Charging station load prediction method and system considering multiple influence factors
By identifying multiple influencing factors and combining LSTM and integrated learning models, the problem of incomplete factors in traditional charging station load prediction is solved, more accurate and efficient load prediction is achieved, and the efficiency of charging station operation and management is improved.
Patent Information
- Application Number
- CN202410917857.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-10
- Publication Date
- 2024-11-26
AI Technical Summary
The research on load prediction of traditional charging stations is incomplete and difficult to process big data, resulting in inaccurate predictions.
Multiple influencing factors are used to divide them, combine LSTM model and ensemble learning, and a random forest model is constructed to predict the charging station load through GAN data augmentation and feature engineering.
It improves the accuracy and comprehensiveness of charging station load prediction, improves operational and management efficiency, and ensures data quality and model generalization capabilities.
Smart Images

Figure CN119029838A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of electric power regulation and control, and in particular to a method and system for predicting the load of a charging station taking multiple influencing factors into consideration. Background Art
[0002] In terms of influencing factors, traditional research on charging station load forecasting has problems such as incomplete consideration of factors and unscientific division of influencing factors. In terms of forecasting methods, traditional research mainly uses traditional regression statistical models, which are difficult to handle massive amounts of data in the big data era and difficult to achieve accurate forecasting.
[0003] With the widespread layout of charging stations, the use of electric vehicle charging is increasing, which in turn has an increasingly greater impact on the power grid load. This invention aims to study and analyze the impact of five main influencing factors on the charging station load, namely natural factors, user factors, charging facility factors, network and power factors, management and scheduling factors, and use machine learning to more accurately predict the charging station load.
[0004] The technical background of charging station load forecasting covers data collection and storage, time series analysis, machine learning, multivariate regression analysis and intelligent optimization algorithms. Currently, many researchers are making continuous progress in the above fields. The comprehensive application of these technologies and methods can improve the accuracy and reliability of charging station load forecasting and provide strong support for the operation and management of charging stations.
[0005] Therefore, this prediction method can solve the accuracy and comprehensiveness of charging station load prediction and break through technical difficulties such as data processing, feature extraction and time series modeling. By comprehensively considering multiple influencing factors and using LSTM model and ensemble learning model for prediction, more accurate load prediction can be achieved, thus improving the operation and management efficiency of charging stations. Summary of the invention
[0006] In view of the above-mentioned problems, the present invention is proposed.
[0007] Therefore, in order to solve the above technical problems, the present invention provides the following technical solutions: a method for predicting the load of a charging station considering multiple influencing factors, comprising: identifying and classifying the main influencing factors affecting the load of the charging station;
[0008] Collect the main influencing data, preprocess the main influencing factor data, and perform feature engineering conversion on the preprocessed main influencing factor data;
[0009] The GAN model is used for data enhancement, and the training set is expanded with the original data. The LSTM prediction model is established using the selected feature data, and the expanded training set is used to train and evaluate the LSTM prediction model.
[0010] A random forest model is constructed, and the prediction results of the LSTM sub-model are used as input features to make the final prediction through ensemble learning.
[0011] As a preferred solution of the charging station load prediction method considering multiple influencing factors described in the present invention, wherein: the main influencing factors include natural factors, user factors, charging facility factors, network and power factors, and management and scheduling factors;
[0012] The natural factors include weather, temperature, rainfall and holiday data;
[0013] The user factors include charging frequency, duration, location, time distribution and remaining power data;
[0014] The charging facility factors include the number, type and power data of charging piles;
[0015] The network and power factors include grid connection capabilities, stability and pricing mechanism data;
[0016] The management and dispatch factors include load management strategies, operating modes and availability management data.
[0017] As a preferred solution of the charging station load prediction method considering multiple influencing factors described in the present invention, wherein: the data preprocessing includes outlier processing and data standardization;
[0018] The outlier processing includes:
[0019] For errors in the data, use statistical methods and K-means algorithm to detect and process outliers;
[0020] Construct two cluster centers a = {a1, a2}, which are the daily data collected by the charging station and the extreme data, a j The set is M j ;
[0021] Import the relevant center data into the corresponding data set. The Euclidean distance calculation formula from the center to the data set is as follows:
[0022]
[0023] Among them, x i For clustered data, x j is the center of the cluster, b(x i ,x j ) is the Euclidean distance from the data to the center;
[0024] According to the clustering algorithm, the relevant data is imported into the data cluster where the nearest cluster center is located;
[0025] Iterate all the data in the set of the nearest cluster center to the cluster center to achieve the minimum distance from each data to the iterated center. The calculation formula for the cluster center with the minimum distance is as follows:
[0026]
[0027] Among them, a j is the jth cluster center, M j are all data points in the jth cluster;
[0028] Repeat the iteration until all data is imported and the calculation of K clusters is completed;
[0029] The data standardization includes:
[0030] Perform Z standardization to convert all data into a standard normal distribution form;
[0031] The calculation formula for Z-standardization is as follows:
[0032]
[0033] where x i (i=1,2,...,n) is the i-th sample data, Z i is the standardized value, μ is the sample population mean, and σ is the sample standard deviation; the calculation formulas for the sample population mean and sample standard deviation are as follows:
[0034]
[0035] Among them, n is the total number of sample data.
[0036] As a preferred solution of the charging station load forecasting method considering multiple influencing factors described in the present invention, wherein: the feature engineering conversion includes natural factor feature conversion, user factor feature conversion, charging facility factor feature conversion, network and power factor feature conversion, management and scheduling factor feature conversion and data time alignment;
[0037] Transform and encode the non-numerical features in the main factors, and convert the various factors into a form suitable for input into the LSTM model;
[0038] Associate various data with the charging load data of the charging station to align the time of the two and establish a relationship model between each classification factor and the charging load;
[0039] The natural factor characteristic conversion includes:
[0040] The rainfall factor is converted into a binary feature, with rainfall set to 1 and no rainfall set to 0;
[0041] Convert the holiday factor into a binary feature, setting holidays to 1 and non-holidays to 0;
[0042] The user factor feature conversion includes:
[0043] The charging location is represented as the number of the charging pile, and the charging location is encoded as a discrete category feature to consider the impact of different charging locations on the load;
[0044] The time distribution of charging activities in a day is divided into different time periods, and different time periods are used as features to consider the impact of time on load;
[0045] The charging facility factor feature conversion includes:
[0046] Charging stations have different types of charging facilities, and unique hot encoding is used to convert different types of charging facilities into binary features;
[0047] The network and power factor feature conversion includes:
[0048] The ratio of the actual load of the charging station to the line capacity indicates the connection capacity with the power grid. The greater the line capacity utilization rate, the stronger the connection capacity with the power grid.
[0049] At the same time, the voltage standard deviation over a period of time is calculated. The lower the standard deviation, the more stable the power grid is, and vice versa, the more unstable the power grid is.
[0050] The time is divided into peak period and valley period, and the peak and valley electricity prices are encoded as features. When the time is in the valley period, the value is assigned to 0, and when it is in the peak period, the value is assigned to 1;
[0051] The management and scheduling factor feature conversion includes:
[0052] By calculating the standard deviation of the load data, a smaller standard deviation means a smaller load fluctuation, and the load fluctuation is used to indicate the smoothness of the load data;
[0053] Because the operation modes of charging stations are different, we use unique hot encoding to convert these modes into binary features. The operation modes include self-operation and third-party operation. We set self-operation to 1 and third-party operation to 0.
[0054] Calculate availability indicators based on equipment uptime and downtime;
[0055] Time alignment of data includes,
[0056] Based on the results of timestamp alignment and time range matching, other types of data are associated with charging load data;
[0057] Use timestamp as primary key to join or merge two data sets, each timestamp has corresponding charging load data and other class data.
[0058] As a preferred solution of the charging station load prediction method considering multiple influencing factors described in the present invention, wherein: the data enhancement using the GAN model includes:
[0059] Use the GAN model to enhance the influencing factors and load data and expand the training set;
[0060] Define the network structure of the generator (G) and the discriminator (D). The input of the generator network G is the noise vector z, and the output is the synthesized charging load and influencing factor data.
[0061] The input of the discriminator network D is real data or synthetic data output by the generator, and the output is the true or false classification result;
[0062] Perform adversarial training on the generator G and the discriminator D, and iteratively train G and D alternately until the Nash equilibrium is reached;
[0063] After the training is completed, the generator G is used to generate a large amount of synthetic charging load and influencing factor data;
[0064] Combine the generated data with the original training data to expand the training set;
[0065] In the case of missing values, the GAN model is used to generate a large amount of similar data to make up for the missing values;
[0066] For the data generated by the GAN model, the absolute error and relative error are used to judge the accuracy of data processing;
[0067] The calculation formula of the absolute error is as follows:
[0068]
[0069] Among them, AE is the absolute error, y(i) is the actual value, is the predicted value;
[0070] The calculation formula of the relative error is as follows:
[0071]
[0072] Among them, RE is the relative error.
[0073] As a preferred solution of the charging station load prediction method considering multiple influencing factors described in the present invention, wherein: the use of selected feature data to establish an LSTM prediction model includes:
[0074] The calculation formula of the LSTM prediction model is as follows:
[0075] S t =f(U / x t +W / S t-1 )
[0076] O t =g(V / S t )
[0077] Among them, S t-1 , S t is the memory information data at different times, x t is the input data, O t is the data output, f and g are activation functions, W, U and V are the automatic learning weight matrices;
[0078] LSTM prediction model parameter settings include:
[0079] The load data of the charging station is matched with the features one by one, and one neuron is input into the input layer of the LSTM prediction model, and the input neuron is defined as a;
[0080] The number of output neurons is related to the number of data modules in the output layer. There are 24 groups of sample points extracted in one day, and there are 24 output neurons. The 24 output neurons are defined as b;
[0081] The calculation formula for the number of neurons in the hidden layer of the LSTM prediction model is as follows:
[0082]
[0083] Where, E is the number of neurons, η is the correction coefficient;
[0084] Assign activation function, using Tanh, Relu and Sigmoid functions;
[0085] The calculation formula of the Tanh function is as follows:
[0086]
[0087] The calculation formula of the ReLU function is as follows:
[0088] g(z)=max(0,z)
[0089] The calculation formula of the Sigmoid function is as follows:
[0090]
[0091] Among them, z is the weighted sum received by the neuron;
[0092] Use historical data for model training and validation data for evaluation;
[0093] The predictive performance of the evaluation model was evaluated using the error evaluation metric;
[0094] The error evaluation indicators include mean absolute percentage error, mean absolute error and root mean square error. The error evaluation indicator calculation formula is as follows:
[0095]
[0096] in, is the predicted value, specifically y i is the true value, specifically y={y1,y2,…,y n}, n is the number of samples.
[0097] As a preferred solution of the charging station load prediction method considering multiple influencing factors described in the present invention, wherein: the final prediction through ensemble learning includes:
[0098] Based on the five influencing factors of natural factors, user factors, charging facility factors, network and power factors, and management and scheduling factors, the LSTM model was used to predict them respectively, and the prediction outputs of five LSTM sub-models were obtained;
[0099] Construct random forest input features, use the prediction output results of the five LSTM sub-models as the input features of the random forest model, and construct a 5-dimensional feature vector, including natural factor prediction values, user factor prediction values, charging facility prediction values, network and power prediction values, and management and scheduling prediction values;
[0100] Train the random forest model and obtain multiple bootstrap samples from the original training data through the bootstrap sampling method;
[0101] For each bootstrap sample, a decision tree model is trained. Some features are randomly selected for splitting during training, and multiple trained decision trees are combined into a random forest model.
[0102] Perform integrated prediction, input the new sample to be predicted into the trained random forest model, and the random forest inputs the sample into each decision tree to obtain multiple prediction results;
[0103] By voting, the predictions of multiple decision trees are combined to obtain the final integrated prediction output;
[0104] Model evaluation and optimization: Use the test set to evaluate the predictive performance indicators of the random forest model and adjust the hyperparameters of the random forest based on the evaluation results.
[0105] The random forest algorithm includes information entropy, conditional information entropy, information gain, and prediction error;
[0106] The information entropy is used to measure the quality of feature segmentation, and the calculation formula is as follows:
[0107]
[0108] Where Y is a set of categories, which in this patent represents five different influencing factors, and p(y) is the probability of category y;
[0109] The conditional information entropy is used to measure the quality index of the category probability after given feature segmentation, and the calculation formula is as follows:
[0110]
[0111] The information gain is used to measure the reduction of information entropy due to feature segmentation, and the calculation formula is as follows:
[0112] IG(X,Y)=H(Y)-H(Y|X)
[0113] The prediction error is an indicator used to measure the prediction performance of the model for test data, and is measured using square error or cross entropy.
[0114] Another object of the present invention is to provide a charging station load prediction system that considers multiple influencing factors. To solve the above technical problems, the present invention provides the following technical solutions: A charging station load prediction system that considers multiple influencing factors includes: a data acquisition module, a data preprocessing module, a feature engineering module, a data enhancement module, a model training module, and an integrated prediction module;
[0115] The data acquisition module is used to identify and classify the main influencing factors affecting the charging station load, and collect data on the main influencing factors;
[0116] The data preprocessing module is used to preprocess the collected data of the main influencing factors
[0117] The feature engineering module is used to transform and encode the non-numerical features in the main factors into a form suitable for input into the LSTM model, and to time-align the various data with the charging load data of the charging station to establish a relationship model between each classification factor and the charging load;
[0118] The data enhancement module is used to enhance the influencing factors and load data using the GAN model to expand the training set;
[0119] The model training module is used to establish and train the LSTM prediction model, input feature data, output prediction values, perform model training on historical data, and use verification data to evaluate model performance;
[0120] The integrated prediction module is used to perform integrated prediction, input the samples to be predicted into the trained random forest model, and integrate the predictions of multiple decision trees to obtain the final prediction output.
[0121] A computer device includes a memory and a processor, wherein the memory stores a computer program, and is characterized in that when the processor executes the computer program, the steps of the charging station load prediction method considering multiple influencing factors as described above are implemented.
[0122] A computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the steps of the charging station load prediction method considering multiple influencing factors as described above are implemented.
[0123] Beneficial effects of the present invention: The present invention proposes a charging station load prediction method that takes multiple influencing factors into consideration, and realizes accurate prediction of charging station load through multiple steps, thereby improving the operating efficiency of charging stations and the stability of power grids; identifies and divides natural factors, user factors, charging facility factors, network and power factors, and management and scheduling factors, thereby realizing classified management and analyzing the impact of different factors on loads; data preprocessing and feature engineering conversion ensure high and consistent data quality, and cleans data through outlier processing and Z-standardization processing, and converts it into feature data suitable for model input; uses a GAN model for data enhancement to generate high-quality synthetic data, solves the problem of insufficient and missing data, and significantly improves the generalization ability of the model; constructs an LSTM prediction model and a random forest model for multi-level prediction, wherein the LSTM model processes time series data, and the random forest model comprehensively predicts the results of various factors, thereby improving prediction accuracy and reliability; the organic combination of the comprehensive steps improves the accuracy and comprehensiveness of charging station load prediction, and provides reliable data support for optimizing charging station management and power grid stability; BRIEF DESCRIPTION OF THE DRAWINGS
[0124] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0125] Figure 1 This is a schematic diagram of the structure of a terminal in the first embodiment of the present invention;
[0126] Figure 2 It is a schematic diagram of a flow chart in the first embodiment of the present invention;
[0127] Figure 3FIG. 4 is a schematic diagram of the structure of a computer device in a third embodiment of the present invention.
[0128] Figure 4 It is a prediction schematic diagram in the first embodiment of the present invention; DETAILED DESCRIPTION
[0129] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.
[0130] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.
[0131] Example 1
[0132] Reference Figure 1-Figure 4 , which is an embodiment of the present invention, provides a charging station load prediction method considering multiple influencing factors.
[0133] First, the charging station load prediction method considering multiple influencing factors provided in this application can be applied to Figure 1 In the terminal shown. Figure 1 As shown, the terminal may include one or two ( Figure 1 Only one is shown in the figure) processor and memory for storing data, wherein the processor may include but is not limited to a processing system such as a microprocessor MCU or a programmable logic device FPGA. The above terminal may also include a transmission device and an input and output device for communication functions. It can be understood by ordinary technicians in this field that Figure 1 The structure shown is only for illustration and does not limit the structure of the above terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.
[0134] The memory can be used to store computer programs, such as the computer program corresponding to the charging station load prediction method considering multiple influencing factors in this embodiment. The processor executes various functional applications and data processing by running the computer program stored in the memory, that is, implementing the above method. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage systems, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely disposed relative to the processor, and these remote memories may be connected to the terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
[0135] The transmission device is used to receive or send data via a network. The above-mentioned network includes a wireless network provided by a communication provider of the terminal. In one example, the transmission device includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device can be a radio frequency (Radio Frequency, referred to as RF) module, which is used to communicate with the Internet wirelessly.
[0136] like Figure 2 As shown, the embodiment of the present invention provides a charging station load prediction method considering multiple influencing factors, and the method is applied to Figure 1 The terminal in is used as an example to illustrate, including the following steps:
[0137] S1: Identify and classify the main factors affecting the charging station load.
[0138] S1.1: The main factors affecting the load of charging stations include natural factors, user factors, charging facility factors, network and power factors, and management and scheduling factors;
[0139] Natural factors include weather, temperature, rainfall, and holiday data;
[0140] User factors include charging frequency, duration, location, time distribution, and remaining power data;
[0141] Charging facility factors include the number, type and power data of charging piles;
[0142] Network and power factors include data on grid connectivity, stability and pricing mechanisms;
[0143] Management and dispatch factors include load management strategies, operating models, and availability management data;
[0144] Natural factors include:
[0145] Since users have different travel modes and charging convenience requirements in rainy and non-rainy conditions, and thus their charging needs and charging locations will also change, whether it rains or not will affect the charging load of the charging station;
[0146] Because the length of time users use the air conditioner (for example, users use the air conditioner for longer periods of time in summer and winter) and the battery life (for example, lower temperatures may weaken the activity of lithium ions) vary under different temperature conditions, power consumption and charging frequency will also change. Therefore, temperature changes will also affect the load of the charging station.
[0147] User factors include:
[0148] Whether the remaining power is sufficient or not will lead to different charging urgency, which will cause the user's charging time (users may choose to increase the charging time if the remaining power is too low) and charging location to change;
[0149] Different destination types will lead to differences in users' willingness to charge and length of stay, which will lead to changes in users' charging locations (for example, users may prefer to use home charging piles or charging stations near their homes after getting home from work) and charging times. Therefore, different destination types will affect the load of different charging stations.
[0150] Charging facility factors include:
[0151] Currently, most electric vehicle charging stations mainly provide two charging methods: ordinary slow charging and high-voltage fast charging;
[0152] Slow charging has a lower power and has less impact on the load of the charging station, while high-voltage fast charging is instantaneous high voltage and has a great impact on the load.
[0153] Network and power factors include,
[0154] The connection capabilities of a charging station also determine its load capacity, i.e. how many electric vehicles it can provide charging services for at the same time;
[0155] Charging stations with high connection capabilities can charge more electric vehicles at the same time, thus having a greater load capacity.
[0156] Management and scheduling factors include,
[0157] The load management and optimization strategies of charging stations will also have an impact on the load;
[0158] Through reasonable load management and optimization strategies, the charging rate, charging sequence and charging strategy of the charging station can be adjusted to balance the load demand and grid capacity;
[0159] Study in detail the specific content of each type of factor and its possible impact on the load of charging stations, such as weather changes in natural factors and charging frequency in user factors.
[0160] S2: Collect the main influencing data, preprocess the main influencing factor data, and perform feature engineering conversion on the preprocessed main influencing factor data.
[0161] S2.1: Data preprocessing includes outlier processing and data standardization;
[0162] Outlier processing includes:
[0163] In view of the errors in the data, statistical methods and K-means algorithm are used to detect and process outliers.
[0164] Due to human factors (such as misoperation by the monitor, frequent plugging and unplugging by the user, damage, etc.) or machine failure, the load data of the charging station will inevitably have extreme values. For abnormal data, this paper uses the most commonly used clustering algorithm based on Euclidean distance - K-means algorithm, which is characterized by classifying data into K groups according to the distance of the samples;
[0165] The data within a group should be as close as possible, and the data between groups should be as far apart as possible;
[0166] Construct two cluster centers a = {a1, a2}, which are the daily data collected by the charging station and the extreme data, a j The set is M j ;
[0167] Import the relevant center data into the corresponding data set. The Euclidean distance calculation formula from the center to the data set is as follows:
[0168]
[0169] Among them, x i For clustered data, x j is the center of the cluster, b(x i ,x j ) is the Euclidean distance from the data to the center; and according to the clustering algorithm, the relevant data is imported into the data cluster where the nearest center is located;
[0170] Then iterate the center of the cluster, and iterate this data center through all the data in the set where the center of the cluster is located, and finally achieve the minimum distance from each data to the iterated center, thus obtaining
[0171]
[0172] Among them, a j is the center of the jth cluster, M j are all the data points in the jth cluster.
[0173] Iterate according to the above process until all data is imported and the calculation of K clusters is completed.
[0174] S2.2: Data standardization includes:
[0175] Perform Z standardization to convert all data into a standard normal distribution form for unified processing and analysis;
[0176] Z-standardization is a commonly used standardization process. Its operating principle is to convert data at different levels into Z scores so that they can be compared under a unified standard. The calculation formula is as follows:
[0177]
[0178] where x i (i=1,2,...,n) is the i-th sample data, Z i is the standardized value, μ is the sample population mean, and σ is the sample standard deviation; the calculation formulas for the sample population mean and sample standard deviation are as follows:
[0179]
[0180] Where n is the total number of sample data;
[0181] The sample data is Z-standardized through the above process, and all sample data are under a unified standard, so as to ensure subsequent comparative analysis.
[0182] S2.3: Feature engineering includes,
[0183] Transform and encode the non-numerical features in the main factors, so as to convert various factors into a form suitable for input into the LSTM model;
[0184] Correlate the various data with the charging load data of the charging station to ensure that the two are time-aligned, so that a relationship model between each classification factor and the charging load can be established;
[0185] The transformation of natural factor characteristics includes:
[0186] The rainfall factor is converted into a binary feature, with rainfall set to 1 and no rainfall set to 0;
[0187] Convert the holiday factor into a binary feature, setting holidays to 1 and non-holidays to 0;
[0188] User factor feature conversion includes:
[0189] The charging location is represented as the number of the charging pile, and the charging location is encoded as a discrete category feature to consider the impact of different charging locations on the load;
[0190] The time distribution of charging activities in a day is divided into different time periods, and different time periods are used as features to consider the impact of time on load;
[0191] Charging facility factor feature conversion includes:
[0192] Charging stations have different types of charging facilities, and unique hot encoding is used to convert these factors into binary features. For example, there are two charging modes: fast charging and slow charging. Fast charging is set to 1 and slow charging is set to 0;
[0193] Network and power factor feature conversion includes,
[0194] The ratio of the actual load of the charging station to the line capacity indicates the connection capacity with the power grid. The greater the line capacity utilization rate, the stronger the connection capacity with the power grid.
[0195] Calculate the voltage standard deviation over a period of time. The lower the standard deviation, the more stable the power grid is, and vice versa.
[0196] The time is divided into peak period and valley period, and the peak and valley electricity prices are encoded as features. When the time is in the valley period, the value is assigned to 0, and when it is in the peak period, the value is assigned to 1;
[0197] Management and scheduling factor feature transformations include,
[0198] By calculating the standard deviation of the load data, a smaller standard deviation means a smaller load fluctuation, and the load fluctuation is used to indicate the smoothness of the load data;
[0199] Because the operation modes of charging stations are different, we use unique hot encoding to convert these modes into binary features. For example, there are two main operation modes: self-operation and third-party operation. We set self-operation to 1 and third-party operation to 0.
[0200] The availability index is calculated based on the equipment's operating time and downtime. For example, if the equipment's operating time is 98 hours and its downtime is 2 hours, the characteristic of the availability index is 0.98, and the calculation formula is as follows:
[0201]
[0202] Time alignment of data includes,
[0203] Based on the results of timestamp alignment and time range matching, other types of data are associated with charging load data;
[0204] Use the timestamp as the primary key to join or merge the two data sets to ensure that each timestamp has corresponding charging load data and other class data.
[0205] S3: Use the GAN model for data enhancement, and expand the training set with the original data. Use the selected feature data to establish the LSTM prediction model, and use the expanded training set to train and evaluate the LSTM prediction model.
[0206] S3.1: Use the GAN model to enhance the influencing factors and load data, expand the training set, and prepare for the subsequent LSTM model training;
[0207] Since the data on charging load and its influencing factors are insufficient or unevenly distributed, the GAN model is used to generate more representative synthetic data to expand the training set;
[0208] GAN can learn the potential distribution of charging load data and generate new data that is consistent with the statistical characteristics of real data to make up for the shortcomings of the original data;
[0209] GAN is a generative model consisting of a generator and a discriminator. Its goal is to generate realistic samples through adversarial training.
[0210] The generator and discriminator of the generative model both contain input layer, hidden layer and output layer. The specific structure and parameters are shown in the following table:
[0211] Table 1 GAN generation model structure table
[0212]
[0213] Define the network structure of the generator (G) and the discriminator (D). The input of the generator network G is the noise vector z, and the output is the synthesized charging load and influencing factor data.
[0214] The input of the discriminator network D is real data or synthetic data output by the generator, and the output is the true or false classification result;
[0215] Next, adversarial training is performed, training G and D alternately and iteratively until the Nash equilibrium is reached, that is, neither G nor D can continue to improve performance;
[0216] After the training is completed, the generator G is used to generate a large amount of synthetic charging load and influencing factor data;
[0217] Combine the generated data with the original training data to expand the training set and apply it to the subsequent charging load prediction model training;
[0218] In the case of missing values, some studies will use the average value to fill the missing values, while the application of the GAN model can generate a large amount of similar data to fill the missing values and avoid large errors;
[0219] For the data generated by the GAN model, the absolute error and relative error are used to judge the accuracy of data processing;
[0220] The absolute error is calculated as follows:
[0221]
[0222] Among them, AE is the absolute error, y(i) is the actual value, is the predicted value;
[0223] The calculation formula of relative error is as follows:
[0224]
[0225] Among them, RE is the relative error;
[0226] Error calibration can be used to compare the effects of generating data to make up for missing data and replacing missing data with the average value, as shown in the following table.
[0227] Table 2 Effect of using GAN model to make up for missing data
[0228] Applying GAN model to make up for missing data Average instead RE 3.4% 21.6% AE 2.84 7.62
[0229] It can be seen from the table that the error value caused by filling the missing data with the mean is more than 20%, which is too high. However, using the generative model to fill the missing values can significantly reduce the error and greatly improve the accuracy of the prediction.
[0230] S3.2: LSTM model includes,
[0231] The LSTM model performs well in time series forecasting tasks, especially for processing data with long-term dependencies;
[0232] Many of the above factors are not simple linear relationships, and the LSTM model has a strong ability to capture complex nonlinear relationships;
[0233] The LSTM model has higher accuracy in predicting the load of charging stations.
[0234] The calculation formula of the LSTM prediction model is as follows:
[0235] S t =f(U*x t +W*S t-1 )
[0236] Ot =g(V*S t )
[0237] Among them, S t-1 , S t is the memory information data at different times, x t is the input data, O t is the data output, f and g are activation functions, W, U and V are the automatic learning weight matrices;
[0238] The long short-term memory model can effectively solve the problem that the RNN model is prone to collapse when the amount of data is too large. It can learn more efficiently and process and predict data.
[0239] The long short-term memory model parameter settings include:
[0240] The load data of the charging station corresponds to the characteristics one by one, and one neuron is input into the input layer of the long short-term memory model, named a;
[0241] The number of output neurons is related to the number of data modules in the output layer. Since there are 24 groups of sample points extracted in one day, there are 24 output neurons, named b.
[0242] The formula for calculating the number of neurons in the hidden layer of the long short-term memory model is as follows:
[0243]
[0244] Among them, η is the correction coefficient. The number of neurons in the hidden layer determines the complexity and modeling ability of LSTM. Its size range is a natural number between 1 and 10, and the error is reduced by controlling the number of neurons.
[0245] Assign activation function. The activation function uses Tanh, Relu and Sigmoid functions. The activation function can better improve the nonlinear expression of the model.
[0246] The activation functions used include Tanh, ReLU and Sigmoid functions, one of their main functions is to introduce nonlinear transformation;
[0247] Combinations of linear transformations are still linear, while nonlinear transformations can handle more complex data distributions and tasks;
[0248] The calculation formula of the Tanh function is as follows:
[0249]
[0250] The calculation formula of the ReLU function is as follows:
[0251] g(z)=max(0,z)
[0252] The calculation formula of the Sigmoid function is as follows:
[0253]
[0254] Among them, z is the weighted sum received by the neuron, that is, the sum of the product of the input signal and the corresponding weight, and the value range of the input value z is [-1,1];
[0255] S3.3: Use historical data for model training and validation data for evaluation and optimization;
[0256] The optimization uses error evaluation indicators to evaluate the predictiveness of the model;
[0257] The error evaluation indicators include mean absolute percentage error, mean absolute error and root mean square error. The error evaluation indicator calculation formula is as follows:
[0258]
[0259] in, is the predicted value, specifically y i is the true value, specifically y={y1,y2,…,y n}, n is the number of samples;
[0260] RMSE root mean square error is kept at the same level as the true value. The larger the RMSE value, the greater the error. It can more intuitively measure the deviation between the predicted value and the true value;
[0261] MAPE (mean absolute percentage error) normalizes each point and expresses the difference from the true value in percentage. MAPE can effectively reduce the impact of outliers. Its value range is [0, +∞). The smaller the MAPE error value, the higher the prediction accuracy of the model.
[0262] MAE mean absolute error is used to calculate the deviation between the observed value and the average value. It mainly observes the degree of dispersion. The larger the value, the greater the error.
[0263] S4: Build a random forest model, use the prediction results of the LSTM sub-model as input features, make final predictions through ensemble learning, and optimize model parameters to improve prediction performance.
[0264] S4.1: The above work predicts five types of influencing factors, namely natural factors, user factors, charging facility factors, network and power factors, and management and scheduling factors, through the LSTM model;
[0265] The random forest algorithm in the ensemble learning model is used to integrate these five factors, comprehensively consider natural factors, user factors, charging facility factors, network and power factors, management and scheduling factors, and make an overall load forecast for the charging station;
[0266] Random forest, as an integrated learning algorithm, improves classification accuracy by building multiple decision trees and voting;
[0267] It has strong robustness and anti-overfitting capabilities, and can effectively process high-dimensional, nonlinear power safety feature data. The core of the algorithm mainly includes information entropy, conditional information entropy, information gain and prediction error;
[0268] Information entropy is an indicator used to measure the quality of feature segmentation. The calculation formula is as follows:
[0269]
[0270] Where Y is a set of categories, which in this patent represents five different influencing factors, and p(y) is the probability of category y;
[0271] Conditional information entropy is a quality indicator used to measure the probability of a given feature segmentation category. The calculation formula is as follows:
[0272]
[0273] Information gain is an indicator used to measure the reduction of information entropy caused by feature segmentation. The calculation formula is as follows:
[0274] IG(X,Y)=H(Y)-H(Y|X)
[0275] Prediction error is an indicator used to measure the prediction performance of the model for test data, measured by square error or cross entropy;
[0276] The steps to apply random forest for ensemble learning are as follows:
[0277] Construct random forest input features, use the prediction results of the previous LSTM sub-model as the input features of the random forest model, and construct a 5-dimensional feature vector, including natural factor prediction values, user factor prediction values, charging facility prediction values, network and power prediction values, and management and scheduling prediction values;
[0278] Train the random forest model and obtain multiple bootstrap samples from the original training data through the bootstrap sampling method;
[0279] For each bootstrap sample, a decision tree model is trained. During training, some features are randomly selected for splitting, and the trained multiple decision trees are combined into a random forest model.
[0280] Perform integrated prediction, input the new sample to be predicted into the trained random forest model, and the random forest inputs the sample into each decision tree to obtain multiple prediction results;
[0281] By voting, the predictions of multiple decision trees are combined to obtain the final integrated prediction output;
[0282] Model evaluation and optimization, use the test set to evaluate the predictive performance indicators of the random forest model, and adjust the hyperparameters of the random forest based on the evaluation results.
[0283] Example 2
[0284] As an embodiment of the present invention, a charging station load prediction system considering multiple influencing factors is provided, comprising: a data acquisition module, a data preprocessing module, a feature engineering module, a data enhancement module, a model training module, and an integrated prediction module;
[0285] The data acquisition module is used to identify and classify the main factors affecting the load of the charging station and collect the data of the main factors;
[0286] The data preprocessing module is used to preprocess the collected data of the main influencing factors
[0287] The feature engineering module is used to transform and encode the non-numerical features in the main factors into a form suitable for input into the LSTM model, and to time-align the various data with the charging load data of the charging station to establish a relationship model between each classification factor and the charging load;
[0288] The data enhancement module is used to enhance the influencing factors and load data using the GAN model and expand the training set;
[0289] The model training module is used to build and train the LSTM prediction model, input feature data, output prediction values, perform model training on historical data, and use validation data to evaluate model performance;
[0290] The integrated prediction module is used to perform integrated prediction. The samples to be predicted are input into the trained random forest model, and the predictions of multiple decision trees are combined to obtain the final prediction output.
[0291] For the specific limitations of the charging station load prediction system considering multiple influencing factors, please refer to the limitations of the charging station load prediction method considering multiple influencing factors mentioned above, which will not be repeated here. Each module in the above-mentioned charging station load prediction system considering multiple influencing factors can be implemented in whole or in part through software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.
[0292] Example 3
[0293] Reference Figure 3 , which is the third embodiment of the present invention. Based on the first two embodiments, the embodiment of the present invention provides a computer device, which may be a server, and its internal structure diagram may be as shown in FIG. Figure 3 As shown. The computer device includes a processor, a memory and a network interface connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium.
[0294] The database of the computer device is used to store action detection data. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the steps in any of the above-mentioned sparse tensor operation acceleration method embodiments are implemented.
[0295] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0296] In one embodiment, an embodiment of the present invention provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps in any of the above-mentioned sparse tensor operation acceleration method embodiments are implemented.
[0297] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).
[0298] Example 4
[0299] As an embodiment of the present invention, a charging station load prediction method considering multiple influencing factors is provided. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.
[0300] First, identify and classify the main factors affecting the load of charging stations, including natural factors, user factors, charging facility factors, network and power factors, and management and scheduling factors. Among them, natural factors include weather, temperature, rainfall, and holiday data; user factors include charging frequency, duration, location, time distribution, and remaining power data; charging facility factors include the number, type, and power of charging piles; network and power factors include grid connection capabilities, stability, and pricing mechanism data; management and scheduling factors include load management strategies, operating models, and availability management data.
[0301] Then, the data of the above main influencing factors were collected and preprocessed. Data preprocessing includes outlier processing and data standardization. Outlier processing uses the K-means clustering algorithm to divide the data into two clusters, normal data and extreme data, and process them separately. Data standardization uses the Z-Score standardization method to convert the data into a standard normal distribution form.
[0302] Next, the preprocessed data of the main influencing factors are transformed into feature engineering. The non-numerical features are encoded and converted into a form suitable for input into the LSTM model. At the same time, the data is time-aligned to make the timestamps of the various influencing factor data consistent with the load data.
[0303] In order to expand the training set, this embodiment uses the GAN model to enhance the influencing factors and load data. Define the generator and discriminator network structure, conduct adversarial training until the Nash equilibrium is reached, and then use the generator to generate a large amount of synthetic data, which is combined with the original training data to form an expanded training set.
[0304] Use the expanded training set to build an LSTM prediction model. Set the number of input and output neurons of the LSTM model, adjust the number of neurons in the hidden layer, select appropriate activation functions such as Tanh, ReLU and Sigmoid, and perform model training and performance evaluation.
[0305] Finally, a random forest model is constructed, and the five prediction results of the LSTM sub-model are used as the input features of the random forest. The final prediction output is obtained through self-sampling and ensemble learning. The prediction performance of the model is evaluated and optimized, and the experimental data is recorded in the following table:
[0306] Table 1 Charging station load prediction experimental data
[0307]
[0308] It can be seen from the above table that the method of the present invention can accurately predict the load situation of the charging station.
[0309] Comparing the prediction results of a single LSTM model with the prediction results of the random forest model through ensemble learning, the prediction performance of the random forest model is better. This is because the random forest model combines the prediction results of multiple LSTM sub-models and improves the generalization ability through ensemble learning. For example, for objects A, C, and F, the prediction values of the random forest model are obviously closer to the true values.
[0310] In addition, the introduction of GAN model for data enhancement also improves the generalization performance of the model. By generating a large amount of synthetic data to expand the training set, the model can learn more potential data distribution patterns, thereby improving the accuracy and robustness of the prediction.
[0311] It is worth noting that the method of the present invention comprehensively considers multiple influencing factors and can fully capture the key features that affect the load of charging stations. From the data table, it can be seen that natural factors such as temperature and rainfall, as well as human factors such as charging facilities, have an impact on the final load forecast.
[0312] Compared with the prior art, the method of the present invention is innovative in that:
[0313] A system influencing factor classification framework is proposed, which comprehensively covers the main factors affecting the charging station load.
[0314] Through feature engineering, various influencing factor data are converted into a form suitable for input into the machine learning model and associated with the load data.
[0315] It combines multiple advanced technologies such as GAN data enhancement, LSTM time series modeling and random forest ensemble learning to form an efficient and robust prediction model.
[0316] The model was optimized and adjusted through evaluation indicators such as information gain and prediction error to improve the prediction accuracy.
[0317] In general, the method of the present invention has effectively solved the problem that the existing technology is difficult to accurately predict the load of charging stations, provides important support for the intelligent scheduling and energy management of power systems, and has important theoretical significance and application value.
[0318] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.
Claims
1. A charging station load prediction method considering multiple influencing factors, characterized in that: include: Identify and classify the main factors affecting the charging station load; Collect the main influencing data, preprocess the main influencing factor data, and perform feature engineering conversion on the preprocessed main influencing factor data; The GAN model is used for data enhancement, and the training set is expanded with the original data. The LSTM prediction model is established using the selected feature data, and the expanded training set is used to train and evaluate the LSTM prediction model. A random forest model is constructed, and the prediction results of the LSTM sub-model are used as input features to make the final prediction through ensemble learning.
2. The method for predicting charging station load considering multiple influencing factors as claimed in claim 1, characterized in that: The main influencing factors include natural factors, user factors, charging facility factors, network and power factors, and management and scheduling factors; The natural factors include weather, temperature, rainfall and holiday data; The user factors include charging frequency, duration, location, time distribution and remaining power data; The charging facility factors include the number, type and power data of charging piles; The network and power factors include grid connection capabilities, stability and pricing mechanism data; The management and dispatch factors include load management strategies, operating modes and availability management data.
3. The method for predicting charging station load considering multiple influencing factors as claimed in claim 2, characterized in that: The data preprocessing includes outlier processing and data standardization; The outlier processing includes: For errors in the data, use statistical methods and K-means algorithm to detect and process outliers; Construct two cluster centers a = {a1, a2}, which are the daily data collected by the charging station and the extreme data, a j The set is M j ; Import the relevant center data into the corresponding data set. The Euclidean distance calculation formula from the center to the data set is as follows: Among them, x i For clustered data, x j is the center of the cluster, b(x i ,x j ) is the Euclidean distance from the data to the center; According to the clustering algorithm, the relevant data is imported into the data cluster where the nearest cluster center is located; Iterate all the data in the set of the nearest cluster center to the cluster center to achieve the minimum distance from each data to the iterated center. The calculation formula for the cluster center with the minimum distance is as follows: Among them, a j is the jth cluster center, M j are all data points in the jth cluster; Repeat the iteration until all data is imported and the calculation of K clusters is completed; The data standardization includes: Perform Z standardization to convert all data into a standard normal distribution form; The calculation formula for Z-standardization is as follows: where x i (i=1,2,...,n) is the i-th sample data, Z i is the standardized value, μ is the sample population mean, and σ is the sample standard deviation; the calculation formulas for the sample population mean and sample standard deviation are as follows: Among them, n is the total number of sample data.
4. The method for predicting charging station load considering multiple influencing factors as claimed in claim 3, characterized in that: The feature engineering conversion includes natural factor feature conversion, user factor feature conversion, charging facility factor feature conversion, network and power factor feature conversion, management and scheduling factor feature conversion and data time alignment; Transform and encode the non-numerical features in the main factors, and convert the various factors into a form suitable for input into the LSTM model; Associate various data with the charging load data of the charging station to align the time of the two and establish a relationship model between each classification factor and the charging load; The natural factor characteristic conversion includes: The rainfall factor is converted into a binary feature, with rainfall set to 1 and no rainfall set to 0; Convert the holiday factor into a binary feature, setting holidays to 1 and non-holidays to 0; The user factor feature conversion includes: The charging location is represented as the number of the charging pile, and the charging location is encoded as a discrete category feature to consider the impact of different charging locations on the load; The time distribution of charging activities in a day is divided into different time periods, and different time periods are used as features to consider the impact of time on load; The charging facility factor feature conversion includes: Charging stations have different types of charging facilities, and unique hot encoding is used to convert different types of charging facilities into binary features; The network and power factor feature conversion includes: The ratio of the actual load of the charging station to the line capacity indicates the connection capacity with the power grid. The greater the line capacity utilization rate, the stronger the connection capacity with the power grid. At the same time, the voltage standard deviation over a period of time is calculated. The lower the standard deviation, the more stable the power grid is, and vice versa, the more unstable the power grid is. The time is divided into peak period and valley period, and the peak and valley electricity prices are encoded as features. When the time is in the valley period, the value is assigned to 0, and when it is in the peak period, the value is assigned to 1; The management and scheduling factor feature conversion includes: By calculating the standard deviation of the load data, a smaller standard deviation means a smaller load fluctuation, and the load fluctuation is used to indicate the smoothness of the load data; Because the operation modes of charging stations are different, we use unique hot encoding to convert these modes into binary features. The operation modes include self-operation and third-party operation. We set self-operation to 1 and third-party operation to 0. Calculate availability indicators based on equipment uptime and downtime; Time alignment of data includes, Based on the results of timestamp alignment and time range matching, other types of data are associated with charging load data; Use timestamp as primary key to join or merge two data sets, each timestamp has corresponding charging load data and other class data.
5. The method for predicting charging station load considering multiple influencing factors as claimed in claim 4, characterized in that: The data enhancement using the GAN model includes: Use the GAN model to enhance the influencing factors and load data and expand the training set; Define the network structure of the generator (G) and the discriminator (D). The input of the generator network G is the noise vector z, and the output is the synthesized charging load and influencing factor data. The input of the discriminator network D is real data or synthetic data output by the generator, and the output is the true or false classification result; Perform adversarial training on the generator G and the discriminator D, and iteratively train G and D alternately until the Nash equilibrium is reached; After the training is completed, the generator G is used to generate a large amount of synthetic charging load and influencing factor data; Combine the generated data with the original training data to expand the training set; In the case of missing values, the GAN model is used to generate a large amount of similar data to make up for the missing values; For the data generated by the GAN model, the absolute error and relative error are used to judge the accuracy of data processing; The calculation formula of the absolute error is as follows: Among them, AE is the absolute error, y(i) is the actual value, is the predicted value; The calculation formula of the relative error is as follows: Among them, RE is the relative error.
6. The method for predicting charging station load considering multiple influencing factors as claimed in claim 5, characterized in that: The use of selected feature data to establish an LSTM prediction model includes: The calculation formula of the LSTM prediction model is as follows: S t =f(U*x t +W*S t-1 ) O t =g(V / S t ) Among them, S t-1 , S t is the memory information data at different times, x t is the input data, O t is the data output, f and g are activation functions, W, U and V are the automatic learning weight matrices; LSTM prediction model parameter settings include: The load data of the charging station is matched with the features one by one, and one neuron is input into the input layer of the LSTM prediction model, and the input neuron is defined as a; The number of output neurons is related to the number of data modules in the output layer. There are 24 groups of sample points extracted in one day, and there are 24 output neurons. The 24 output neurons are defined as b; The calculation formula for the number of neurons in the hidden layer of the LSTM prediction model is as follows: Where, E is the number of neurons, η is the correction coefficient; Assign activation function, using Tanh, Relu and Sigmoid functions; The calculation formula of the Tanh function is as follows: The calculation formula of the ReLU function is as follows: g(z)=max(0,z) The calculation formula of the Sigmoid function is as follows: Among them, z is the weighted sum received by the neuron; Use historical data for model training and validation data for evaluation; The predictive performance of the evaluation model was evaluated using the error evaluation metric; The error evaluation indicators include mean absolute percentage error, mean absolute error and root mean square error. The error evaluation indicator calculation formula is as follows: in, is the predicted value, specifically y i is the true value, specifically y={y1,y2,…,y n }, n is the number of samples.
7. The method for predicting charging station load considering multiple influencing factors as claimed in claim 6, characterized in that: The final prediction by ensemble learning includes: Based on the five influencing factors of natural factors, user factors, charging facility factors, network and power factors, and management and scheduling factors, the LSTM model was used to predict them respectively, and the prediction outputs of five LSTM sub-models were obtained; Construct random forest input features, use the prediction output results of the five LSTM sub-models as the input features of the random forest model, and construct a 5-dimensional feature vector, including natural factor prediction values, user factor prediction values, charging facility prediction values, network and power prediction values, and management and scheduling prediction values; Train the random forest model and obtain multiple bootstrap samples from the original training data through the bootstrap sampling method; For each bootstrap sample, a decision tree model is trained. Some features are randomly selected for splitting during training, and multiple trained decision trees are combined into a random forest model. Perform integrated prediction, input the new sample to be predicted into the trained random forest model, and the random forest inputs the sample into each decision tree to obtain multiple prediction results; By voting, the predictions of multiple decision trees are combined to obtain the final integrated prediction output; Model evaluation and optimization: Use the test set to evaluate the predictive performance indicators of the random forest model and adjust the hyperparameters of the random forest based on the evaluation results. The random forest algorithm includes information entropy, conditional information entropy, information gain, and prediction error; The information entropy is used to measure the quality of feature segmentation, and the calculation formula is as follows: Where Y is a set of categories, which in this patent represents five different influencing factors, and p(y) is the probability of category y; The conditional information entropy is used to measure the quality index of the category probability after given feature segmentation, and the calculation formula is as follows: The information gain is used to measure the reduction of information entropy due to feature segmentation, and the calculation formula is as follows: IG(X,Y)=H(Y)-H(Y|X) The prediction error is an indicator used to measure the prediction performance of the model for test data, and is measured using square error or cross entropy.
8. A system using the charging station load prediction method considering multiple influencing factors as described in any one of claims 1 to 7, characterized in that: include: Data acquisition module, data preprocessing module, feature engineering module, data enhancement module, model training module, integrated prediction module; The data acquisition module is used to identify and classify the main influencing factors affecting the charging station load, and collect data on the main influencing factors; The data preprocessing module is used to preprocess the collected data of the main influencing factors The feature engineering module is used to transform and encode the non-numerical features in the main factors into a form suitable for input into the LSTM model, and to time-align the various data with the charging load data of the charging station to establish a relationship model between each classification factor and the charging load; The data enhancement module is used to enhance the influencing factors and load data using the GAN model to expand the training set; The model training module is used to establish and train the LSTM prediction model, input feature data, output prediction values, perform model training on historical data, and use verification data to evaluate model performance; The integrated prediction module is used to perform integrated prediction, input the samples to be predicted into the trained random forest model, and integrate the predictions of multiple decision trees to obtain the final prediction output.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the charging station load prediction method considering multiple influencing factors described in any one of claims 1 to 7 are implemented.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the charging station load prediction method considering multiple influencing factors described in any one of claims 1 to 7 are implemented.
Citation Information
Cited By
Substation metering system state prediction method and device and electronic equipment
CN120012979A
Substation metering system state prediction method, device and electronic equipment
CN120012979B
Method for constructing power load accurate prediction model by using big data
CN120031335A
Method and device for identifying influence factors of power grid admitting ability under polymorphic load
CN120216925A
Power grid regional load prediction system based on big data
CN120218307A
