An electric vehicle load forecasting method based on variational mode decomposition

Abstract

This invention discloses an electric vehicle (EV) load forecasting method based on variational mode decomposition (MODE), relating to the field of EV charging load forecasting technology. The method includes: collecting and preprocessing historical EV charging load data to generate an EV charging load dataset; generating candidate decomposed mode component sequences using a variational mode decomposition algorithm based on the EV charging load dataset, and selecting the optimal combination of variational mode decomposition parameters using a particle swarm optimization algorithm; decomposing the EV charging load dataset into decomposed mode component sequences using the optimal combination of variational mode decomposition parameters; and constructing a differential autoregressive moving average model using a time series probabilistic forecasting method based on the decomposed mode component sequences, and obtaining the probability distribution results of the predicted load values. This invention enhances the quantification of uncertainty in load forecasting, provides probabilistic support for power grid dispatching, and improves the reliability of forecasting.

Description

Technical Field

[0001] This invention relates to the field of electric vehicle charging load prediction technology, and in particular to an electric vehicle load prediction method based on variational mode decomposition. Background Technology

[0002] In the field of electric vehicle charging load forecasting, conventional methods mainly rely on time series analysis, machine learning, and traditional statistical models for load characteristic modeling and short-term forecasting. Existing technologies typically employ methods such as autoregressive integral moving average, support vector machines, or long short-term memory networks to train and predict based on historical charging power, time characteristics, and user behavior data, thereby supporting grid dispatching and energy management. These methods possess certain predictive capabilities under stationary sequences and explicit characteristic conditions and are widely used in power load forecasting and optimal control.

[0003] In the field of electric vehicle load forecasting based on variational mode decomposition, traditional methods often struggle to effectively decompose the multi-scale fluctuation characteristics implicit in load sequences, resulting in limited prediction accuracy for scenarios involving the coupling of sudden load changes and periodic fluctuations. Furthermore, most methods only provide deterministic point prediction results, lacking a quantitative assessment of prediction uncertainty, making it difficult to support risk-aware grid dispatching decisions. Especially in scenarios where complex user behavior is coupled with the external environment, prediction bias can easily affect the reliability of dispatching commands. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an electric vehicle load prediction method based on variational mode decomposition to solve the problems of insufficient multi-scale feature decomposition and lack of uncertainty quantification.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a method for predicting electric vehicle load based on variational mode decomposition, comprising: Collect historical data on electric vehicle charging load and preprocess it to generate an electric vehicle charging load dataset. Based on the electric vehicle charging load dataset, candidate decomposition mode component sequences are generated by variational mode decomposition algorithm, and the optimal parameter combination of variational mode decomposition is selected by particle swarm optimization algorithm. The electric vehicle charging load dataset is decomposed into a sequence of decomposed mode components by using the optimal parameter combination of variational mode decomposition. Based on the decomposed modal component sequence, a differential autoregressive moving average model is constructed using a time series probability prediction method, and the probability distribution of the predicted load value is obtained. Aggregate all predicted load value probability distribution results and perform random sampling through Monte Carlo simulation to generate a complete electric vehicle charging load forecast; The complete electric vehicle charging load forecast results are input into the hierarchical scheduling algorithm, which adjusts the charging power allocation according to user needs and outputs charging scheduling instructions.

[0007] As a preferred embodiment of the electric vehicle load prediction method based on variational mode decomposition described in this invention, the historical data of electric vehicle charging load includes charging power values, charging start and end timestamps, user identifiers, charging pile identifiers, and vehicle battery state of charge information.

[0008] As a preferred embodiment of the electric vehicle load prediction method based on variational mode decomposition described in this invention, the electric vehicle charging load dataset refers to the data generated by cleaning and removing invalid values ​​from historical electric vehicle charging load data and then normalizing the data.

[0009] As a preferred embodiment of the electric vehicle load prediction method based on variational mode decomposition described in this invention, the steps for generating candidate decomposed mode component sequences based on the electric vehicle charging load dataset using a variational mode decomposition algorithm are as follows: Based on the electric vehicle charging load dataset, a particle swarm algorithm is used to initialize the particle swarm and generate candidate parameter combinations. Based on the candidate parameter combinations, the electric vehicle charging load dataset is decomposed using a variational mode decomposition algorithm to generate candidate decomposed mode component sequences.

[0010] As a preferred embodiment of the electric vehicle load prediction method based on variational mode decomposition described in this invention, the step of selecting the optimal combination of variational mode decomposition parameters using a particle swarm optimization algorithm is as follows: The approximate signal is reconstructed using the candidate decomposed mode component sequences, and the reconstruction error between the reconstructed approximate signal and the electric vehicle charging load dataset is calculated. The fitness function is constructed based on the reconstruction error value, and the particle state is updated to select the optimal parameter combination for variational mode decomposition.

[0011] As a preferred embodiment of the electric vehicle load prediction method based on variational mode decomposition described in this invention, the steps of decomposing the electric vehicle charging load dataset into a sequence of decomposed mode components through the optimal parameter combination of variational mode decomposition are as follows: Based on the optimal parameter combination of variational mode decomposition, the electric vehicle charging load dataset is input into the variational mode decomposition algorithm, and the final time domain signal components are generated iteratively through the alternating direction multiplier method. The final time-domain signal components are converted into frequency-domain signals using Hilbert transform, and the amplitude and phase information are separated. Based on amplitude and phase information, the alternating direction multiplier method is driven to converge, generating a sequence of decomposed mode components.

[0012] As a preferred embodiment of the electric vehicle load prediction method based on variational mode decomposition described in this invention, the construction of the differential autoregressive moving average model refers to the construction by analyzing the autocorrelation characteristics of the decomposed mode component sequence through drawing autocorrelation plots and partial autocorrelation plots.

[0013] As a preferred embodiment of the electric vehicle load prediction method based on variational mode decomposition described in this invention, the steps for obtaining the probability distribution results of the predicted load value are as follows: Calculate the coefficients of the differential autoregressive moving average model and generate the conditional probability distribution parameters for future load values; Monte Carlo random sampling is performed on the conditional probability distribution parameters of future load values ​​to obtain the probability distribution results of predicted load values.

[0014] As a preferred embodiment of the electric vehicle load forecasting method based on variational mode decomposition described in this invention, the steps of aggregating the probability distribution results of all predicted load values ​​and generating a complete electric vehicle charging load forecast through Monte Carlo simulation are as follows. Based on the probability distribution of predicted load values, total load sample values ​​are generated through Monte Carlo simulation, and point prediction sequences and probability prediction sequences are generated through statistical analysis. Use the plotting library to draw point prediction sequences and probability prediction sequences as optional visualizations; Integrate point prediction sequences, probability prediction sequences, and optional visualization charts into a complete electric vehicle charging load forecast.

[0015] As a preferred embodiment of the electric vehicle load forecasting method based on variational mode decomposition described in this invention, the steps of inputting the complete electric vehicle charging load forecasting results into a hierarchical scheduling algorithm, adjusting the charging power allocation according to user demand, and outputting charging scheduling instructions are as follows. Collect the set of scheduling requests submitted by users, and combine them with the complete electric vehicle charging load forecast. Then, use mathematical programming methods to solve for the optimal matching scheme between grid constraints and user needs. Transform the optimal matching scheme into a pre-scheduling plan; Execute the pre-scheduled plan and monitor the actual load of charging piles, overlaying the data in optional visualization charts and outputting the updated optional visualization charts; Based on the updated optional visualization charts and probability prediction sequences, the final power value is generated through weight correction. Power values ​​are allocated based on the final power values, and charging scheduling instructions are generated.

[0016] The beneficial effects of this invention are as follows: by optimizing the key parameters of variational mode decomposition using the particle swarm optimization algorithm, the optimal parameter combination is selected, and the multi-scale fluctuation characteristics in the electric vehicle charging load sequence are accurately decomposed, thereby improving the accuracy and efficiency of the decomposition; by constructing a differential autoregressive moving average model for the decomposed mode component sequence using a time series probabilistic prediction method, the uncertainty quantification of load forecasting is enhanced, providing probabilistic support for power grid dispatching and improving the reliability of forecasting. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a load prediction method for electric vehicles based on variational mode decomposition.

[0019] Figure 2 This is a flowchart of data acquisition and preprocessing.

[0020] Figure 3 This is a flowchart of variational mode decomposition and parameter optimization.

[0021] Figure 4 This is a flowchart of load forecasting and scheduling. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4This is one embodiment of the present invention, which provides a method for predicting electric vehicle load based on variational mode decomposition, including the following steps: S1. Collect historical data on electric vehicle charging load and preprocess it to generate an electric vehicle charging load dataset. Historical data on electric vehicle charging load includes charging power values, charging start and end timestamps, user identifiers, charging pile identifiers, and vehicle battery state of charge information. Data cleaning and invalid value removal are performed on historical electric vehicle charging load data, and electric vehicle charging load dataset is generated through data normalization. Furthermore, based on historical data of electric vehicle charging load, missing values ​​are identified and filled using linear interpolation; outliers and values ​​outside the reasonable range are detected and removed using a standard deviation-based method; duplicate records are removed and format errors are corrected; and timestamps are aligned through data warping operations to generate an electric vehicle charging load dataset.

[0026] S2. Based on the electric vehicle charging load dataset, candidate decomposition mode component sequences are generated using the variational mode decomposition algorithm, and the optimal parameter combination for variational mode decomposition is selected using the particle swarm optimization algorithm. Based on the electric vehicle charging load dataset, a particle swarm algorithm is used to initialize the particle swarm and generate candidate parameter combinations. Furthermore, based on the electric vehicle charging load dataset, a particle swarm is initialized using a particle swarm optimization algorithm. The particle swarm optimization algorithm randomly generates a set of particles and randomly generates an initial position for each particle. At the same time, an initial velocity is randomly initialized for each particle, generating candidate parameter combinations.

[0027] It should be noted that the initial position is usually determined directly by the numerical values ​​of the candidate parameter combination; the initial velocity is usually set to zero and a small range of random values ​​(e.g., the range [-0.1, 0.1]).

[0028] Based on the candidate parameter combinations, the electric vehicle charging load dataset is decomposed using a variational mode decomposition algorithm to generate candidate decomposed mode component sequences. Furthermore, for each candidate parameter combination, a constrained variational problem (including the objective function and constraints) is defined using a variational mode decomposition algorithm. The constrained variational problem is solved iteratively using the alternating direction multiplier method, gradually approximating the optimal solution by alternately updating each modal component and Lagrange multiplier. In each iteration, the center frequency and bandwidth of each modal component are extracted using Hilbert transform, and the candidate time-domain signal components are converted into frequency-domain signals, separating amplitude and phase information, and driving the convergence of the alternating direction multiplier method. The relative error norm of all modal components before and after updating is obtained using the L2 norm calculation method. When the relative error norm is less than the convergence threshold (e.g., 1×10⁻⁶), the convergence is achieved. -3When convergence is determined, the candidate decomposed mode component sequence is output.

[0029] It should be noted that the objective function extracts the instantaneous frequency of each modal component through Hilbert transform and is set to minimize the sum of the bandwidths of all modal components. The constraint condition is set to the sum of the bandwidths of all modal components equal to the original signal. The convergence threshold is set based on the iteration limit of the particle swarm optimization algorithm and the problem complexity, and its value range is usually a continuous value of 10. -2 Up to 10 -4 .

[0030] The approximate signal is reconstructed using the candidate decomposed mode component sequences, and the reconstruction error between the reconstructed approximate signal and the electric vehicle charging load dataset is calculated. Furthermore, based on the candidate decomposed mode component sequences, the values ​​of all mode component sequences at each time point are summed by linear superposition to generate a reconstructed approximate signal. Then, the squared difference between the reconstructed signal value at each time point and the electric vehicle charging load dataset is calculated using the mean square error formula, and the average of the squared differences at all time points is obtained to output the reconstruction error value.

[0031] The fitness function is constructed based on the reconstruction error value, the particle state is updated, and the optimal parameter combination for variational mode decomposition is selected. Furthermore, based on the reconstruction error value and the learning factor preset by the particle swarm optimization algorithm, a fitness function is constructed, and the fitness value is obtained through reciprocal mapping. Using the particle swarm optimization algorithm, the current fitness value is compared with the individual's historical best fitness value and the group's historical best fitness value (if the current value is higher, the individual's historical best position or the group's historical best fitness value is updated). Based on the individual's historical best position and the group's historical best position, the particle's motion direction is adjusted through velocity update formulas and position update formulas. The global exploration is balanced through a decreasing inertia weight strategy, and higher fitness regions are iteratively searched. When the termination condition is met (the maximum number of iterations is reached or the change in fitness value is less than the change threshold), the candidate parameter combination represented by the particle with the highest fitness is selected as the optimal parameter combination for variational mode decomposition.

[0032] It should be noted that the learning factor is usually chosen as a symmetric value in the range [1.5, 2.5] during particle swarm optimization initialization. The learning factor includes individual learning factor and social learning factor. The maximum number of iterations is usually set based on processor performance, memory capacity, and particle swarm size, and is determined to be a fixed value through computational resource trade-offs, typically ranging from [50, 500]. The change threshold is set as a very small constant (e.g., 1 × 10⁻⁶) based on the algorithm's convergence accuracy requirements. -4 The iteration is terminated when the fitness value stabilizes during consecutive iterations, ensuring that the change in fitness value is negligible. The value range is typically a continuous interval [1×10].-3 1×10 -6 ].

[0033] The speed update formula is expressed as follows: ; in, For particle identifiers; This represents the number of iterations. For particles In the number of iterations The velocity vector at +1; This is the inertial weight, and its value range is usually a continuous value [0.4, 0.9]. For particles In the number of iterations The velocity vector at time; This is the individual learning factor, and its value range is usually a continuous value [1.5, 2.5]. The values ​​are random numbers (to increase the randomness of the search), and are usually uniformly distributed in the range of [0, 1]. For particles The individual's historical optimal position; For particles In the number of iterations Position vector at time; This is the social learning factor, and its value range is usually a continuous value [1.5, 2.5]. To be independent of The random numbers are usually uniformly distributed in the range [0, 1]. This represents the group's historically optimal position.

[0034] S3. Decompose the electric vehicle charging load dataset into a sequence of decomposed mode components by using the optimal parameter combination of variational mode decomposition. Based on the optimal parameter combination of variational mode decomposition, the electric vehicle charging load dataset is input into the variational mode decomposition algorithm, and the final time domain signal components are generated iteratively through the alternating direction multiplier method. Furthermore, based on the optimal parameter combination of variational mode decomposition, the electric vehicle charging load dataset is input into the variational mode decomposition algorithm; the constrained variational problem is solved iteratively by the alternating direction multiplier method, and the optimal solution is gradually approximated by alternatingly updating each modal component and Lagrange multiplier; the final time-domain signal component is output.

[0035] The final time-domain signal components are converted into frequency-domain signals using Hilbert transform, and the amplitude and phase information are separated. Furthermore, the Hilbert transform signal of the final time-domain signal component is obtained through Hilbert transform; a frequency-domain signal is constructed based on the original time-domain signal component and the Hilbert transform signal, and the amplitude information is extracted from the frequency-domain signal as the modulus of the frequency-domain signal, and the phase information is extracted as the argument of the frequency-domain signal.

[0036] It should be noted that the original time-domain signal components refer to the non-alternating updated modal component sequence generated iteratively by the alternating direction multiplier method.

[0037] Based on amplitude and phase information, the alternating direction multiplier method is driven to converge, generating a sequence of decomposed mode components; Furthermore, the bandwidth constraints of the modal components are adjusted based on the amplitude information, and the instantaneous frequency is obtained based on the time derivative of the phase information to drive the center frequency update; the alternating direction multiplier method updates the modal components and Lagrange multipliers alternately based on amplitude and phase information, gradually reducing the relative error norm before and after the update of all modal components; when the relative error norm is less than the minimum value (e.g., 1×10⁻⁶), the frequency is adjusted accordingly. -6 When the alternating direction multiplier method converges, the decomposed mode component sequence is output.

[0038] S4. Based on the decomposed modal component sequence, construct the differential autoregressive moving average model through the time series probability prediction method, and obtain the probability distribution results of the predicted load value. The autocorrelation characteristics of the decomposed modal component sequences are analyzed by plotting autocorrelation and partial autocorrelation plots, and a differential autoregressive moving average model is constructed. Furthermore, based on the decomposed modal component sequences, the autocorrelation function is obtained through the Pearson correlation coefficient formula, and the partial autocorrelation function is obtained through the recursive least squares method. Autocorrelation plots and partial autocorrelation plots are then plotted. The autocorrelation characteristics are analyzed based on the autocorrelation plots and partial autocorrelation plots. The autoregression order and moving average order are determined based on the partial autocorrelation plots and the cutoff points of the autocorrelation plots. The stationarity of the decomposed modal component sequences is judged through the ADF test, and a differential autoregressive moving average model is constructed.

[0039] It should be noted that the cutoff point refers to the critical lag order at which the correlation coefficient first falls into the confidence interval in the autocorrelation plot and partial autocorrelation plot, and the subsequent lag order does not fluctuate. The confidence interval is set in the autocorrelation plot and partial autocorrelation plot through statistical hypothesis testing, and its range usually changes dynamically with the sample size.

[0040] The coefficients of the differential autoregressive moving average model are calculated by maximum likelihood estimation, and the conditional probability distribution parameters of future load values ​​are generated by variance propagation. Furthermore, based on the residual sequences in the decomposed modal component sequences, the sum of squares of the residual sequences is calculated, along with the residual sequence length and residual variance. The log-likelihood function formula is used to integrate the residual sequence length, residual variance, and the ratio of the residual sum of squares to the residual variance to construct a likelihood function. The quasi-Newton method is used for iterative optimization to find the maximum value of the likelihood function, obtaining the coefficients of the differential autoregressive moving average model, and training the differential autoregressive moving average model. Based on the differential autoregressive moving average model coefficients, the conditional mean of the load values ​​at future time points is calculated using a linear recursive formula, and the conditional variance is calculated using the cumulative error variance formula, outputting the conditional probability distribution parameters of the future load values.

[0041] Monte Carlo random sampling is performed on the conditional probability distribution parameters of future load values ​​to obtain the probability distribution results of predicted load values; Furthermore, based on the conditional probability distribution parameters of future load values, a large number of random samples are generated for each time point using the Monte Carlo random sampling method, and random numbers are generated using the inverse transformation sampling method. The random numbers are scaled to the target distribution using the standard normal distribution quantile function to generate a sample set of predicted load values. Subsequently, the sample set of predicted load values ​​is analyzed using the kernel density estimation method and histogram statistics method (such as the Scott rule for selecting the bin width) to generate the probability distribution results of predicted load values.

[0042] S5. Aggregate the probability distribution results of all predicted load values ​​and perform random sampling through Monte Carlo simulation to generate a complete electric vehicle charging load forecast; Based on the probability distribution of predicted load values, total load sample values ​​are generated through Monte Carlo simulation, and point prediction sequences and probability prediction sequences are generated through statistical analysis. Furthermore, based on the probability distribution results of the predicted load values, the total load sample values ​​are aggregated and generated through Monte Carlo simulation. The steps are as follows: a synchronous random number seed is generated for each time point, and sample values ​​are synchronously extracted from the probability distribution results of the predicted load values ​​through inverse transformation sampling. The total load sample values ​​are obtained through the cross-modal summation formula. This process is repeated to generate a large number of total load sample values. The arithmetic mean of the total load sample value set at each time point is calculated through statistical analysis to generate a point prediction sequence, and fixed quantiles (e.g., 5% and 95%) are calculated to generate a probability prediction sequence.

[0043] Use the plotting library to draw point prediction sequences and probability prediction sequences as optional visualizations; Furthermore, the point prediction sequence can be plotted as a time-load value line chart using a plotting library (e.g., the horizontal axis is the timestamp sequence and the vertical axis is the point prediction value sequence), and the upper and lower bounds of the confidence interval of the probability prediction sequence can be plotted as semi-transparent banded areas and superimposed on the line chart to generate optional visualization charts.

[0044] Integrate point prediction sequences, probability prediction sequences, and optional visualization charts into a complete electric vehicle charging load forecast; Furthermore, based on point prediction sequences, probability prediction sequences, and optional visualization charts, the point prediction sequences and probability prediction sequences are stored in a structured data format through a data encapsulation method, and the optional visualization charts are integrated into the structured data format as embedded resources to generate a complete electric vehicle charging load forecast.

[0045] S6. Input the complete electric vehicle charging load prediction results into the hierarchical scheduling algorithm, adjust the charging power allocation according to user needs, and output charging scheduling instructions.

[0046] Collect the set of scheduling requests submitted by users, and combine them with the complete electric vehicle charging load forecast. Then, use mathematical programming methods to solve for the optimal matching scheme between grid constraints and user needs. Furthermore, the system collects a set of scheduling requests submitted by users (including user identifiers, charging pile identifiers, charging start and end timestamps, and user-requested electricity amounts). Combined with a complete electric vehicle charging load forecast, the objective function is set to minimize the total grid operating cost using mathematical programming methods. Constraints are integrated, including grid power balance constraints, transformer capacity constraints, line current carrying capacity constraints, and user charging demand constraints. A solver is invoked to analyze the objective function and constraints. Mathematical algorithms, including the interior-point method, simplex method, and branch and bound method, are used for iterative calculations. The system searches the space satisfying all constraints to find the optimal variable values ​​that allow the objective function to reach its optimal state, i.e., the optimal solution to the objective function, thus generating the optimal matching scheme between grid constraints and user demand.

[0047] It should be noted that mathematical programming methods refer to translating the power grid dispatch problem (including user demand, grid security, and cost minimization) into a mathematical framework by defining variables, setting objective functions, and listing constraints. Grid power balance constraints originate from the engineering application of Kirchhoff's current law, requiring the total charging power (such as the total power of charging piles) to match the transformer output capacity in real time. Transformer capacity constraints are based on the thermal stability limits of power equipment, setting peak power limits. Line current carrying capacity constraints are derived from the conductor temperature rise equation, converting the current safety range into power constraints. User charging demand constraints are determined by the user service agreement terms, with the core being to meet the requested electricity volume. Solver refers to specialized software tools used to automatically solve the mathematical framework, such as CPLEX and Gurobi.

[0048] The optimal matching scheme is transformed into a pre-scheduling plan using a structured mapping method; Furthermore, the variable values ​​in the optimal matching scheme are parsed, and data mapping is performed based on the predefined plan template. The variable values ​​in the optimal matching scheme are filled into the corresponding fields of the predefined plan template to generate a structured data table. Then, the data integrity is verified through format validation rules, and the pre-scheduled plan is output.

[0049] It should be noted that the plan template is based on the power grid dispatch protocol and charging pile parameter settings, including the charging pile identifier field, time window field, and power value field.

[0050] Execute the pre-scheduled plan and monitor the actual load of charging piles, overlaying the data in optional visualization charts and outputting the updated optional visualization charts; Furthermore, based on the charging pile identifier field, time window field, and power value field in the pre-scheduled plan, a power allocation instruction is sent to the charging pile controller; then, the actual load of the charging pile is monitored in real time through smart meters, and the instantaneous power value at each time point is collected (e.g., collected once every 15 minutes) to generate an actual load sequence; the actual load sequence is overlaid on an optional visualization chart in scatter plot form, and the data points of the actual load sequence are aligned with the time axis of the optional visualization chart to output an updated optional visualization chart.

[0051] Based on the updated optional visualization charts and probability prediction sequences, the final power value is generated through weight correction. Furthermore, based on the updated optional visualization chart and the probability prediction sequence, the deviation between the data points of the actual load sequence in the updated optional visualization chart and the corresponding confidence intervals of the probability prediction sequence is calculated. Based on the deviation, a correction weight is generated using an exponentially weighted moving average algorithm. The correction weight and the predicted values ​​in the point prediction sequence are adjusted using a linear interpolation formula to generate a corrected power value sequence as the final power value.

[0052] Power values ​​are allocated based on the final power values, and charging scheduling instructions are generated.

[0053] Furthermore, based on the charging pile identifier and time window field in the scheduling request set, the final power value is mapped to the corresponding charging pile identifier and time period to generate a charging scheduling instruction. The instruction format follows a predefined plan template and is sent to the charging pile controller for power allocation via a communication protocol.

[0054] This embodiment also provides a computer device applicable to the electric vehicle load prediction method based on variational mode decomposition, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the electric vehicle load prediction method based on variational mode decomposition as proposed in the above embodiment.

[0055] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0056] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the electric vehicle load prediction method based on variational mode decomposition as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0057] In summary, this invention achieves the following: by optimizing the key parameters of variational mode decomposition using the particle swarm optimization algorithm and selecting the optimal parameter combination, it accurately decomposes the multi-scale fluctuation characteristics in the electric vehicle charging load sequence, improving the accuracy and efficiency of the decomposition; and by constructing a differential autoregressive moving average model for the decomposed mode component sequence using a time series probabilistic prediction method, it enhances the quantification of uncertainty in load forecasting, provides probabilistic support for power grid dispatching, and improves the reliability of forecasting.

[0058] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for predicting electric vehicle load based on variational mode decomposition, characterized in that: include, Collect historical data on electric vehicle charging load and preprocess it to generate an electric vehicle charging load dataset. Based on the electric vehicle charging load dataset, candidate decomposition mode component sequences are generated by variational mode decomposition algorithm, and the optimal parameter combination of variational mode decomposition is selected by particle swarm optimization algorithm. The electric vehicle charging load dataset is decomposed into a sequence of decomposed mode components by using the optimal parameter combination of variational mode decomposition. Based on the decomposed modal component sequence, a differential autoregressive moving average model is constructed using a time series probability prediction method, and the probability distribution of the predicted load value is obtained. Aggregate all predicted load value probability distribution results and perform random sampling through Monte Carlo simulation to generate a complete electric vehicle charging load forecast; The complete electric vehicle charging load forecast results are input into the hierarchical scheduling algorithm, which adjusts the charging power allocation according to user needs and outputs charging scheduling instructions.

2. The electric vehicle load prediction method based on variational mode decomposition as described in claim 1, characterized in that: The historical data of electric vehicle charging load includes charging power values, charging start and end timestamps, user identifiers, charging pile identifiers, and vehicle battery state of charge information.

3. The electric vehicle load prediction method based on variational mode decomposition as described in claim 2, characterized in that: The electric vehicle charging load dataset refers to the dataset generated by cleaning and removing invalid values ​​from historical electric vehicle charging load data and then normalizing the data.

4. The electric vehicle load prediction method based on variational mode decomposition as described in claim 3, characterized in that: Based on the electric vehicle charging load dataset, candidate decomposed mode component sequences are generated using a variational mode decomposition algorithm. The steps are as follows: Based on the electric vehicle charging load dataset, a particle swarm algorithm is used to initialize the particle swarm and generate candidate parameter combinations. Based on the candidate parameter combinations, the electric vehicle charging load dataset is decomposed using a variational mode decomposition algorithm to generate candidate decomposed mode component sequences.

5. The electric vehicle load prediction method based on variational mode decomposition as described in claim 4, characterized in that: The steps for selecting the optimal parameter combination for variational mode decomposition using the particle swarm optimization algorithm are as follows: The approximate signal is reconstructed using the candidate decomposed mode component sequences, and the reconstruction error between the reconstructed approximate signal and the electric vehicle charging load dataset is calculated. The fitness function is constructed based on the reconstruction error value, and the particle state is updated to select the optimal parameter combination for variational mode decomposition.

6. The electric vehicle load prediction method based on variational mode decomposition as described in claim 5, characterized in that: The process of decomposing the electric vehicle charging load dataset into a sequence of decomposed mode components using the optimal parameter combination through variational mode decomposition is as follows: Based on the optimal parameter combination of variational mode decomposition, the electric vehicle charging load dataset is input into the variational mode decomposition algorithm, and the final time domain signal components are generated iteratively through the alternating direction multiplier method. The final time-domain signal components are converted into frequency-domain signals using Hilbert transform, and the amplitude and phase information are separated. Based on amplitude and phase information, the alternating direction multiplier method is driven to converge, generating a sequence of decomposed mode components.

7. The electric vehicle load prediction method based on variational mode decomposition as described in claim 6, characterized in that: The construction of the differential autoregressive moving average model refers to the process of analyzing the autocorrelation characteristics of the decomposed modal component sequences by drawing autocorrelation and partial autocorrelation plots.

8. The electric vehicle load prediction method based on variational mode decomposition as described in claim 7, characterized in that: The steps for obtaining the probability distribution of the predicted load value are as follows: Calculate the coefficients of the differential autoregressive moving average model and generate the conditional probability distribution parameters for future load values; Monte Carlo random sampling is performed on the conditional probability distribution parameters of future load values ​​to obtain the probability distribution results of predicted load values.

9. The electric vehicle load prediction method based on variational mode decomposition as described in claim 8, characterized in that: The process involves aggregating the probability distribution results of all predicted load values ​​and performing random sampling through Monte Carlo simulation to generate a complete electric vehicle charging load forecast. The steps are as follows: Based on the probability distribution of predicted load values, total load sample values ​​are generated through Monte Carlo simulation, and point prediction sequences and probability prediction sequences are generated through statistical analysis. Use the plotting library to draw point prediction sequences and probability prediction sequences as optional visualizations; Integrate point prediction sequences, probability prediction sequences, and optional visualization charts into a complete electric vehicle charging load forecast.

10. The electric vehicle load prediction method based on variational mode decomposition as described in claim 9, characterized in that: The steps for inputting the complete electric vehicle charging load forecast results into the hierarchical scheduling algorithm, adjusting the charging power allocation according to user demand, and outputting charging scheduling instructions are as follows: Collect the set of scheduling requests submitted by users, and combine them with the complete electric vehicle charging load forecast. Then, use mathematical programming methods to solve for the optimal matching scheme between grid constraints and user needs. Transform the optimal matching scheme into a pre-scheduling plan; Execute the pre-scheduled plan and monitor the actual load of charging piles, overlaying the data in optional visualization charts and outputting the updated optional visualization charts; Based on the updated optional visualization charts and probability prediction sequences, the final power value is generated through weight correction. Power values ​​are allocated based on the final power values, and charging scheduling instructions are generated.

Images