Power distribution network electric vehicle charging load prediction method and system

Abstract

The invention discloses a power distribution network electric vehicle charging load prediction method, and belongs to the technical field of urban power grids. The method comprises the following steps: according to a coordinate set of charging stations in a region, historical charging loads of the charging stations and vehicle position data, in combination with a Thiessen polygon method, calculating charging load data of time-space distribution, and performing preprocessing; performing self-adaptive decomposition on the preprocessed charging load data by adopting a self-adaptive empirical mode decomposition algorithm based on a dynamic average envelope and an energy entropy change rate, and obtaining charging load data of a historical charging station after decomposition; according to the charging load data of the historical charging station, constructing a CycleNet model; and inputting actual charging load data into the constructed CycleNet model to predict the charging load of the charging station, thereby realizing the charging load prediction of the electric vehicle in the power distribution network. According to the method, refined representation of the load space-time coupling characteristic is realized, the component decomposition accuracy of the non-stationary charging load signal is improved, and the charging load prediction accuracy is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of urban power grids, and more specifically, relates to a method and system for predicting the charging load of electric vehicles in a distribution network. Background Art

[0002] With the rapid development of electric vehicles, the spatio-temporal distribution characteristics of the charging load have an increasingly significant impact on the operation of the distribution network. Traditional charging load prediction methods have low geographical visualization, often isolate the spatial characteristics from the time characteristics, resulting in the spatio-temporal interaction relationship not being fully captured, and not fully considering spatio-temporal coupling factors such as the geographical location of charging stations and the dynamic distribution of vehicles, resulting in limited prediction accuracy and low geographical visualization of load spatio-temporal coupling, which is not conducive to subsequent geographical visualization such as the planning and site selection of charging stations and the optimization of service areas.

[0003] In the prior art, the charging load data often has noise interference, non-stationarity and abnormal fluctuations, and the traditional empirical mode decomposition algorithm uses a fixed threshold for mode decomposition, making it difficult to adaptively adjust the envelope fitting parameters, resulting in unclear physical meanings of the decomposed components and difficult to effectively extract the internal characteristics of the load.

[0004] In addition, conventional prediction models (such as LSTM, ARIMA) have insufficient ability to co-model periodic characteristics and non-linear residual patterns, and cannot adapt to the complex dynamic changes of the charging load. Summary of the Invention

[0005] To solve the deficiencies in the prior art, the present invention provides a method and system for predicting the charging load of electric vehicles in a distribution network. Based on the charging station coordinate set in the region, the historical charging load data of the charging stations, and the vehicle position data in the region, combined with the Thiessen polygon method, calculate the charging load data considering spatio-temporal distribution; use an adaptive empirical mode decomposition algorithm based on dynamic average envelope and energy entropy change rate to adaptively decompose the preprocessed charging load data, and obtain the historical charging load data of the charging stations after decomposition; construct a CycleNet model according to the historical charging load data of the charging stations; input the actual charging load data into the constructed CycleNet model to predict the charging load of the charging stations, and realize the prediction of the charging load of electric vehicles in the distribution network. It realizes the refined characterization of the load spatio-temporal coupling characteristics and improves the decomposition accuracy of the components of the non-stationary charging load signal; the constructed CycleNet model solves the defect of insufficient joint modeling of the periodic component and the residual component of the charging load of the electric vehicle charging station by the traditional model, and significantly improves the prediction accuracy.

[0006] The present invention adopts the following technical solutions.

[0007] The first aspect of the present invention provides a method for predicting the charging load of electric vehicles in a distribution network, specifically including: Based on the charging station coordinate set within the region, the historical charging load data of the charging stations, and the vehicle position data within the region, combined with the Thiessen polygon method, calculate the charging load data considering spatio-temporal distribution; Preprocess the charging load considering spatio-temporal distribution; Use the adaptive empirical mode decomposition algorithm based on dynamic average envelope and energy entropy change rate to adaptively decompose the preprocessed charging load data, and obtain the charging load data of historical charging stations after decomposition; Construct a CycleNet model according to the charging load data of historical charging stations; Input the actual charging load data into the constructed CycleNet model for predicting the charging load of charging stations, and realize the prediction of electric vehicle charging load in the distribution network.

[0008] Preferably, the calculating the charging load data considering spatio-temporal distribution based on the charging station coordinate set within the region, the historical charging load data of the charging stations, and the vehicle position data within the region, combined with the Thiessen polygon method, specifically includes: Obtain the charging station coordinate set within the region, the historical charging load data of the charging stations, and the vehicle position data within the region; Solve the Thiessen polygons corresponding to all charging stations in the charging station coordinate set within the region; Overlay the vehicle position data within the region with the solved Thiessen polygons to obtain the number of vehicles within each Thiessen polygon; Solve the average charging power per vehicle according to the historical charging load data of the charging stations; Solve the reference load of the charging station according to the number of vehicles within each Thiessen polygon and the average charging power per vehicle, and multiply the reference load of the charging station by the time weight to solve and obtain the charging load data considering spatio-temporal distribution.

[0009] Preferably, the solving the Thiessen polygons corresponding to all charging stations in the charging station coordinate set within the region specifically includes: Solve all points within the region whose distance to charging station s i is less than or equal to the distance to other charging stations, and obtain the Thiessen polygons corresponding to all charging stations.

[0010] Preferably, the solving the reference load of the charging station according to the number of vehicles within each Thiessen polygon and the average charging power per vehicle, and multiplying the reference load of the charging station by the time weight to solve and obtain the charging load data considering spatio-temporal distribution includes: According to the number of vehicles s i within the Thiessen polygon of charging station N i and the average charging power of a single vehicle at time t r(t) multiply to obtain the charging stations i Base load at time t Q i (t) ; Construct a time weight according to the ratio of the historical average load at time t to the average load over the entire period; For the charging station s i Base load at time t Q i (t) Multiply by the time weight to obtain the charging load with spatio-temporal distribution.

[0011] Preferably, the preprocessing of the charging load considering spatio-temporal distribution specifically includes: Use the moving average method to zero-mean the charging load considering spatio-temporal distribution to obtain zero-mean data; Replace the outliers in the zero-mean data with the mean values of adjacent time periods; Smooth the noise of the replaced data and perform data normalization to obtain the preprocessed charging load data.

[0012] Preferably, the adaptive empirical mode decomposition algorithm based on dynamic average envelope and energy entropy change rate is used to adaptively decompose the preprocessed charging load data, specifically including: Initialize the residual signal according to the preprocessed charging load data, detect the extreme values of the current residual signal and correct the extreme values; Set the charging pile online rate as the coefficient of the corrected extreme value, perform B-spline envelope fitting on the corrected extreme value, and solve the dynamic average envelope; Set the difference between the current residual signal and the dynamic average envelope as the candidate component, and solve the normalized root mean square error and the energy entropy change rate; Set the condition that both the standard deviation threshold and the energy entropy change rate are less than the corresponding set thresholds as the stopping criterion. If the candidate component meets the stopping criterion, set the current candidate component as the nth IMF component. Otherwise, re-perform dynamic envelope fitting, update the candidate component, and repeat the steps until the stopping criterion is met; Judge whether the residual signal satisfies a monotonic function. If not, return the re-generated residual signal and perform cyclic decomposition of the residual signal until the residual signal satisfies a monotonic function, terminate the decomposition, and obtain the decomposed historical charging load data set.

[0013] Preferably, the detecting the extreme values of the current residual signal and correcting the extreme values specifically includes: Detect the extreme values of the current residual signal. If the maximum value exceeds the rated power of the charging pile, correct the maximum value to the upper limit value of the device. If the minimum value is negative, set the minimum value to zero.

[0014] Preferably, constructing a CycleNet model based on the charging load data of historical charging stations specifically includes: Normalize the input charging load data of historical charging stations and set the historical window length L in , and the number of future time steps L to be predicted out ; Set a learnable periodic function , where C is initialized to all zeros and L represents the period length; Circularly expand the learnable periodic function according to a vector of all 1s with length L out to obtain the periodic component C repeat ; Combine the periodic component C repeat , and use MLP to perform residual prediction on the charging load data of historical charging stations; Based on the residual prediction result and the periodic component C repeat perform prediction synthesis to obtain the charging load prediction result; Solve the mean square error based on the predicted result of the charging load and the actual charging load value, set the mean square error as the loss function, and combine the optimizer to train the learnable periodic function C and the residual network parameters θ in the residual prediction process to solve the optimal parameters and construct the CycleNet model.

[0015] Preferably, combining the periodic component C repeat , and using MLP to perform residual prediction on the charging load data of historical charging stations specifically includes: Normalize the charging load data x of historical charging stations history , and intercept the last L history data x in x in [-L history :], where L in is the historical window length, and set L in to be an integer multiple of L; in Calculate the residual between the last L history data x in x in [-L history :] and the periodic component C in ; repeat According to the calculated residual result, construct a two-layer MLP residual network to solve the residual prediction result. According to the calculated residual result, construct a two-layer MLP residual network to solve the residual prediction result.

[0016] The second aspect of the present invention provides a distribution network electric vehicle charging load prediction system that runs a distribution network electric vehicle charging load prediction method, including: ​A charging load acquisition module, which is used to calculate the charging load data considering spatio-temporal distribution according to the charging station coordinate set in the area, the historical charging load data of the charging station, and the vehicle position data in the area, in combination with the Thiessen polygon method; A preprocessing module, which is used to preprocess the charging load considering spatio-temporal distribution; An adaptive decomposition module, which is used to adaptively decompose the preprocessed charging load data by using an adaptive empirical mode decomposition algorithm based on dynamic average envelope and energy entropy change rate, and obtain the charging load data of historical charging stations after decomposition; A model construction module, which is used to construct a CycleNet model according to the charging load data of historical charging stations; A load prediction module, which is used to input the actual charging load data into the constructed CycleNet model to predict the charging load of the charging station, and realize the prediction of the electric vehicle charging load of the distribution network.

[0017] Compared with the prior art, the beneficial effects of the present invention at least include: By combining the Thiessen polygon method with vehicle position data, dynamically divide the service range of charging stations, accurately quantify the impact of regional vehicle distribution on the load, solve the problem of rough spatial distribution modeling in traditional methods, and combine time weight correction to achieve refined characterization of the spatio-temporal coupling characteristics of the load. The Thiessen polygon method has a high degree of visualization, which is conducive to subsequent geographical visualization optimizations such as charging station planning and site selection and service range optimization; Adopt dynamic average envelope fitting based on the online rate of charging piles and the energy entropy change rate criterion, adaptively adjust the extreme value correction and envelope generation process of empirical mode decomposition, effectively suppress mode mixing, and improve the decomposition accuracy of the components of non-stationary charging load signals; The constructed CycleNet model explicitly models the load periodic characteristics through a learnable periodic function, and combines the MLP residual network to capture non-linear residual patterns, solves the defect of insufficient joint modeling of the periodic component and the residual component of the charging load of electric vehicle charging stations in traditional models, and significantly improves the prediction accuracy. Description of the Drawings

[0018] Figure 1 is a schematic diagram of a prediction process for the electric vehicle charging load of a distribution network provided according to an embodiment of the present invention; Figure 2 is a schematic diagram of a process for constructing a charging load with spatio-temporal characteristics by the Thiessen polygon method provided according to an embodiment of the present invention; Figure 3 is a schematic diagram of an adaptive empirical mode decomposition algorithm process provided according to an embodiment of the present invention. Detailed Embodiments

[0019] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. The described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the spirit of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0020] As Figure 1 shown, Embodiment 1 of the present invention provides a method for predicting the charging load of electric vehicles in a distribution network, which specifically includes: Step 1, as Figure 2 shown, based on the charging station coordinate set in the region, the historical charging load data of the charging stations, and the vehicle position data in the region, combined with the Thiessen polygon method, calculate the charging load data considering the spatio-temporal distribution.

[0021] In a preferred but non-limiting embodiment of the present invention, Step 1 includes: Step 1.1, solve the Thiessen polygons corresponding to all charging stations in the charging station coordinate set in the region; Further preferably, Step 1.1 includes: (1) In the formula, V i represents the set of all points in the region whose distance to charging station s i is less than or equal to the distance to other charging stations, p represents a point in the region, indicating that p is an arbitrary point in the two-dimensional plane (region), represents the two-dimensional Euclidean space, d(p, s i ) represents the distance from a point in the region to charging station s i , d(p, s j ) represents the distance from a point in the region to the remaining charging station s j .

[0022] Step 1.2, overlay the vehicle position data in the region with the Thiessen polygons solved in Step 1.1 to obtain the number of vehicles in each Thiessen polygon; Step 1.3, solve the average charging power per vehicle based on the historical charging load data of the charging stations. The average charging power per vehicle is expressed by the following formula: (2) In the formula, r(t) represents the average charging power of a single vehicle at time t, L i (t) Represents the load on the charging station s i at time t, Σ L i (t) represents the sum of the loads of all charging stations at time t, N veh represents the total number of vehicles.

[0023] Step 1.4: According to the number of vehicles in each Thiessen polygon obtained in Step 1.2 and the average charging power per vehicle obtained in Step 1.3, solve for the reference load of the charging station, multiply it by the historical charging load data of the charging station, and solve for the charging load data considering the spatio-temporal distribution.

[0024] Further preferably, Step 1.4 includes: Step 1.4.1: According to the charging station s i the number of vehicles within the Thiessen polygon of N i and the average charging power per vehicle at time t r(t) multiply to obtain the reference load of the charging station s i at time t Q i (t) , Step 1.4.2: Construct a time weight according to the ratio of the historical average load at time t to the average load over the entire period; Step 1.4.3: Multiply the reference load of the charging station s i at time t Q i (t) by the time weight to obtain the charging load considering the spatio-temporal distribution.

[0025] Step 2: Preprocess the charging load considering the spatio-temporal distribution obtained in Step 1.

[0026] In a preferred but non-limiting embodiment of the present invention, Step 2 includes: Step 2.1: Use the moving average method to zero-mean the charging load considering the spatio-temporal distribution to obtain zero-mean data; Step 2.2: Replace the outliers in the zero-mean data obtained in Step 2.1 with the mean values of adjacent time periods; Step 2.3: Smooth the replaced data for noise and perform data normalization to obtain the preprocessed charging load data.

[0027] Step 3, as Figure 3 shown, the charging load data preprocessed in Step 2 is adaptively decomposed by an adaptive empirical mode decomposition algorithm based on the dynamic average envelope and the energy entropy change rate to obtain the charging load data of the historical charging station.

[0028] In a preferred but non-limiting embodiment of the present invention, Step 3 includes: Step 3.1, initialize the residual signal according to the preprocessed charging load data, detect the extreme values of the current residual signal and correct the extreme values.

[0029] Further preferably, Step 3.1 includes: Step 3.1.1, initialize the residual signal according to the preprocessed charging load data, which is expressed by the following formula: (4) In the formula, r0(t) represents the initialized residual signal, x prep (t) represents the preprocessed charging load data.

[0030] Step 3.1.2, detect the extreme values of the current residual signal r n (t) and correct the extreme values. If the maximum value exceeds the rated power of the charging pile, correct the maximum value to the upper limit value of the device. If the minimum value is negative (for example, but not limited to being caused by digital noise), set the minimum value to zero.

[0031] Step 3.2, set the charging pile online rate as the coefficient of the corrected extreme value, and perform B-spline envelope fitting according to the extreme values obtained in Step 3.1 to obtain the dynamic average envelope P env (t) , which is expressed by the following formula: (5) In the formula, α(t) represents the charging pile online rate at time t, U(t) represents the charging power peak envelope generated by connecting the maximum value points, that is, the upper envelope; L(t) represents the charging power valley envelope generated by connecting the minimum value points, that is, the lower envelope.

[0032] Step 3.3, set the difference between the current residual signal and the dynamic average envelope as the new candidate component, and solve the normalized root mean square error and the energy entropy change rate.

[0033] Further preferably, Step 3.3 includes: Step 3.3.1, set the difference between the current residual signal and the dynamic average envelope as the candidate component h(t), which is expressed by the following formula: (6) In the formula, r n (t) represents the current residual signal.

[0034] Step 3.3.2, solve the normalized root mean square error and the energy entropy change rate according to the candidate component in Step 3.3.1. The normalized root mean square error is expressed by the following formula: (7) In the formula, SD k represents the normalized root mean square error, h k (t) represents the candidate component at the k-th iteration.

[0035] The energy entropy change rate is expressed by the following formula: (8) In the formula, represents the energy entropy change rate, H k represents the energy entropy at the k-th iteration, represents the energy entropy of the candidate component, represents time The proportion of the local energy at to the total energy of the candidate component, represents the value of the current candidate component at time , T represents the total length of the time series.

[0036] Step 3.4, set that both the standard deviation threshold and the energy entropy change rate are less than the corresponding set thresholds as the stopping criterion. If the candidate component meets the stopping criterion, set the current candidate component as the n-th IMF component IMF n (t), otherwise, re-perform the dynamic envelope fitting, update the candidate component, and repeat the steps until the stopping criterion is met; Step 3.5, according to the n-th IMF component IMF n (t) obtained in Step 3.4, update the residual signal, which is expressed by the following formula: (9) Determine whether the residual signal satisfies a monotonic function. If not, return the regenerated residual signal to step 3.12 for cyclic decomposition of the residual signal until the residual signal satisfies the monotonic function, terminate the decomposition, and obtain the charging load data of the historical charging station.

[0037] Step 4: Construct a CycleNet model based on the charging load data of the historical charging station.

[0038] In a preferred but non-limiting embodiment of the present invention, step 4 includes: Step 4.1: Normalize the input charging load data of the historical charging station, and set the historical window length L in , and the number of future time steps L out .

[0039] Step 4.2: Set a learnable periodic function , initialized to all zeros, where L represents the period length, set to 24 hours for a daily period, i.e., L = 24, represents a real matrix of length L, represents the set of real numbers.

[0040] Step 4.3: Generate a periodic component C out aligned with the prediction window length L repeat through cyclic extension of the learnable periodic function in step 4.2, expressed by the following formula: (10) In the formula, C repeat represents the periodic component, 1 Lout represents a vector of all 1s of length L out , L out represents the prediction window length, set to an integer multiple of L, and used to represent the number of future time steps that the model expects to predict.

[0041] Step 4.4: Combine the periodic component C repeat in step 4.3, and use MLP to perform residual prediction on the charging load data of the historical charging station to obtain the residual prediction result.

[0042] Further preferably, step 4.4 includes: Step 4.4.1: After normalizing the charging load data x history of the historical charging station, intercept the last L history data x in in the charging load data x history [-L in :], L inis the historical window length, used to represent the number of past time steps input to the model, set as L in is an integer multiple of L. For example, but not limited to, when L = 24, L in = 96; Step 4.4.2, calculate the charging load data x of the historical charging station history in the last L in data x history [-L in :] and the periodic component C repeat of the residual, expressed by the following formula: (11) In the formula, x res represents the residual, x history [-L in :] represents the last L history data in the charging load data x of the historical charging station in , C repeat [-L in :] represents the last L repeat data in the periodic component C in .

[0043] Step 4.4.3, according to the calculated residual results, construct a two-layer MLP residual network, expressed by the following formula: (12) In the formula, f θ (x res ) represents the residual prediction result, represents the residual network parameters, , W2 represents the linear transformation matrix that maps the hidden layer to the predicted residual, ReLU represents the activation function, x res represents the residual, W1 represents the linear transformation matrix that maps the residual to the hidden layer, b1 represents the activation threshold of the neurons in the hidden layer, b2 represents the baseline offset of the output layer.

[0044] Step 4.5, based on the residual prediction result and the periodic component C in Step 4.4 repeat perform prediction synthesis, expressed by the following formula: (13) In the formula, ypred Represents the predicted value of the charging load.

[0045] Step 4.6: Solve the mean square error based on the predicted value of the charging load and the actual charging load value obtained in Step 4.5, set the mean square error as the loss function, and combine the optimizer AdamW to train the learnable periodic function C and the residual network parameters to obtain the optimal parameters and construct the CycleNet model.

[0046] Step 5: Input the actual charging load data into the CycleNet model constructed in Step 4 to predict the charging load of the charging station, and realize the prediction of the electric vehicle charging load of the distribution network.

[0047] Obtain the data of an electric bus charging station in a certain city, including N charging piles and M buses. Collect the charging load data and location data of the electric buses every 15 minutes, and collect 96 data within 24 hours a day. Randomly select the charging load and location data of a certain day of the electric bus charging station, and use the BP neural network, wavelet neural network and the present invention for prediction. The MAPE prediction accuracies of the BP neural network, wavelet neural network and the present invention are 4.87%, 1.97% and 0.99% respectively. The predicted values of the BP neural network prediction results fluctuate greatly, and the ability to capture peaks and valleys is weak. The wavelet neural network still has a delay during the fast-changing period. The prediction results of the present invention are significantly better than those of the BP neural network and the wavelet neural network.

[0048] Embodiment 2 of the present invention provides a distribution network electric vehicle charging load prediction system, which runs the distribution network electric vehicle charging load prediction method described in Embodiment 1, including: A charging load acquisition module, configured to calculate the charging load data considering the spatio-temporal distribution based on the charging station coordinate set in the region, the historical charging load data of the charging station, and the vehicle location data in the region, in combination with the Thiessen polygon method; A preprocessing module, configured to preprocess the charging load considering the spatio-temporal distribution; An adaptive decomposition module, configured to adaptively decompose the preprocessed charging load data by using an adaptive empirical mode decomposition algorithm based on the dynamic average envelope and the energy entropy change rate, and obtain the historical charging load data of the charging station after decomposition; A model construction module, configured to construct a CycleNet model according to the historical charging load data of the charging station; A load prediction module, configured to input the actual charging load data into the constructed CycleNet model to predict the charging load of the charging station, and realize the prediction of the electric vehicle charging load of the distribution network.

[0049] Compared with the prior art, the beneficial effects of the present invention at least include: By using the Thiessen polygon method in combination with vehicle position data, the service range of charging stations is dynamically divided, the impact of regional vehicle distribution on load is accurately quantified, the problem of rough spatial distribution modeling in traditional methods is solved, and combined with time weight correction, a refined characterization of the spatio-temporal coupling characteristics of the load is achieved. The Thiessen polygon method has a high degree of visualization, which is conducive to subsequent geographical visualization optimizations such as charging station planning and site selection and service range optimization; By using dynamic average envelope fitting based on the online rate of charging piles and the criterion of the change rate of energy entropy, the extreme value correction and envelope generation process of empirical mode decomposition are adaptively adjusted, effectively suppressing mode mixing and improving the decomposition accuracy of the components of non-stationary charging load signals; The constructed CycleNet model explicitly models the periodic characteristics of the load through a learnable periodic function, and combines the MLP residual network to capture non-linear residual patterns, solving the defect of insufficient joint modeling of the periodic component and the residual component of the charging load of electric vehicle charging stations in traditional models, and significantly improving the prediction accuracy.

[0050] This disclosure may be a system, method, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions thereon for causing a processor to implement various aspects of this disclosure.

[0051] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: the specific implementation manners of the present invention can still be modified or equivalently replaced, and any modification or equivalent replacement that does not depart from the spirit and scope of the present invention shall be covered by the protection scope of the claims of the present invention.

Claims

1. A method for predicting the charging load of electric vehicles in a distribution network, characterized in that: Based on the charging station coordinate set in the region, the historical charging load data of the charging stations, and the vehicle position data in the region, combined with the Thiessen polygon method, calculate the charging load data considering the spatio-temporal distribution; Preprocess the charging load considering the spatio-temporal distribution; Use the adaptive empirical mode decomposition algorithm based on the dynamic average envelope and the energy entropy change rate to adaptively decompose the preprocessed charging load data. After decomposition, obtain the charging load data of historical charging stations, specifically including: initialize the residual signal based on the preprocessed charging load data, detect the extreme values of the current residual signal and correct the extreme values; Set the charging pile online rate as the coefficient of the corrected extreme value, perform B-spline envelope fitting on the corrected extreme value, and solve the dynamic average envelope; Set the difference between the current residual signal and the dynamic average envelope as the candidate component, and solve the normalized root mean square error and the energy entropy change rate; Set the condition that both the standard deviation threshold and the energy entropy change rate are less than the corresponding set thresholds as the stop criterion. If the candidate component meets the stop criterion, set the current candidate component as the nth IMF component. Otherwise, re-perform the dynamic envelope fitting, update the candidate component, and repeat the steps until the stop criterion is met; Judge whether the residual signal meets the monotonic function. If not, return the re-generated residual signal and perform cyclic decomposition of the residual signal until the residual signal meets the monotonic function, terminate the decomposition, and obtain the set of decomposed historical charging load data; Construct a CycleNet model based on the charging load data of historical charging stations; Input the actual charging load data into the constructed CycleNet model to predict the charging load of the charging stations, and realize the prediction of the charging load of electric vehicles in the distribution network.

2. A method for predicting the charging load of electric vehicles in a distribution network according to claim 1, characterized in that: The calculation of the charging load data considering the spatio-temporal distribution based on the charging station coordinate set in the region, the historical charging load data of the charging stations, and the vehicle position data in the region, combined with the Thiessen polygon method, specifically includes: Solve the Thiessen polygons corresponding to all charging stations in the charging station coordinate set in the region; Overlay the vehicle position data in the region with the solved Thiessen polygons to obtain the number of vehicles in each Thiessen polygon; Solve the average charging power per vehicle based on the historical charging load data of the charging stations; Solve the reference load of the charging station according to the number of vehicles in each Thiessen polygon and the average charging power per vehicle, and multiply the reference load of the charging station by the time weight to solve the charging load data considering the spatio-temporal distribution.

3. A method for predicting the charging load of electric vehicles in a distribution network according to claim 2, characterized in that: The solution of the Thiessen polygons corresponding to all charging stations in the charging station coordinate set in the region specifically includes: The set of points within the solution region whose distances to the charging station s i are less than or equal to their distances to other charging stations, to obtain the Thiessen polygons corresponding to all charging stations.

4. A method for predicting the charging load of electric vehicles in a distribution network according to claim 1, characterized in that: Solving the benchmark load of the charging station according to the number of vehicles in each Thiessen polygon and the average charging power of a single vehicle, and multiplying the benchmark load of the charging station by the time weight to solve the charging load data considering the spatio-temporal distribution, including: According to the vehicle count within the Thiessen polygon of the charging station s i and the average charging power of a single vehicle at time t N i multiply them to obtain the reference load of the charging station r(t) at time t s i Q i (t) ;​ Constructing the time weight according to the ratio of the historical average load at time t to the average load over the entire period; The charging station s i The reference load at time t Q i (t) is multiplied by the time weight to obtain the charging load with spatio-temporal distribution.

5. A method for predicting the charging load of electric vehicles in a distribution network according to claim 1, characterized in that: Preprocessing the charging load considering the spatio-temporal distribution, specifically including: Using the moving average method to zero-mean the charging load considering the spatio-temporal distribution to obtain zero-mean data; Replacing the outliers in the zero-mean data with the mean values of adjacent time periods; Smoothing the noise of the replaced data and normalizing the data to obtain the preprocessed charging load data.

6. A method for predicting the charging load of electric vehicles in a distribution network according to claim 5, characterized in that: Detecting and correcting the extreme values of the current residual signal, specifically including: Detecting the extreme values of the current residual signal. If the maximum value exceeds the rated power of the charging pile, correcting the maximum value to the equipment upper limit value. If the minimum value is negative, setting the minimum value to zero.

7. A method for predicting the charging load of electric vehicles in a distribution network according to claim 1, characterized in that: Constructing a CycleNet model according to the charging load data of historical charging stations, specifically including: Normalize the charging load data of the input historical charging stations and set the historical window length L in , and the expected number of future time steps L to be predicted out ; Set a learnable periodic function , where C is initialized to all zeros and L represents the period length; According to the all-ones vector of length L out circularly extend the learnable periodic function to obtain the periodic component C repeat ; Combined with the periodic component C repeat , the residual prediction of the charging load data of historical charging stations is carried out by using MLP; Based on the residual prediction results and the periodic component C repeat Perform prediction synthesis to obtain the charging load prediction results; Solving the mean square error based on the prediction result of the charging load and the actual charging load value, setting the mean square error as the loss function, and training the learnable periodic function C and the residual network parameters θ in the residual prediction process in combination with the optimizer to solve the optimal parameters and construct the CycleNet model.

8. A method for predicting the charging load of electric vehicles in a distribution network according to claim 7, characterized in that: The combined periodic component C repeat , and the residual prediction of the charging load data of the historical charging stations is performed using MLP, specifically including: Normalize the charging load data x of historical charging stations history and intercept the last L history data x in in x history [-L in :], where L in is the historical window length, and L in is set to be an integer multiple of L; Calculate x history The last L in data in x history [-L in :] and the periodic component C repeat of the residual; Constructing a two-layer MLP residual network according to the calculated residual result to obtain the residual prediction result.

9. A system for predicting the charging load of electric vehicles in a distribution network, running a method for predicting the charging load of electric vehicles in a distribution network according to any one of claims 1-8, characterized in that: A charging load acquisition module, configured to calculate the charging load data considering the spatio-temporal distribution according to the charging station coordinate set in the region, the historical charging load data of the charging station, and the vehicle position data in the region, in combination with the Thiessen polygon method; A preprocessing module, configured to preprocess the charging load considering the spatio-temporal distribution; An adaptive decomposition module, configured to adaptively decompose the preprocessed charging load data by using an adaptive empirical mode decomposition algorithm based on the dynamic average envelope and the energy entropy change rate, and obtain the charging load data of historical charging stations after decomposition; A model construction module, configured to construct a CycleNet model according to the charging load data of historical charging stations; A load prediction module, configured to input the actual charging load data into the constructed CycleNet model to predict the charging load of the charging station, and realize the prediction of the charging load of electric vehicles in the distribution network.

Images