar day combination strategy
CN116169670A (en) Short-term non-resident load prediction method and system based on improved neural network
Xie et al. A multi-information fusion model for short term load forecasting of an architectural complex considering spatio-temporal characteristics
Bao et al. Short-term electricity price forecasting based on empirical mode decomposition and deep neural network
CN115481788B (en) Phase change energy storage system load prediction method and system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination