Power prediction data enhancement method and power prediction method for electric vehicle charging station
The automatic data enhancement framework generates high-quality virtual samples, which solves the problems of low data and low quality of electric vehicle charging stations, and realizes accurate prediction of charging power of charging stations.
Patent Information
- Application Number
- CN202311302105.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-10-09
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2043-10-09
AI Technical Summary
The small size and low quality of the electric vehicle charging station data sample size make it difficult to accurately predict charging power using big data models. The existing data enhancement methods cannot effectively generate virtual samples that conform to the actual charging behavior.
Using an automatic data augmentation framework, through uniformly distributed sampling, centralized sampling from preset time series augmentation functions is generated, such as jitter, scaling, time distortion, amplitude perturbation and phase perturbation, to form a virtual sample to ensure the virtual sample High quality set and suitable for training big data model.
A large number of high-quality virtual samples were generated, the sample size was effectively expanded, the prediction accuracy of the big data model was improved, the problems of low data and low quality were solved, and the charging power of electric vehicle charging stations was achieved.
Smart Images

Figure CN117493878B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of data analysis of electric vehicle charging stations, and more specifically, relates to a power prediction data enhancement method and a power prediction method of an electric vehicle charging station. Background Art
[0002] Reducing carbon emissions is one of the most basic actions to address climate change, and electrification of the transportation system is another effective way to reduce carbon emissions. With the development of manufacturing technology and related key technologies, the reliability and efficiency of electric vehicles have been significantly improved. Therefore, all countries hope to achieve a high penetration rate of electric vehicles in the near future. It is predicted that the number of electric vehicles will continue to grow, and electric vehicle charging stations also have an important mission to reduce emissions.
[0003] Accurately predicting the charging power of electric vehicle charging stations is of great significance in supporting the operation planning of electric vehicles. Since the charging behavior of electric vehicles is highly uncertain and the power characteristics of different charging stations are different, it is difficult to predict the power of electric vehicle charging stations. In recent years, big data methods, such as deep learning methods, have achieved remarkable performance in the field of energy forecasting. In addition to good performance, another advantage of these data-driven methods is that they can learn customized model parameters based on the historical data of each charging station.
[0004] In order to effectively learn model parameters, big data models rely on a large number of high-quality data samples. However, in underdeveloped charging stations or charging stations facing monitoring equipment failures, data transmission problems, etc., charging data may not be stored or transmitted, which makes the amount of data samples available for model learning small. In response to the problem of small data samples, existing methods often use data enhancement to expand the data set. However, these methods often use a single, fixed data enhancement method to generate virtual samples, while electric vehicle charging behavior is uncertain and the power characteristics of charging stations are also different, which makes the generated virtual samples not effectively applied to the training of big data models. Some scholars have also proposed using complex networks such as TimeGAN for data enhancement. However, in order to ensure the quality of the generated virtual samples, it is also necessary to first use a certain amount of real samples to train these data enhancement networks, which is difficult to achieve when the amount of charging data at the charging station itself is small. In addition, existing charging stations often allow users to input charging data by themselves, which may not be consistent with the actual charging behavior, resulting in poor quality of the collected charging data. In general, how to effectively solve the problem of small and low-quality data at electric vehicle charging stations is a very important issue for accurately predicting the charging power of electric vehicle charging stations using big data models. Summary of the invention
[0005] In view of the defects of the prior art and the need for improvement, the present invention provides a power prediction data enhancement method and a power prediction method for an electric vehicle charging station, aiming to solve the problem that the data of electric vehicle charging stations is small and of low quality, and the charging power of electric vehicle charging stations cannot be accurately predicted using big data models.
[0006] To achieve the above object, according to one aspect of the present invention, a method for enhancing power prediction data of an electric vehicle charging station is provided, comprising:
[0007] (S1) Collect the total charging power of the charging station at each time of the day as a real sample to obtain a real sample set;
[0008] (S2) For each real sample, V virtual samples are generated to obtain a virtual sample set, thus completing data enhancement;
[0009] For any real sample, the methods of generating a virtual sample include:
[0010] (S21) sampling S time series enhancement functions from a preset time series enhancement function set according to uniform distribution sampling;
[0011] (S22) sequentially executing the S time series enhancement functions obtained by sampling on the real samples to obtain virtual samples;
[0012] Among them, V and S are both positive integers, and S is greater than 1 and less than the size of the time series enhancement function set.
[0013] Further, step (S1) comprises:
[0014] (S11) collecting charging data of each charging pile in the charging station; the charging data includes power-on time, power-off time and charging energy;
[0015] (S12) For each charging data, the average charging power in the charging time period is calculated, and the sum of the average charging powers of all charging piles at the same time is calculated to obtain the total charging power of the charging station at that time. The total charging power of the charging station at each time constitutes the load time series of the charging station;
[0016] (S13) Divide the load time series by day to obtain the total charging power of the charging station at each time of the day, and use it as a real sample to obtain a real sample set.
[0017] Furthermore, each time series enhancement function in the time series enhancement function set has the same enhancement strength M.
[0018] Furthermore, the time series enhancement function set includes: jitter, scaling, time distortion, amplitude distortion, amplitude disturbance and phase disturbance, and the specific operation of each time series enhancement function is as follows:
[0019] Jitter: Add zero-mean Gaussian noise to the time series;
[0020] Scaling: changing the size of a certain segment of data within a sample by multiplying it by a random scalar;
[0021] Time Warp: Smoothly warp the time intervals between samples;
[0022] Amplitude distortion: convolve a certain section of data in the sample with a smooth curve with a mean of 1;
[0023] Amplitude perturbation: Add zero-mean Gaussian noise to the amplitude of the sample frequency domain;
[0024] Phase perturbation: Add zero-mean Gaussian noise to the phase values of the samples in the frequency domain.
[0025] Furthermore, S, V and M are determined as follows:
[0026] Taking S, V and M as the variables to be optimized and maximizing the accuracy improvement of the virtual sample set on the actual power prediction task as the objective function, an optimization problem is established;
[0027] The optimization problem is solved using a preset parameter optimization algorithm to obtain the values of S, V and M.
[0028] According to another aspect of the present invention, a method for establishing a power prediction model for an electric vehicle charging station is provided, comprising:
[0029] Establish a time series prediction model to predict the total charging power at each time of each day in the future time period based on the historical data of the total charging power at each time of each day;
[0030] The time series prediction model is trained using a virtual sample set established by the power prediction data enhancement method for an electric vehicle charging station provided by the present invention, and the parameters of the trained time series model are fine-tuned using a real sample set established by the power prediction data enhancement method for an electric vehicle charging station provided by the present invention to obtain a power prediction model for an electric vehicle charging station.
[0031] According to another aspect of the present invention, a power prediction method for an electric vehicle charging station is provided, comprising:
[0032] The historical data of the total charging power at each time of the day are collected and input into the power prediction model established by the power prediction model establishment method of the electric vehicle charging station provided by the present invention to obtain the prediction results of the total charging power at each time of the day in the future time period.
[0033] According to another aspect of the present invention, a power prediction data enhancement system for an electric vehicle charging station is provided, comprising:
[0034] A computer-readable storage medium for storing a computer program;
[0035] And a processor, which is used to read the computer program stored in the computer-readable storage medium and execute the power prediction data enhancement method for the electric vehicle charging station provided by the present invention.
[0036] According to another aspect of the present invention, a power prediction model establishment system for an electric vehicle charging station is provided, comprising:
[0037] A computer-readable storage medium for storing a computer program;
[0038] And a processor, which is used to read the computer program stored in the computer-readable storage medium and execute the power prediction model establishment method for the electric vehicle charging station provided by the present invention.
[0039] According to another aspect of the present invention, there is provided a power prediction system for an electric vehicle charging station, comprising:
[0040] A computer-readable storage medium for storing a computer program;
[0041] And a processor, which is used to read the computer program stored in the computer-readable storage medium and execute the power prediction method for the electric vehicle charging station provided by the present invention.
[0042] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects:
[0043] (1) The present invention uses the total charging power of the electric vehicle charging station at each time of each day as a sample. The sample is a pure digital time series and can be effectively applied to data analysis tasks. At the same time, the charging load of the electric vehicle charging station has obvious daily periodicity. The sample determined by the present invention can accurately reflect the daily charging load characteristics of the electric vehicle charging station. Therefore, the charging data collected by the present invention is of high quality, which is conducive to the big data model to complete the data analysis task; on this basis, the present invention proposes an automatic data enhancement framework that can integrate multiple time series enhancement functions. Under this data enhancement framework, each time data enhancement is performed, multiple time series enhancement functions will be selected according to probability, thereby fully considering the time correlation of the charging data, effectively generating richer and more complex enhancement functions, and generating more and richer virtual samples while maintaining the time dynamics of the original data. While expanding the sample size, the quality of the sample is effectively guaranteed. In general, the present invention can generate a large number of high-quality virtual samples, effectively solving the problem that the electric vehicle charging station data is small and of low quality, and the charging power of the electric vehicle charging station cannot be accurately predicted using the big data model.
[0044] (2) The present invention records the power-on time, power-off time and charging energy of each charging pile each time, and calculates the average charging power of each charging pile at each time based on the recorded charging data, thereby obtaining the average charging power of the entire charging station at each time, and then divides it to obtain the average charging power at each time of each day. In this way, real samples are obtained, and the character and digital mixed data format of the charging session type is effectively converted into a pure digital time series, which can be effectively applied to data analysis tasks.
[0045] (3) In the preferred embodiment of the present invention, the time series enhancement function set specifically includes six enhancement functions: jitter, scaling, time distortion, amplitude distortion, amplitude disturbance and phase disturbance. Experiments show that the new enhancement function obtained by combining these six enhancement functions can effectively capture the time correlation of electric vehicle charging data, and the virtual samples generated by acting on real samples are consistent with the actual charging behavior of electric vehicle charging stations.
[0046] (4) The present invention takes the number S of enhancement functions selected for each data enhancement, the number V of virtual samples generated by each real sample, and the enhancement strength M of each enhancement function as variables to be optimized, and takes maximizing the accuracy improvement of the virtual sample set for the actual power prediction task as the objective function, establishes an optimization problem, and determines the values of S, V and M by solving the optimization problem, thereby ensuring that when the established virtual sample set is used for training a big data model, the prediction accuracy of the trained model can be maximized.
[0047] (5) Based on the virtual sample set obtained through data enhancement, the present invention adopts a two-stage establishment method when establishing a model for electric vehicle charging station power prediction: in the first stage, virtual samples with a large amount of data are used to train the model, so that the model parameters can reach a better state; in the second stage, real samples with a small amount of data are used to fine-tune the model parameters, so that the model can learn the actual charging power characteristics and optimize the model prediction performance. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] Figure 1 A schematic diagram of a method for enhancing power prediction data of an electric vehicle charging station provided by an embodiment of the present invention;
[0049] Figure 2 A flow chart of a method for establishing a power prediction model for an electric vehicle charging station provided in an embodiment of the present invention;
[0050] Figure 3 A schematic diagram showing a comparison of the effects of the prediction model on a public data set before and after the electric vehicle charging station data enhancement method provided by an embodiment of the present invention;
[0051] Figure 4A schematic diagram showing a comparison of the convergence levels of the prediction model on a validation data set before and after the electric vehicle charging station data enhancement method provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0052] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.
[0053] In the present invention, the terms "first", "second", etc. (if any) in the present invention and the accompanying drawings are used to distinguish similar objects but not necessarily to describe a specific order or sequence.
[0054] In order to effectively solve the problem that electric vehicle charging station data is small and of low quality, and the charging power of electric vehicle charging stations cannot be accurately predicted using big data models, the present invention provides a power prediction data enhancement method and a power prediction method for electric vehicle charging stations. The overall idea is: based on the uncertainty of electric vehicle charging behavior and the differences in power characteristics of charging stations, a new automatic data enhancement framework is proposed, which integrates a variety of time series enhancement functions that can capture the time correlation of charging station charging data, generates richer and more complex enhancement functions, and generates more and richer virtual samples while maintaining the time dynamics of the original data. While expanding the sample size, the quality of the samples is effectively guaranteed.
[0055] The following are examples.
[0056] Embodiment 1:
[0057] A power prediction data enhancement method for electric vehicle charging stations, such as Figure 1 As shown, it includes: the following steps (S1) to (S2).
[0058] In this embodiment, step (S1) specifically includes: collecting the total charging power of the charging station at each time of the day as a real sample to obtain a real sample set.
[0059] Considering that the traditional charging session data is in a mixed character and digital data format, which is inconvenient to process, this embodiment provides an effective data conversion method to convert the mixed character and digital data format of the charging session into a pure digital time series. Specifically, Figure 1 As shown, step (S1) of this embodiment includes:
[0060] (S11) collecting charging data of each charging pile in the charging station; the charging data includes power-on time, power-off time and charging energy;
[0061] In the traditional recording method, the user fills in the charging plan by himself, but in actual situations, the user may leave before the scheduled power-off time is reached, which makes the collected data inconsistent with the actual charging behavior. This embodiment records the power-on time, power-off time and charging energy of the actual charging behavior, which can effectively avoid this situation;
[0062] (S12) For each charging data, the average charging power in the charging time period is calculated, and the sum of the average charging powers of all charging piles at the same time is calculated to obtain the total charging power of the charging station at that time. The total charging power of the charging station at each time constitutes the load time series of the charging station;
[0063] In the actual charging process, the charging power of a single charging pile does not change much during a single charging process. This embodiment directly calculates the average charging power at each moment in the charging time period according to the calculated power-on time, power-off time and charging energy, and uses it as the actual charging power at each moment. This can accurately obtain the charging load of the charging station at each moment without installing a real-time power sensor device.
[0064] (S13) Dividing the load time series by day, obtaining the total charging power of the charging station at each time of the day, and using it as a real sample to obtain a real sample set;
[0065] The charging load of an electric vehicle charging station has an obvious daily periodicity. This embodiment divides the samples into days, which can accurately reflect the daily charging load characteristics of the electric vehicle charging station. The real sample finally obtained is a pure digital time series, which can be effectively applied to data analysis tasks.
[0066] Based on the real sample set obtained in step (S1), Figure 1 As shown, step (S2) of this embodiment specifically includes: for each real sample, generating V virtual samples, obtaining a virtual sample set, and completing data enhancement; wherein V is a positive integer.
[0067] For any real sample, the methods of generating a virtual sample include:
[0068] (S21) sampling S time series enhancement functions from a preset time series enhancement function set according to uniform distribution sampling;
[0069] Where S is a positive integer, S is greater than 1 and less than the size of the time series enhancement function set;
[0070] In order to effectively capture the time correlation of electric vehicle charging data, as a preferred implementation, in this embodiment, the time series enhancement function set includes: jitter, scaling, time distortion, amplitude distortion, amplitude disturbance and phase disturbance, and the specific operations of each time series enhancement function are as follows:
[0071] Jitter: Add zero-mean Gaussian noise to the time series;
[0072] Scaling: changing the size of a certain segment of data within a sample by multiplying it by a random scalar;
[0073] Time Warp: Smoothly warp the time intervals between samples;
[0074] Amplitude distortion: convolve a certain section of data in the sample with a smooth curve with a mean of 1;
[0075] Amplitude perturbation: Add zero-mean Gaussian noise to the amplitude of the sample frequency domain;
[0076] Phase perturbation: Add zero-mean Gaussian noise to the phase value of the sample frequency domain;
[0077] Each enhancement function is used for controlling the enhancement strength of hyperparameters with different meanings. In order to facilitate the determination of the preferred enhancement strength of each enhancement function, as a preferred implementation, in this embodiment, each time series enhancement function has the same enhancement strength M, and its value range is an integer within 10;
[0078] (S22) S time series enhancement functions obtained by sampling are sequentially executed on the real samples to obtain virtual samples.
[0079] Through the above method, this embodiment provides a new automatic data enhancement framework. Under this data enhancement framework, each time data enhancement is performed, multiple time series enhancement functions are selected according to probability, thereby fully considering the time correlation of charging data, effectively generating richer and more complex enhancement functions, and generating more and richer virtual samples while maintaining the time dynamics of the original data. While expanding the sample size, the quality of the sample is effectively guaranteed.
[0080] In the automatic data enhancement framework used in this embodiment, there are only three parameters, S, V and M. Parameter V determines how many virtual samples will be generated from a real sample; parameter S determines how many times a virtual sample is generated after each real sample is enhanced; parameter M determines the global strength of the enhancement transformation. Compared with the method of using complex networks such as TimeGAN for data enhancement, in this embodiment, data enhancement is simple to execute, has high computational efficiency, has a small number of parameters, and is easy to optimize.
[0081] The specific values of S, V and M can be set based on experience. However, in order to further improve the positive effect of the virtual sample set on the actual power prediction task, as a preferred implementation, in this embodiment, S, V and M are determined as follows:
[0082] Taking S, V and M as the variables to be optimized and maximizing the accuracy improvement of the virtual sample set on the actual power prediction task as the objective function, an optimization problem is established;
[0083] The optimization problem is solved using a preset parameter optimization algorithm to obtain the values of S, V and M.
[0084] Through the above method, the values of S, V and M determined in this embodiment can ensure that when the established virtual sample set is used for training a big data model, the prediction accuracy of the trained model can be maximized.
[0085] In general, this embodiment uses charging data to calculate the total charging power of the charging station at each time of each day as a real sample, and proposes a new automatic data enhancement framework, which can effectively generate richer and more complex enhancement functions, while maintaining the temporal dynamics of the original data. More and richer virtual samples are generated, while expanding the sample size. At the same time, the quality of the sample is effectively guaranteed, and the problem of insufficient and low-quality data on electric vehicle charging stations and the inability to use big data models to accurately predict the charging power of electric vehicle charging stations is effectively solved.
[0086] Embodiment 2:
[0087] A method for establishing a power prediction model for an electric vehicle charging station, such as Figure 2 As shown, including:
[0088] Establish a time series prediction model to predict the total charging power at each time of each day in the future time period based on the historical data of the total charging power at each time of each day;
[0089] The time series prediction model is trained using a virtual sample set established by the power prediction data enhancement method for an electric vehicle charging station provided in the above embodiment 1, and the parameters of the trained time series model are fine-tuned using a real sample set established by the power prediction data enhancement method for an electric vehicle charging station provided in the above embodiment 1 to obtain a power prediction model for an electric vehicle charging station.
[0090] As an optional implementation, in this embodiment, the selected time series prediction model is a long short-term memory network (LSTM). The model architecture is a fully connected neural network with 3 layers, 20 neurons in each layer, and the outermost cascade output dimension is 1. The update algorithm of the long short-term memory network uses the back propagation algorithm over time to update the parameters of the deep recurrent neural network by calculating the full gradient, specifically:
[0091]
[0092] Among them, E is the loss of the data batch calculation; d represents the calculation of the full gradient; η represents the learning rate. The larger the learning rate, the larger the amplitude of the single-step parameter update; θ l Represents the parameters of the lth layer of the deep neural network.
[0093] It should be noted that the above network and parameter updating method is only an optional implementation method and should not be understood as the only limitation of the present invention. In some other embodiments of the present invention, other time series prediction models such as ARIMA can also be used.
[0094] The virtual sample set established through data enhancement cannot completely replace the real sample. This embodiment, based on the virtual sample set obtained through data enhancement, adopts a two-stage establishment method when establishing a model for power prediction of electric vehicle charging stations: in the first stage, the model is trained using virtual samples with a large amount of data, so that the model parameters can reach a better state; in the second stage, the model parameters are fine-tuned using real samples with a small amount of data, so that the model can learn the actual charging power characteristics and optimize the model prediction performance.
[0095] Embodiment 3:
[0096] A power prediction method for an electric vehicle charging station, comprising:
[0097] The historical data of the total charging power at each time of each day is collected and input into the power prediction model established by the power prediction model establishment method of the electric vehicle charging station provided in the above embodiment 2 to obtain the prediction results of the total charging power at each time of each day in the future time period.
[0098] Embodiment 4:
[0099] A power prediction data enhancement system for an electric vehicle charging station, comprising:
[0100] A computer-readable storage medium for storing a computer program;
[0101] And a processor, which is used to read the computer program stored in the computer-readable storage medium and execute the power prediction data enhancement method for the electric vehicle charging station provided in the above embodiment 1.
[0102] Embodiment 5:
[0103] A power prediction model establishment system for an electric vehicle charging station, comprising:
[0104] A computer-readable storage medium for storing a computer program;
[0105] And a processor, which is used to read the computer program stored in the computer-readable storage medium and execute the method for establishing the power prediction model of the electric vehicle charging station provided in the above embodiment 2.
[0106] Embodiment 6:
[0107] A power prediction system for an electric vehicle charging station, comprising:
[0108] A computer-readable storage medium for storing a computer program;
[0109] And a processor, which is used to read the computer program stored in the computer-readable storage medium and execute the power prediction method for the electric vehicle charging station provided in the above embodiment 3.
[0110] The beneficial effects that can be achieved by the present invention are verified and explained below in conjunction with specific electric vehicle charging station power prediction results.
[0111] First, two public datasets of adaptive charging networks in California were collected. The adaptive charging network on the Caltech campus is located in a parking lot and has 54 electric vehicle charging facilities. The charging facilities at Caltech are open to the public and are often used by non-Caltech drivers. Since the parking lot is close to the campus gym, many drivers charge their electric vehicles when they exercise in the morning or evening. The charging facilities at the Jet Propulsion Laboratory include 52 electric vehicle charging stations in a parking lot. Unlike Caltech, access to the Jet Propulsion Laboratory is restricted and only employees can use the charging system. The Jet Propulsion Laboratory is a representative of workplace charging, while Caltech is a hybrid of workplace and public charging. The penetration rate of electric vehicles at the Jet Propulsion Laboratory is also quite high. This leads to high utilization of electric vehicle charging stations and temporary plans by drivers to move their electric vehicles to free up plugs for other drivers after charging.
[0112] In order to test the effect of virtual samples, the charging sessions at the Caltech campus and the Jet Propulsion Laboratory were divided into four seasonal data sets. Finally, the data sets were divided into training data, validation data, and test data in a ratio of 6:2:2 for model validation.
[0113] In order to verify the effect of the present invention, a big data prediction model is constructed. For each season, two models are established, namely, model 1 and model 2, where:
[0114] Model 1 is trained using the original training data;
[0115] Model 2 is established by using the method for establishing a power prediction model for an electric vehicle charging station provided in the above-mentioned embodiment 2;
[0116] The prediction effects of both Model 1 and Model 2 were verified on the test set, and the performance of the two models was compared. The root mean square error index was selected to quantify the prediction performance. Among them, the big data prediction model is a long short-term memory deep neural network model.
[0117] The results of using the test set to verify Model 1 and Model 2 are as follows Figure 3 As shown, according to Figure 3 The results show that without using the data enhancement method proposed by the present invention, the accuracy of the power prediction model obtained by the final training is significantly reduced. More specifically, without using the data enhancement method provided by the present invention, the root mean square error of the four seasons on the California Institute of Technology dataset increases by 27.53%, 26.64%, 15.81% and 16.20% respectively. On the spray paint propulsion laboratory dataset, the root mean square error increases more significantly, specifically, the root mean square error in the four seasons increases by 48.93%, 36.35%, 29.88% and 32.41% respectively. Figure 3 The results shown indicate that, under the premise of considering only the original training data, the model can easily overfit to small-scale training samples, which may weaken the capabilities of the deep neural network and even have a negative effect. The data enhancement method provided by the present invention plays a very important role in improving the prediction performance.
[0118] In order to further illustrate the effect of the electric vehicle charging station data enhancement method proposed in the present invention, the training process of the spring case of the Caltech dataset is visualized. Figure 4 The training and validation loss curves of the two models are shown. The training and validation losses are the root mean square error of the model's prediction on the training dataset (for model 2, the training loss includes the loss on the virtual samples) and the validation dataset. The decrease in training loss indicates that the model parameters are continuously updated in the direction of reducing the model's error on the training samples during the training process, while the validation error reflects the model's performance on unknown data, further reflecting the model's generalization ability. Figure 4 It can be seen that the validation loss of model 1 is always higher than the training loss, which indicates that when the model is trained with only limited real-world data, it will overfit the training samples. After integrating the data enhancement method of the present invention, the performance of model 2 on the validation set is significantly improved, and the validation loss curve converges to a significantly lower value. Therefore, the electric vehicle charging station data enhancement method proposed in the present invention can help the model achieve better performance in big data applications.
[0119] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for establishing a power prediction model for an electric vehicle charging station. It is characterized in that include: Establish a time series prediction model to predict the total charging power at each time of each day in the future time period based on the historical data of the total charging power at each time of each day; The time series prediction model is trained using a virtual sample set established by the power prediction data enhancement method for electric vehicle charging stations, and the parameters of the trained time series model are fine-tuned using a real sample set established by the power prediction data enhancement method for electric vehicle charging stations to obtain a power prediction model for electric vehicle charging stations; The power prediction data enhancement method of the electric vehicle charging station includes: (S1) collecting the total charging power of the charging station at each time of the day as a real sample to obtain a real sample set; The step (S1) comprises: (S11) collecting charging data of each charging pile in the charging station; the charging data includes power-on time, power-off time and charging energy; (S12) For each charging data, the average charging power in the charging time period is calculated, and the sum of the average charging powers of all charging piles at the same time is calculated to obtain the total charging power of the charging station at that time, and the total charging power of the charging station at each time constitutes the load time series of the charging station; (S13) dividing the load time series by day to obtain the total charging power of the charging station at each time of the day, taking it as a real sample to obtain a real sample set; (S2) For each real sample, V virtual samples are generated to obtain a virtual sample set, thus completing data enhancement; For any real sample, the methods of generating a virtual sample include: (S21) sampling S time series enhancement functions from a preset time series enhancement function set according to uniform distribution sampling; (S22) sequentially executing the S time series enhancement functions obtained by sampling on the real samples to obtain virtual samples; Wherein, V and S are both positive integers, S is greater than 1 and less than the size of the time series enhancement function set; each time series enhancement function in the time series enhancement function set has the same enhancement strength M, and S, V and M are determined as follows: Taking S, V and M as the variables to be optimized and maximizing the accuracy improvement of the virtual sample set on the actual power prediction task as the objective function, an optimization problem is established; The optimization problem is solved using a preset parameter optimization algorithm to obtain the values of S, V and M.
2. The method for establishing a power prediction model for an electric vehicle charging station according to claim 1, It is characterized in that The time series enhancement function set includes: jitter, scaling, time distortion, amplitude distortion, amplitude disturbance and phase disturbance, and the specific operation of each time series enhancement function is as follows: Jitter: Add zero-mean Gaussian noise to the time series; Scaling: changing the size of a certain segment of data within a sample by multiplying it by a random scalar; Time Warp: Smoothly warp the time intervals between samples; Amplitude distortion: convolve a certain section of data in the sample with a smooth curve with a mean of 1; Amplitude perturbation: Add zero-mean Gaussian noise to the amplitude of the sample frequency domain; Phase perturbation: Add zero-mean Gaussian noise to the phase values of the samples in the frequency domain.
3. A power prediction method for electric vehicle charging stations, It is characterized in that include: The historical data of the total charging power at each time of each day is collected and input into the power prediction model established by the power prediction model establishment method of the electric vehicle charging station according to claim 1 or 2, so as to obtain the prediction result of the total charging power at each time of each day in the future time period.
4. A power prediction model establishment system for electric vehicle charging stations, It is characterized in that include: A computer-readable storage medium for storing a computer program; And a processor, which is used to read the computer program stored in the computer-readable storage medium and execute the power prediction model establishment method for the electric vehicle charging station according to claim 1 or 2.
5. A power prediction system for electric vehicle charging stations, It is characterized in that include: A computer-readable storage medium for storing a computer program; and a processor, which is used to read the computer program stored in the computer-readable storage medium and execute the power prediction method for the electric vehicle charging station as claimed in claim 3.
Citation Information
Patent Citations
Electric vehicle charging station load prediction method based on support vector machine
CN103400203A
Electric vehicle charging load prediction method based on Monte Carlo and deep learning
CN110570014A
Charging station electricity consumption prediction method based on deep clustering model
CN112785029A
Time sequence prediction method and device, equipment and storage medium
CN113344129A
Fan icing generation power data enhancement method based on TimeGAN deep learning method
CN115994325A
