A short-term electric vehicle charging load probability prediction method based on composite quantile regression

Abstract

The application discloses a short-term electric vehicle charging load probability prediction method based on composite quantile regression. First, four types of data, including electric vehicle historical charging load, air temperature, weather type and day type, are obtained, and the data is quantified and normalized for pretreatment. Then, based on the set model structure and model input, and a specific loss function with a penalty term, an LSTM neural network composite quantile regression (CQRLSTM) prediction model is constructed. The historical data is used to train the prediction model, the to-be-predicted data is input into the model to obtain a series of quantile prediction results, the upper and lower limits of the prediction confidence interval are obtained from the model output results, and the prediction curve of the electric vehicle charging load is obtained. Finally, based on the output prediction results and the kernel density estimation method, the charging load probability density curve is generated. The obtained prediction curve and prediction interval can be used as the input quantity of the ordered charging strategy, and help the design and operation of the ordered charging strategy.

Description

Technical Field

[0001] This invention relates to the field of electric vehicle charging load forecasting, and in particular to a short-term electric vehicle charging load probability forecasting method based on composite quantile regression. Background Technology

[0002] With the continuous development of new energy technologies, new energy industries such as electric vehicles and solar energy will have an increasingly significant impact on daily life. Electric vehicles are characterized by being green, energy-saving, environmentally friendly, and highly efficient, making them a key area for future development.

[0003] In recent years, the rapid increase in electric vehicle penetration has led to a series of problems, such as excessive peak loads on residential transformers and insufficient transformer capacity. Existing residential transformers cannot meet the needs of large-scale, uncontrolled electric vehicle loads. Ordered charging technology for electric vehicles is an effective way to solve these problems. The implementation and operation of most ordered charging strategies require obtaining forecasts of future electric vehicle charging loads.

[0004] Electric vehicle charging loads are highly random and volatile, making point-based load predictions insufficient to reflect all their characteristics. Probabilistic prediction, however, can provide the upper and lower limits of the confidence interval for the charging load, as well as its probability distribution, offering greater reference value. Common neural network probabilistic prediction methods are based on quantile regression (QR) principles, but these models can only output a single quantile. In contrast, composite quantile regression (CQR) models can output multiple predicted quantiles simultaneously. Currently, research on probabilistic prediction of electric vehicle charging loads is limited, particularly in the area of ​​neural network composite quantile regression. Summary of the Invention

[0005] This invention provides a short-term electric vehicle charging load probability prediction method based on composite quantile regression. Based on historical data of electric vehicle charging load, considering the influence of temperature, weather type and day type, and combining composite quantile regression and LSTM neural network, it simultaneously predicts multiple quantiles of future electric vehicle charging load to obtain the upper and lower limits of the confidence interval and the probability distribution of charging load.

[0006] The technical problem solved by this invention is achieved through the following technical solution:

[0007] A short-term electric vehicle charging load probability prediction method based on composite quantile regression includes the following steps:

[0008] S1. Collect feature data related to electric vehicle charging load prediction, and preprocess the feature data to construct a training dataset; the feature data includes historical charging load of electric vehicles, temperature, weather type and day type;

[0009] S2. Establish a load probability prediction model based on composite quantile regression and LSTM neural network. The model input is the electric vehicle charging load sequence [x1,…,x] over the past day. n [and temperature, weather type, and day type sequences that match the charging load sequence in time, where n is the number of sequence points; the load probability prediction model simultaneously predicts n future time points each time, and k quantiles need to be predicted simultaneously for each time point. The output of the model after inverse normalization is:]

[0010]

[0011] Among them, Q yi (τ j |x i ) represents the input x predicted by the model. i The corresponding electric vehicle charging load y i τ j Quantiles, τ j For the j-th quantile, [y1,…,y n ] indicates the relationship with input x i The corresponding actual values ​​of the electric vehicle charging load to be predicted, i = 1, 2, ..., n;

[0012] With the goal of minimizing the loss function, the load probability prediction model is trained using the training dataset to obtain the final load probability prediction model.

[0013] S3. When making actual forecasts, the electric vehicle charging load sequence [x1,…,x] for the day before the forecast date will be used. n The temperature, weather type, and day type sequences that match the charging load sequence are input into the final load probability prediction model to obtain the quantile matrix Q of the model output. Y Then, for any i-th future time point, according to the specified confidence interval, from the quantile matrix Q... Y Extract the upper and lower bounds of the confidence interval for future charging load from the i-th row, and use the quantile matrix Q τ The 0.5 quantile in the i-th row is used as the predicted value of future charging load, thus forming the prediction curve of future charging load;

[0014] S4. Based on the specified prediction confidence level, and using the quantile matrix Q output by the model... Y The probability distribution of electric vehicle charging load at each future time point is generated using the kernel density estimation method based on the predicted k quantiles for each future time point.

[0015] Preferably, when the feature data is used to construct training samples, the weather type needs to be quantized to a value between 0 and 1, while the day type is set to 0 and 1 for weekdays and rest days, respectively.

[0016] Preferably, the model structure of the load probability prediction model in step S2 is as follows:

[0017] The charging load sequence is processed through input layer 1, a one-dimensional convolutional layer, a pooling layer, and a fully connected layer to obtain intermediate data A1. The temperature, weather type, and day type sequences are merged into a 3×n matrix, which is then processed through input layer 2 and a fully connected layer to obtain intermediate data A2. Intermediate data A1 and intermediate data A2 are concatenated in a stitching layer, then passed through an LSTM layer, a Dropout layer, and a fully connected layer. Finally, the output layer, composed of a fully connected layer and a shaping layer, outputs the quantile prediction result Q. Y .

[0018] Preferably, the loss function used to train the load probability prediction model is:

[0019]

[0020] Wherein, L(τ) j ) is related to τ j The relevant function has the following form:

[0021]

[0022]

[0023]

[0024] M(x i ) represents the k quantiles predicted at the i-th time point. The relevant penalty term used to solve the quantile crossover problem is the penalty term M(x). i The expression for ) is:

[0025]

[0026] Where λ is a constant.

[0027] Preferably, the load probability prediction model aims to predict the charging load quantiles of electric vehicles over the next 24 hours. It simultaneously predicts n = 96 future time points, with adjacent time points spaced 15 minutes apart. For each time point, k = 19 quantiles need to be predicted simultaneously, i.e., for τ... j Quantile prediction is performed for values ​​in the range [0.05, ..., 0.95].

[0028] Preferably, in step S4, the prediction confidence level needs to be determined from Q. Y For each future time point, extract m quantiles from the k quantiles predicted, and then use these m quantiles to generate the probability distribution of electric vehicle charging load.

[0029]

[0030] Among them, y l is the l-th quantile among the extracted m quantiles; h is the kernel bandwidth; and K(·) is the kernel function used to construct the kernel density estimate.

[0031] Preferably, the kernel function K(·) adopts the Epanechnikov kernel function, whose general formula is:

[0032]

[0033] Preferably, the core bandwidth h should be found using a grid search cross-validation method.

[0034] Preferably, in S3, the specified confidence interval is a 90% confidence interval, and the i-th future time point needs to be determined from the quantile matrix Q. Y Extract from the i-th row and These are used as the upper and lower limits of the 90% confidence interval, respectively.

[0035] Preferably, in S4, the specified prediction confidence level is 90%, which needs to be derived from Q. Y Extracting m = 19 quantiles from the k = 19 quantiles predicted for each future time point, and using τ j Generate the probability distribution of electric vehicle charging load for each future time point ∈ [0.05,…,0.95]; or specify a prediction confidence level of 80%, and need to generate the probability distribution from Q. Y From the k=19 quantiles predicted at each future time point, extract the m=17 quantiles, and then use τ j Generate the electric vehicle charging load probability distribution for each future time point in the range [0.10,…,0.90].

[0036] This invention provides a short-term electric vehicle charging load probability prediction method based on composite quantile regression. Based on historical data of electric vehicle charging load, it considers the influence of temperature, weather type, and day type, and uses a specific loss function to train the model. By adding a penalty term to the loss function, the quantile crossover problem is solved, and a series of predicted quantiles of future charging load are output. The predicted curve of charging load, the upper and lower limits of the confidence interval of charging load, and the probability distribution of charging load can be obtained. It has good prediction accuracy and reliability. The prediction results can be used as input for ordered charging strategies, which helps in the design and operation of ordered charging strategies. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the method flow in an embodiment of the present invention;

[0038] Figure 2 This is a schematic diagram of the CQRLSTM model structure in an embodiment of the present invention;

[0039] Figure 3 This is a diagram showing the prediction interval and prediction curve effect provided in the embodiments of the present invention;

[0040] Figure 4 This is a graph showing the effect of the prediction probability density function provided in the embodiment of the present invention; Detailed Implementation

[0041] Embodiments of the present invention will now be described in more detail with reference to the accompanying drawings. While some embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various different forms and should not be construed as limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the invention. It should be understood that the accompanying drawings and embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the invention.

[0042] Please see the appendix Figure 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. The inventive concept is as follows: First, four types of data are obtained: historical electric vehicle charging load, temperature, weather type, and day type. These data are then preprocessed by quantization and normalization. Next, based on a defined model structure and input, and a specific loss function with a penalty term, an LSTM neural network composite quantile regression (CQRLSTM) prediction model is constructed. The prediction model is trained using historical data, and the data to be predicted is input into the model to obtain a series of quantile prediction results. The upper and lower limits of the prediction confidence interval are obtained from the model output, thus acquiring the electric vehicle charging load prediction curve. Finally, based on the output prediction results and a kernel density estimation method, a charging load probability density curve is generated. The specific steps of the above prediction method are described in detail below.

[0043] S1. Collect feature data related to electric vehicle charging load prediction, and preprocess this feature data to construct a training dataset. The aforementioned feature data includes historical electric vehicle charging load, temperature, weather type, and day type.

[0044] The aforementioned historical charging load, temperature, weather type, and day type for electric vehicles are all time-series data, which can be segmented according to the required sample length. When constructing training samples, the feature data of different dimensions can be normalized to avoid the impact of differences in units on the final prediction results. Since weather type and day type are not continuous numerical features, weather type can be classified, and each category can be assigned a value in the range of 0 to 1. Day type can be classified into weekdays and rest days, and set to 0 and 1 respectively.

[0045] In an embodiment of the present invention, weather types are divided into 15 categories, and each weather type is quantified to a value between 0 and 1. The quantification method for weather types is shown in Table 1.

[0046] Table 1

[0047] Weather Name Quantized value rainstorm 1.0 heavy snow 0.9 heavy rain 0.9 Moderate snow 0.8 Moderate rain 0.8 Thunderstorm 0.7 haze 0.6 Snowfall 0.6 shower 0.6 Light Snow 0.5 Light rain 0.5 fog 0.5 Negative 0.4 partly cloudy 0.3 clear 0.2

[0048] S2. Establish a load probability prediction model based on composite quantile regression and LSTM neural network (CQRLSTM). The model input is the electric vehicle charging load sequence [x1,…,x] over the past day. n The model also includes temperature, weather type, and day type sequences that match the charging load sequence, where n is the number of sequence points. The load probability prediction model simultaneously predicts n future time points each time, and for each time point, it simultaneously predicts k quantiles of the future electric vehicle charging load. The output of the inverse normalized model is:

[0049]

[0050] in, The input x obtained by the model prediction i The corresponding electric vehicle charging load y i τ j Quantiles, τ j For the j-th quantile, [y1,…,y n ] indicates the relationship with input x i The corresponding actual value of the electric vehicle charging load to be predicted, i = 1, 2, ..., n.

[0051] In an embodiment of the present invention, the structure of the load probability prediction model in step S2 is as shown in the appendix. Figure 2 As shown, the specific internal processing procedure is as follows:

[0052] The charging load sequence is processed through input layer 1, a one-dimensional convolutional layer, a pooling layer, and a fully connected layer to obtain intermediate data A1. The temperature, weather type, and day type sequences are merged into a 3×n matrix, which is then processed through input layer 2 and a fully connected layer to obtain intermediate data A2. Intermediate data A1 and intermediate data A2 are concatenated in a stitching layer, then passed through an LSTM layer, a Dropout layer, and a fully connected layer. Finally, the output layer, composed of a fully connected layer and a shaping layer, outputs the quantile prediction result Q. Y .

[0053] Before the above-mentioned load probability prediction model can be applied to inference, it needs to be trained in advance. The training method is similar to that of traditional neural networks. The goal is to minimize the loss function. The above-mentioned training dataset is used to train the load probability prediction model and obtain the final load probability prediction model.

[0054] In an embodiment of the present invention, the loss function used to train the above-mentioned load probability prediction model is:

[0055]

[0056] Wherein, L(τ) j ) is related to τ j The relevant function has the following form:

[0057]

[0058]

[0059]

[0060] in, That is, the matrix Q output by the load probability prediction model. Y The corresponding value in the table, i.e., the electric vehicle charging load y i τ j Quantiles, which are related to the input x i Related.

[0061] M(x i ) represents the k quantiles predicted at the i-th time point. The relevant penalty terms used to solve the quantile crossover problem.

[0062] The above penalty term M(x) i The expression for ) is:

[0063]

[0064] Wherein, λ is a constant. In the embodiments of the present invention, λ is 0.01.

[0065] The training process of the above-mentioned load probability prediction model can be implemented with reference to existing technologies. Generally, the preprocessed training dataset in step S1 can be divided into a training set and a test set. The load probability prediction model is trained using the training set and tested using the test set. The network parameters are optimized by the optimizer based on the loss function value to obtain the final load probability prediction model.

[0066] S3. When making actual forecasts, the electric vehicle charging load sequence [x1,…,x] for the day before the forecast date will be used. n The quantile matrix Q is obtained from the final load probability prediction model, which is derived from the input of temperature, weather type, and day type sequences that match the charging load sequence in time, and from these sequences. Y .

[0067] Then, for any i-th future time point, according to the specified confidence interval, from the quantile matrix Q... Y Extract the upper and lower bounds of the confidence interval for future charging load from the i-th row, and use the quantile matrix Q Y The 0.5 quantile in the i-th row is used as the predicted value of future charging load, thus forming the prediction curve of future charging load.

[0068] It should be noted that the specified confidence interval, as well as the lengths of the input and output sequences, mentioned in this invention can be specified according to actual prediction requirements. In the embodiments of this invention, the prediction target of the load probability prediction model is the charging load quantile of electric vehicles in the next 24 hours. Simultaneous predictions are made for n = 96 future time points, with an interval of 15 minutes between adjacent time points. For each time point, k = 19 quantiles need to be predicted simultaneously, i.e., for τ... j Quantile prediction is performed within the range [0.05, ..., 0.95]. Simultaneously, a specified confidence interval can be set to a 90% confidence interval. Based on this 90% confidence interval, any i-th future time point needs to be predicted from the quantile matrix Q. Y Extract from the i-th row and These are respectively used as the upper and lower limits of the 90% confidence interval. Combining this confidence interval with the aforementioned 0.5 quantile, the value corresponding to each time point in the prediction curve and its upper and lower confidence interval limits can be determined, thereby generating the predicted value and prediction curve for future charging load. The predicted value and prediction curve generated by the embodiments of the present invention can be used as input to an ordered charging strategy to support the design and operation of the ordered charging strategy.

[0069] S4. Based on the specified prediction confidence level, and using the quantile matrix Q output by the model... Y The probability distribution of electric vehicle charging load at each future time point is generated using the kernel density estimation method based on the predicted k quantiles for each future time point.

[0070] It should be noted that the specified prediction confidence level mentioned in step S4 above can be specified according to the actual prediction requirements. This is because the quantile matrix Q output by the model... Y The model has n×k quantiles, with each column corresponding to k quantiles at a given time point. Therefore, the number of quantiles required to generate the probability distribution of electric vehicle charging load using kernel density estimation differs depending on the specified prediction confidence level. Based on the n×k quantile prediction results from the model output, m quantiles can be extracted from the predicted k quantiles at each time point according to the required prediction confidence interval. These m quantiles are then used to generate the probability distribution of electric vehicle charging load using kernel density estimation, as follows:

[0071]

[0072] Among them, y l Let be the l-th quantile among the m extracted quantiles; m is the total number of extracted predicted quantiles; K(·) is the kernel function. In this embodiment, the Epanechnikov kernel function is selected to construct the kernel density estimate, and its general formula is:

[0073]

[0074] h represents the kernel bandwidth, which significantly impacts the accuracy of nonparametric kernel density estimation. Therefore, a grid search cross-validation method can be used to find the optimal kernel bandwidth. In this invention, the grid search cross-validation method is existing technology and can be implemented through custom programming or various existing tools; therefore, it will not be elaborated further. In the embodiments of this invention, relevant tools from the scikit-learn package are used for implementation.

[0075] Furthermore, as mentioned earlier, the number of quantiles *m* required to be extracted varies depending on the prediction confidence interval. In the embodiments of this invention, if *k* = 19, the number of prediction quantiles *m* extracted can be selected as 19 or 17, corresponding to a specified prediction confidence level of 90% and 80%, respectively. When *m* = 19, τ is used... j The 19 predicted quantiles ∈ [0.05,…,0.95] are used to generate the probability distribution of electric vehicle charging load at each future time point, and the generated probability density curve covers the 90% confidence interval; when m=17, τ is used. j The prediction results of 17 quantiles ∈ [0.1, 0.15, ..., 0.9] generate the probability distribution of electric vehicle charging load at each future time point, and the generated probability density curve covers the 80% confidence interval. Embodiments of this invention will generate probability density curves at two confidence levels for subsequent applications.

[0076] The following example demonstrates the technical effectiveness of generating the electric vehicle charging load probability distribution for each future time point as described in S1 to S4.

[0077] Example

[0078] To verify the effectiveness of the short-term electric vehicle charging load probability prediction method based on composite quantile regression proposed in this invention, this embodiment developed and implemented the method using the Python 3.8 programming language, and completed the testing and verification of this embodiment using a PC equipped with an Intel Core i5-8300H 2.3GHz CPU, an Nvidia GTX 1060 6G graphics card, and 32GB of memory.

[0079] This invention's embodiments are based on historical electric vehicle charging load data from a residential area in Hangzhou for technical verification. The data used in this example covers the period from July 1st to July 31st, 2022, a total of 31 days. A sample table of the normalized data is shown in Table 2.

[0080] Table 2

[0081] time Charging load Daily Type 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

[0082] The data from the first 21 days was used as the training set, and the data from the last 9 days was used as the test set. The model was trained using the training set. After training, it was tested on the test set. Following step S3 above, the 0.5 quantile was taken as the predicted value for future charging load, and the confidence interval was specified as a 90% confidence interval. The corresponding prediction result for the second day is attached. Figure 3 As shown, the predicted curve matches the actual curve well, and the predicted interval completely covers the actual value, indicating high prediction reliability. Further, following step S4 above, taking the prediction results for the second day at six time points (00:00, 04:00, 08:00, 12:00, 16:00, and 20:00), the probability density curves for the 90% and 80% confidence intervals generated by the model are shown in the attached figure. Figure 4 As shown, the predicted probability density curve performs well. This embodiment demonstrates the effectiveness of the method proposed in this invention.

[0083] The above are preferred embodiments of the present invention. It should be noted that those skilled in the art can make several improvements and additions without departing from the principle of the present invention, and these improvements and additions are also considered to be within the scope of protection of the present invention.

Claims

1. A short-term electric vehicle charging load probability prediction method based on composite quantile regression, characterized in that, Includes the following steps: S1. Collect feature data related to electric vehicle charging load prediction, and preprocess the feature data to construct a training dataset; the feature data includes historical charging load of electric vehicles, temperature, weather type and day type; S2. Establish a load probability prediction model based on composite quantile regression and LSTM neural network. The model input is the electric vehicle charging load sequence of the past day. And temperature, weather type, and day type sequences that match the charging load sequence in time, among which The number of points in the sequence; Load probability prediction model for the future Simultaneous forecasting is required at each time point, and each time point needs to be forecasted simultaneously. The output of the model after inverse normalization is: (The quantiles are given.) in, The input obtained by the model prediction Corresponding electric vehicle charging load of Quantiles For the j-th quantile, Representation and Input The corresponding actual value of the electric vehicle charging load to be predicted, i=1,2,…,n; With the goal of minimizing the loss function, the load probability prediction model is trained using the training dataset to obtain the final load probability prediction model. S3. When making actual forecasts, the electric vehicle charging load sequence for the day preceding the forecast date will be used. The temperature, weather type, and day type sequences that match the charging load sequence are input into the final load probability prediction model to obtain the quantile matrix of the model output. Then, for any i-th future time point, according to the specified confidence interval, from the quantile matrix... Extract the upper and lower bounds of the confidence interval for future charging load from the i-th row, and use the quantile matrix. The 0.5 quantile in the i-th row is used as the predicted value of future charging load, thus forming the prediction curve of future charging load; S4. Based on the specified prediction confidence level, calculate the quantile matrix output by the model. The predictions for each future point in time Using the kernel density estimation method, the probability distribution of electric vehicle charging load at each future time point is generated by using the quantiles. In step S4, the prediction confidence level needs to be determined from... The predictions for each future point in time Extract m quantiles from the quantiles, and then use these m quantiles to generate the probability distribution of electric vehicle charging load. : in, h represents the l-th quantile out of the extracted m quantiles; h is the kernel bandwidth. It is the kernel function used to construct kernel density estimates; The kernel function Using the Epanechnikov kernel function, its general formula is: The optimal value of the kernel bandwidth h needs to be found using a grid search cross-validation method.

2. The short-term electric vehicle charging load probability prediction method based on composite quantile regression as described in claim 1, characterized in that: When the feature data is used to construct training samples, the weather type needs to be quantized to a value between 0 and 1, while the day type is set to 0 and 1 for weekdays and rest days, respectively.

3. The short-term electric vehicle charging load probability prediction method based on composite quantile regression as described in claim 1, characterized in that: The model structure of the load probability prediction model described in step S2 is as follows: The charging load sequence is processed through input layer 1, a one-dimensional convolutional layer, a pooling layer, and a fully connected layer to obtain intermediate data A1; the temperature, weather type, and daily type sequences are merged into one. The matrices are fed into input layer 2 and a fully connected layer to obtain intermediate data A2. Intermediate data A1 and A2 are then concatenated in a concatenation layer, followed by an LSTM layer, a Dropout layer, and a fully connected layer. Finally, the output layer, composed of a cascaded fully connected layer and an shaping layer, outputs the quantile prediction result. .

4. The short-term electric vehicle charging load probability prediction method based on composite quantile regression as described in claim 1, characterized in that: The loss function used to train the load probability prediction model is: in, Is with The relevant function has the following form: For the prediction at the i-th time point quantiles The relevant penalty term used to solve the quantile crossover problem, penalty term The expression is: in, It is a constant.

5. The short-term electric vehicle charging load probability prediction method based on composite quantile regression as described in claim 1, characterized in that: The load probability prediction model aims to predict the charging load quantile of electric vehicles over the next 24 hours, and each prediction is based on the future... Simultaneous prediction is required for each time point, with adjacent time points spaced 15 minutes apart. quantiles, that is, for Perform quantile prediction.

6. The short-term electric vehicle charging load probability prediction method based on composite quantile regression as described in claim 1, characterized in that: In S3, the specified confidence interval is a 90% confidence interval, and the i-th future time point needs to be determined from the quantile matrix. Extract from the i-th row and These are used as the upper and lower limits of the 90% confidence interval, respectively.

7. The short-term electric vehicle charging load probability prediction method based on composite quantile regression as described in claim 1, characterized in that: In S4, the specified prediction confidence level is 90%, which needs to be derived from... The predictions for each future point in time Extract m=19 quantiles from the quantiles, and then... Generate the probability distribution of electric vehicle charging load for each future time point; or, if the specified prediction confidence level is 80%, it needs to be generated from... The predictions for each future point in time Extract m=17 quantiles from the quantiles, and then... Generate the electric vehicle charging load probability distribution for each future time point.

Images