Short-term probabilistic power prediction method and system applied to peak regulation scene
By using an LSTM deep learning model and a probabilistic prediction framework, the problem of nonlinear characteristics and uncertainty assessment in electric vehicle charging load forecasting is solved, achieving high-precision short-term probabilistic power forecasting and supporting scientific grid scheduling and equipment configuration.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- STATE GRID JIANGXI ELECTRIC POWER CO LTD
- Filing Date
- 2025-12-30
- Publication Date
- 2026-05-01
AI Technical Summary
Existing electric vehicle charging load forecasting methods struggle to handle nonlinear characteristics and lack assessment of the uncertainty range of forecast results, leading to insufficient robustness of grid dispatching strategies.
A short-term power prediction model is constructed by using a deep learning model based on a long short-term memory network (LSTM) and combining it with a probabilistic prediction framework. This model generates power prediction intervals and daily charging power probability distributions at different confidence levels, and evaluates the peak shaving or valley filling potential of electric vehicles at specified confidence levels.
It improves forecast accuracy, quantifies forecast uncertainty, provides scientific and reliable dispatch decision support, and helps the power grid to rationally allocate equipment and formulate scientific demand response strategies.
Smart Images
Figure CN121965489A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power prediction technology, and in particular to a short-term probabilistic power prediction method and system applied to peak shaving scenarios. Background Technology
[0002] Currently, with the escalating global energy crisis and increasingly prominent environmental issues, electric vehicles (EVs), as a clean and efficient mode of transportation, have been widely promoted by countries around the world. However, the large-scale grid connection of EVs for charging presents significant challenges to the safe and stable operation of power distribution networks due to the highly random, fluctuating, and intermittent nature of the load. Particularly during peak load periods, a large number of disorderly charging activities can lead to transformer overload and voltage exceeding limits.
[0003] On the other hand, electric vehicles, as a distributed energy storage resource, have good adjustability. By implementing demand response (DR), guiding users to charge during off-peak hours and discharge or reduce charging power during peak hours, peak shaving and valley filling can be effectively achieved, assisting the grid in absorbing new energy sources. However, to formulate a scientific and reasonable demand response strategy, it is essential to accurately predict the charging load of electric vehicles.
[0004] Existing charging load forecasting methods are mainly divided into traditional statistical methods and artificial intelligence methods. Traditional methods, such as the ARIMA model, struggle to handle the nonlinear characteristics of charging loads. Artificial intelligence methods, such as Support Vector Machines (SVM) and Neural Networks (NN), while improving prediction accuracy, are mostly limited to deterministic "point predictions," providing only a single, specific predicted value. In practical peak-shaving scenarios, grid dispatchers not only need to know the expected future power but also the range of uncertainty in the forecast results (i.e., the probability distribution) to assess dispatch risks and develop more robust dispatch plans. Summary of the Invention
[0005] The purpose of this invention is to provide a short-term probabilistic power prediction method and system for peak shaving scenarios, aiming to solve at least one of the problems in the background art.
[0006] In a first aspect, the present invention provides a short-term probabilistic power prediction method applied in peak shaving scenarios, the method comprising:
[0007] Historical charging power data of electric vehicle charging stations is obtained, and the historical charging power data is preprocessed. A sample set including time series features and time labels is constructed based on the preprocessed historical charging power data, and the sample set is divided into a training set and a test set.
[0008] A short-term power prediction model based on a long short-term memory network is constructed, and the short-term power prediction model is trained according to the training set. The power of future periods is predicted according to the trained short-term power prediction model to obtain multiple point prediction results.
[0009] The prediction error of the trained short-term power prediction model on the test set is obtained, and the power prediction interval and daily charging power probability distribution under different confidence levels are obtained based on the point prediction results and the prediction error.
[0010] Based on the probability prediction interval and the probability distribution of daily charging power, the potential of electric vehicles to participate in peak shaving or valley filling power regulation at a specified confidence level is evaluated.
[0011] In some embodiments, the step of acquiring historical charging power data of electric vehicle charging stations and preprocessing the historical charging power data includes:
[0012] Collect historical charging power time series data P of electric vehicle charging stations {raw} ={p1, p2, ..., p t}, where p t This represents the historical charging power at time step t;
[0013] Clean the historical charging power time series data;
[0014] For missing data points, linear interpolation is used to complete them:
[0015]
[0016] The power data is mapped to the [0,1] interval using the Min-Max normalization method:
[0017]
[0018] Where, x t p is the normalized power value. min p max These are the minimum and maximum historical charging power values, respectively.
[0019] In some embodiments, the step of constructing a sample set including time-series features and time labels based on preprocessed historical charging power data includes:
[0020] Let the sliding window size be L and the prediction step size be 1;
[0021] For time t, construct the input feature vector X. t and target label Y t :
[0022] X t =[x {t-L} ,x {t-L+1} ,…,x {t-1} ]
[0023] Y t =[H t W t ];
[0024] Among them, H t ∈[0,23] represents the hour corresponding to time t, W t ∈[0,6] represents the weekday corresponding to time t.
[0025] In some embodiments, the step of constructing a short-term power prediction model based on a long short-term memory network includes:
[0026] Construct a short-term power prediction model that includes an input layer, an LSTM layer, a fully connected layer, and an output layer;
[0027] LSTM layers contain forget gates f t Input gate i t and output gate o t Its state update formula is as follows:
[0028] f t =σ(W f ·[h {t-1} ,X t ]+b f )
[0029] i t =σ(W i ·[h {t-1} ,X t ]+b i )
[0030]
[0031] o t =σ(W o ·[h {t-1} ,x t ]+b o )
[0032] h t =o t *tanh(C t );
[0033] Among them, h {t-1} C is the hidden state from the previous moment. t Let represent the cell state, σ be the sigmoid activation function, * denote element-wise multiplication, and W and b be the weight matrix and bias vector, respectively.
[0034] In some embodiments, the step of training the short-term power prediction model based on the training set includes:
[0035] The loss function of the short-term power prediction model is constructed according to the following formula:
[0036]
[0037] Where N is the total number of samples in the training set, L is the loss value, and y i , and are the true value and predicted value of the i-th sample, respectively.
[0038] In some embodiments, the step of obtaining the prediction error of the trained short-term power prediction model on the test set includes:
[0039] Collect the model's predictions on the test set. and the true value y;
[0040] Calculate the predicted residual sequence These are the true value and predicted value of the Mth sample in the test set, respectively.
[0041] The mean μ of the statistically predicted residual series ∈ and standard deviation σ ∈ ;
[0042] Assume the prediction error follows a normal distribution. Construct a probability distribution model.
[0043] In some embodiments, the step of obtaining the power prediction interval and daily charging power probability distribution at different confidence levels based on the point prediction result and the prediction error includes:
[0044] For a given confidence level (1-α), calculate the confidence interval [L] of the predicted value at future time. t U t ]:
[0045]
[0046] in, The upper part of the standard normal distribution Quantiles, L t U t These are the lower and upper limits of the confidence interval, respectively;
[0047] By integrating the predicted power at each time of day, the total daily charging current is obtained, and the total daily charging capacity E is calculated using the error propagation law. {daily} The expected value and variance are used to generate its probability density function.
[0048] In some embodiments, the step of evaluating the potential of an electric vehicle to participate in peak shaving or valley filling power regulation at a specified confidence level based on the probability prediction interval and the daily charging power probability distribution includes:
[0049] Calculate the predicted power for each time period and the historical average power for the same period. The difference ΔP t :
[0050]
[0051] If ΔP t If ΔP > 0, the prediction result at this point is considered to have peak-shaving potential; if ΔP t If the value is ≤0, the predicted value for that point is determined to have no potential for valley filling.
[0052] Plot a power prediction graph that includes prediction curves and confidence intervals for all points, and mark the judgment results at the corresponding time points on the power prediction graph.
[0053] Draw the histogram and fitted curve of the probability density function distribution of daily charging volume.
[0054] Secondly, the present invention provides a short-term probabilistic power prediction system applied in peak shaving scenarios, the system comprising:
[0055] The data acquisition module is used to acquire historical charging power data of electric vehicle charging stations, preprocess the historical charging power data, and construct a sample set including time series features and time labels based on the preprocessed historical charging power data, and divide the sample set into a training set and a test set.
[0056] The model building module is used to build a short-term power prediction model based on a long short-term memory network, train the short-term power prediction model according to the training set, and predict the power of future periods according to the trained short-term power prediction model to obtain multiple point prediction results.
[0057] The confidence interval calculation module is used to obtain the prediction error of the trained short-term power prediction model on the test set, and to obtain the power prediction interval and daily charging power probability distribution under different confidence levels based on the point prediction results and the prediction error.
[0058] The evaluation module is used to evaluate the potential of electric vehicles to participate in peak shaving or valley filling power regulation at a specified confidence level, based on the probability prediction interval and the probability distribution of daily charging power.
[0059] Thirdly, the present invention provides a storage medium that stores one or more programs, which, when executed by a processor, implement the above-described short-term probabilistic power prediction method applied to peak shaving scenarios.
[0060] Fourthly, the present invention provides an electronic device, the electronic device comprising a memory and a processor, wherein:
[0061] The memory is used to store computer programs;
[0062] When the processor executes the computer program stored in the memory, it implements the above-described short-term probabilistic power prediction method applied to peak shaving scenarios.
[0063] Compared with the prior art, the present invention has the following advantages:
[0064] 1. By employing the LSTM deep learning model, it is possible to deeply explore the temporal patterns and nonlinear characteristics in charging load data, and achieve higher prediction accuracy compared to traditional methods.
[0065] 2. The introduction of a probabilistic prediction framework not only provides the predicted point values, but also the confidence interval and probability distribution of the prediction results, quantifying the uncertainty of the prediction and making scheduling decisions more scientific and reliable.
[0066] 3. By generating a probability distribution of daily charging volume, the overall characteristics and fluctuation range of user charging behavior are intuitively displayed, which helps power grid planners to rationally allocate distribution transformer capacity and energy storage equipment.
[0067] 4. By combining predicted values with historical averages in peak shaving scenarios, the algorithm can quickly identify key periods for participating in demand response, thus enhancing its practical value. Attached Figure Description
[0068] Figure 1 This is a flowchart of a short-term probabilistic power prediction method for peak shaving scenarios proposed in an embodiment of the present invention;
[0069] Figure 2 This is a diagram comparing the historical power and predicted power of a charging station.
[0070] Figure 3 A diagram showing the power output of a charging station at different times;
[0071] Figure 4 A probability distribution chart of daily charging volume at charging stations;
[0072] Figure 5 This is a schematic diagram of the structure of a short-term probabilistic power prediction system for peak shaving scenarios proposed in an embodiment of the present invention.
[0073] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation
[0074] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Unless otherwise defined, the technical or scientific terms used herein should have the ordinary meaning understood by those skilled in the art. The terms "comprising" and similar expressions used herein mean that the element or object preceding the word covers the element or object listed after the word and its equivalents, but does not exclude other elements or objects.
[0075] like Figure 1 As shown, an embodiment of the present invention proposes a short-term probabilistic power prediction method for peak shaving scenarios. The method includes steps S101 to S104, wherein:
[0076] Step S101: Obtain historical charging power data of electric vehicle charging stations, preprocess the historical charging power data, and construct a sample set including time series features and time labels based on the preprocessed historical charging power data, and divide the sample set into a training set and a test set.
[0077] It should be noted that the preprocessing process includes: firstly, collecting historical charging power time series data P from electric vehicle charging stations. {raw} ={p1, p2, ..., p t}, where p t This represents the historical charging power at time step t;
[0078] Clean the historical charging power time series data;
[0079] For missing data points, linear interpolation is used to complete them:
[0080]
[0081] The power data is mapped to the [0,1] interval using the Min-Max normalization method:
[0082]
[0083] Where, x t p is the normalized power value. min p maxThese represent the minimum and maximum historical charging power, respectively. Normalization eliminates the influence of different dimensions, accelerating the convergence of the neural network.
[0084] Furthermore, in some embodiments, the sliding window size is set to L, and the prediction step size is 1;
[0085] For time t, construct the input feature vector X. t and target label Y t :
[0086]
[0087] Among them, H t ∈[0,23] represents the hour corresponding to time t, W t ∈[0,6] represents the weekday corresponding to time t.
[0088] Finally, the dataset was divided into training and test sets in an 8:2 ratio.
[0089] Step S102: Construct a short-term power prediction model based on a long short-term memory network, train the short-term power prediction model according to the training set, and predict the power of future periods according to the trained short-term power prediction model to obtain multiple point prediction results.
[0090] It should be noted that in this step, a short-term power prediction module is first constructed, which includes an input layer, an LSTM layer, a fully connected layer, and an output layer. Specifically, this includes:
[0091] Input layer: Receives a data tensor of shape (batch_size, time_steps, features).
[0092] LSTM layer: Contains 50 hidden units, using tanh as the activation function. This layer is responsible for extracting long-term dependencies from the input sequence. Internally, each LSTM unit contains a forget gate, input gate, and output gate, using a gating mechanism to control the retention and updating of information.
[0093] Fully connected layer (Dense): Contains one neuron, used to map the output of the LSTM layer to the final predicted power value.
[0094] The model's optimizer is configured as Adam (learning rate set to 0.001), and the loss function is mean squared error (MSE). The number of training epochs is set to 50, and the batch size is set to 32. During training, the model's performance is monitored using a validation set, and the model parameters with the smallest validation error are saved.
[0095] Furthermore, in some embodiments, the LSTM layer includes a forget gate f tInput gate i t and output gate o t Its state update formula is as follows:
[0096] f t =σ(W f ·[h {t-1} ,X t ]+b f )
[0097] i t =σ(W i ·[h {t-1} ,X t ]+b i )
[0098]
[0099] o t =σ(W o ·[h {t-1} ,x t ]+b o )
[0100] h t =o t *tanh(C t )
[0101] Among them, h {t-1} C is the hidden state from the previous moment. t Let represent the cell state, σ be the sigmoid activation function, * denote element-wise multiplication, and W and b be the weight matrix and bias vector, respectively.
[0102] Furthermore, in some embodiments, the loss function of the short-term power prediction model is constructed according to the following formula:
[0103]
[0104] Where N is the total number of samples in the training set, L is the loss value, and y i , and are the true value and predicted value of the i-th sample, respectively.
[0105] Step S103: Obtain the prediction error of the trained short-term power prediction model on the test set, and obtain the power prediction interval and daily charging power probability distribution under different confidence levels based on the point prediction results and the prediction error.
[0106] It should be noted that the best-trained model is used to predict the charging power for the next week (7 days). A multi-step rolling prediction strategy is adopted: first, the power for the first time point is predicted using the most recent historical window data; then, this predicted value is used as a replacement for the actual value to update the input window, and the prediction for the second time point is continued; and so on, until the prediction for all time steps is completed. Finally, the prediction result is restored to the actual power value using an inverse normalization formula.
[0107] Specifically, the model's predictions on the test set are collected first. And the true value y; calculate the predicted residual sequence Let μ be the true value and predicted value of the Mth sample in the test set, respectively; and let μ be the mean of the predicted residual sequence. ∈ and standard deviation σ ∈ Histogram and QQ plot tests were used to verify that the residuals approximately follow a normal distribution. Therefore, it is assumed that the prediction error follows a normal distribution. Construct a probability distribution model.
[0108] Furthermore, for a given confidence level (1-α), the confidence interval [L] of the predicted value at future time points is calculated. t U t ]:
[0109]
[0110] in, The upper part of the standard normal distribution Quantiles, L t U t These are the lower and upper limits of the confidence interval, respectively;
[0111] By integrating the predicted power at each time of day, the total daily charging current is obtained, and the total daily charging capacity E is calculated using the error propagation law. {daily} The expected value and variance are used to generate its probability density function.
[0112] like Figure 2 and Figure 3 As shown in the figure, the historical actual power curve (blue) and the future predicted power curve (red) are displayed, with the shaded area representing the 95% confidence interval. By comparing the predicted curve with the historical average curve, it can be found that during certain peak periods (such as 18:00-21:00), the predicted power is significantly higher than the historical average, indicating that there is a large peak-shaving potential during this period, and the power grid can issue high electricity prices or incentive signals to guide users to reduce charging. During off-peak periods (such as 02:00-05:00), the predicted power is lower, which can guide users to increase charging to absorb excess wind power and other renewable energy sources.
[0113] Furthermore, the probability distribution of the total daily charging amount is calculated. Assuming the prediction errors at each time point are independent, the variance of the daily charging amount error is the sum of the variances at each time point. Based on this, the following is generated: Figure 4 The daily charging power probability density curve shown reflects the fluctuation of a user's daily charging demand.
[0114] Step S104: Based on the probability prediction interval and the daily charging power probability distribution, evaluate the power regulation potential of electric vehicles to participate in peak shaving or valley filling at a specified confidence level.
[0115] In this step, the predicted power for each time period and the historical average power for the same period are first calculated. The difference ΔP t :
[0116]
[0117] If ΔP t If ΔP > 0, the prediction result at this point is considered to have peak-shaving potential; if ΔP t If the value is ≤0, the predicted value for that point is determined to have no potential for valley filling.
[0118] Plot a power prediction graph that includes prediction curves and confidence intervals for all points, and mark the judgment results at the corresponding time points on the power prediction graph.
[0119] Draw the histogram and fitted curve of the probability density function distribution of daily charging volume.
[0120] In summary, introducing the difference analysis between predicted power and historical average power, along with related visualizations, is a crucial step in ensuring that scheduling decisions have clear data support and practical operability. This is achieved by calculating ΔP. t This system can quantify the degree of deviation of electric vehicle charging load from normal levels at any given moment, thereby scientifically identifying periods with peak-shaving potential and providing a direct basis for formulating differentiated charging incentive or control strategies. Simultaneously, plotting power prediction maps and power probability distribution maps not only helps to intuitively display the spatiotemporal characteristics and uncertainty range of load fluctuations, but also provides operators with important visual analysis tools for assessing adjustment potential and formulating reliable scheduling plans, effectively enhancing the interpretability and decision reliability of the scheduling system.
[0121] In summary, the short-term probabilistic power prediction method applied to peak-shaving scenarios described above has the following advantages:
[0122] 1. By employing the LSTM deep learning model, it is possible to deeply explore the temporal patterns and nonlinear characteristics in charging load data, and achieve higher prediction accuracy compared to traditional methods.
[0123] 2. The introduction of a probabilistic prediction framework not only provides the predicted point values, but also the confidence interval and probability distribution of the prediction results, quantifying the uncertainty of the prediction and making scheduling decisions more scientific and reliable.
[0124] 3. By generating a probability distribution of daily charging volume, the overall characteristics and fluctuation range of user charging behavior are intuitively displayed, which helps power grid planners to rationally allocate distribution transformer capacity and energy storage equipment.
[0125] 4. By combining predicted values with historical averages in peak shaving scenarios, the algorithm can quickly identify key periods for participating in demand response, thus enhancing its practical value.
[0126] like Figure 5 As shown, one embodiment of the present invention proposes a short-term probabilistic power prediction system for peak shaving scenarios, the system comprising:
[0127] The data acquisition module 10 is used to acquire historical charging power data of electric vehicle charging stations, preprocess the historical charging power data, construct a sample set including time series features and time labels based on the preprocessed historical charging power data, and divide the sample set into a training set and a test set.
[0128] The model building module 20 is used to build a short-term power prediction model based on a long short-term memory network, train the short-term power prediction model according to the training set, and predict the power of future periods according to the trained short-term power prediction model to obtain multiple point prediction results.
[0129] The confidence interval calculation module 30 is used to obtain the prediction error of the trained short-term power prediction model on the test set, and to obtain the power prediction interval and daily charging power probability distribution under different confidence levels based on the point prediction results and the prediction error.
[0130] Evaluation module 40 is used to evaluate the potential of electric vehicles to participate in peak shaving or valley filling at a specified confidence level based on the probability prediction interval and the probability distribution of daily charging power.
[0131] In another aspect, the present invention also proposes a storage medium on which one or more programs are stored, which, when executed by a processor, implement the above-described short-term probabilistic power prediction method for peak shaving scenarios.
[0132] In another aspect, the present invention also proposes an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to realize the above-mentioned short-term probabilistic power prediction method applied to peak shaving scenarios.
[0133] Those skilled in the art will understand that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can mean any means that can contain stored, communicated, propagated, or transmitted programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.
[0134] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.
[0135] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.
[0136] While embodiments of the present invention have been described in detail above, it will be apparent to those skilled in the art that various modifications and variations can be made to these embodiments. However, it should be understood that such modifications and variations fall within the scope and spirit of the invention as set forth in the claims. Furthermore, the invention described herein may have other embodiments and can be implemented or carried out in various ways.
Claims
1. A short-term probabilistic power prediction method applied to peak shaving scenarios, characterized in that, The method includes: Historical charging power data of electric vehicle charging stations is obtained, and the historical charging power data is preprocessed. A sample set including time series features and time labels is constructed based on the preprocessed historical charging power data, and the sample set is divided into a training set and a test set. A short-term power prediction model based on a long short-term memory network is constructed, and the short-term power prediction model is trained according to the training set. The power of future periods is predicted according to the trained short-term power prediction model to obtain multiple point prediction results. The prediction error of the trained short-term power prediction model on the test set is obtained, and the power prediction interval and daily charging power probability distribution under different confidence levels are obtained based on the point prediction results and the prediction error. Based on the probability prediction interval and the probability distribution of daily charging power, the potential of electric vehicles to participate in peak shaving or valley filling power regulation at a specified confidence level is evaluated.
2. The short-term probabilistic power prediction method for peak shaving scenarios according to claim 1, characterized in that, The steps of acquiring historical charging power data of electric vehicle charging stations and preprocessing the historical charging power data include: Collect historical charging power time series data P of electric vehicle charging stations {raw} ={p1,p2,...,p t }, where p t This represents the historical charging power at time step t; Clean the historical charging power time series data; For missing data points, linear interpolation is used to complete them: The power data is mapped to the [0,1] interval using the Min-Max normalization method: Where, x t p is the normalized power value. min p max These are the minimum and maximum historical charging power values, respectively.
3. The short-term probabilistic power prediction method for peak shaving scenarios according to claim 2, characterized in that, The step of constructing a sample set including time-series features and time labels based on the preprocessed historical charging power data includes: Let the sliding window size be L and the prediction step size be 1; For time t, construct the input feature vector X. t and target label Y t : X t =[x {t-L} ,x {t-L+1} ,...,,x {t-1} ] Y t =[H t ,W t ]; Among them, H t ∈[0,23] represents the hour corresponding to time t, W t ∈[0,6] represents the weekday corresponding to time t.
4. The short-term probabilistic power prediction method for peak shaving scenarios according to claim 3, characterized in that, The steps for constructing a short-term power prediction model based on a long short-term memory network include: Construct a short-term power prediction model that includes an input layer, an LSTM layer, a fully connected layer, and an output layer; LSTM layers contain forget gates f t Input gate i t and output gate o t Its state update formula is as follows: f t =σ(W f ·[h {t-1} ,X t ]+b f ) i t =σ(W i ·[h {t-1} ,X t ]+b i ) the t =σ(W o ·[h {t-1} ,x t ]+b o ) h t = no t *fish(C) t ); Among them, h {t-1} C is the hidden state from the previous moment. t Let represent the cell state, σ be the sigmoid activation function, * denote element-wise multiplication, and W and b be the weight matrix and bias vector, respectively.
5. The short-term probabilistic power prediction method for peak shaving scenarios according to claim 4, characterized in that, The step of training the short-term power prediction model based on the training set includes: The loss function of the short-term power prediction model is constructed according to the following formula: Where N is the total number of samples in the training set, L is the loss value, and y i , and are the true value and predicted value of the i-th sample, respectively.
6. The short-term probabilistic power prediction method for peak shaving scenarios according to claim 1, characterized in that, The step of obtaining the prediction error of the trained short-term power prediction model on the test set includes: Collect the model's predictions on the test set. and the true value y; Calculate the predicted residual sequence y M , These are the true value and predicted value of the Mth sample in the test set, respectively. The mean μ of the statistically predicted residual series ∈ and standard deviation σ ∈ ; Assume the prediction error follows a normal distribution. Construct a probability distribution model.
7. The short-term probabilistic power prediction method for peak shaving scenarios according to claim 6, characterized in that, The step of obtaining the power prediction interval and daily charging power probability distribution under different confidence levels based on the point prediction result and the prediction error includes: For a given confidence level (1-α), calculate the confidence interval [L] of the predicted value at future time. t U t ]: in, The upper part of the standard normal distribution Quantiles, L t U t These are the lower and upper limits of the confidence interval, respectively; By integrating the predicted power at each time of day, the total daily charging current is obtained, and the total daily charging capacity E is calculated using the error propagation law. {daily} The expected value and variance are used to generate its probability density function.
8. The short-term probabilistic power prediction method for peak shaving scenarios according to claim 6, characterized in that, The step of evaluating the potential of electric vehicles to participate in peak shaving or valley filling power regulation at a specified confidence level based on the probability prediction interval and the daily charging power probability distribution includes: Calculate the predicted power for each time period and the historical average power for the same period. The difference ΔP t : If ΔP t If ΔP > 0, the prediction result at this point is determined to have peak-shaving potential; if ΔP t If the value is ≤0, the predicted value for that point is determined to have no potential for valley filling. Plot a power prediction graph that includes prediction curves and confidence intervals for all points, and mark the judgment results at the corresponding time points on the power prediction graph. Draw the histogram and fitted curve of the probability density function distribution of daily charging volume.
9. A short-term probabilistic power prediction system applied in peak shaving scenarios, characterized in that, The system includes: The data acquisition module is used to acquire historical charging power data of electric vehicle charging stations, preprocess the historical charging power data, and construct a sample set including time series features and time labels based on the preprocessed historical charging power data, and divide the sample set into a training set and a test set. The model building module is used to build a short-term power prediction model based on a long short-term memory network, train the short-term power prediction model according to the training set, and predict the power of future periods according to the trained short-term power prediction model to obtain multiple point prediction results. The confidence interval calculation module is used to obtain the prediction error of the trained short-term power prediction model on the test set, and to obtain the power prediction interval and daily charging power probability distribution under different confidence levels based on the point prediction results and the prediction error. The evaluation module is used to evaluate the potential of electric vehicles to participate in peak shaving or valley filling power regulation at a specified confidence level, based on the probability prediction interval and the probability distribution of daily charging power.
10. A storage medium, characterized in that, The storage medium stores one or more programs that, when executed by a processor, implement the short-term probabilistic power prediction method for peak shaving scenarios as described in any one of claims 1-8.