Multi-time-scale elastic response and rolling optimization regulation and control method for electric vehicle

Abstract

The invention discloses a multi-time-scale elastic response and rolling optimization regulation and control method for an electric vehicle, and relates to the technical field of electric vehicles. Comprising the steps of obtaining historical travel data of the electric vehicle based on a Monte Carlo random sampling method, and generating a historical travel data set; generating a wind-light power prediction result by adopting the wind-light power prediction model; according to the historical travel data set, establishing a dual-target model based on the micro-grid operation cost and the vehicle owner charging cost, and obtaining a basic regulation and control strategy in combination with a wind-solar power prediction result; setting a rolling time window, dynamically adjusting an energy storage charging and discharging power boundary and an electric vehicle charging and discharging interval based on ultra-short-term load prediction data, and correcting a basic regulation and control strategy in real time; and solving the corrected optimization problem by adopting a parallel optimization algorithm, and forming a multi-time scale regulation and control strategy by minimizing the deviation caused by the prediction error. The method can reduce the influence of prediction errors, and improves the operation stability and wind and light absorption rate of the micro-grid.

Description

Technical Field

[0001] This invention relates to the field of electric vehicle technology, and more specifically to a method for multi-timescale elastic response and rolling optimization control of electric vehicles. Background Technology

[0002] With the rapid development of the new energy vehicle industry, the number of electric vehicles on the road continues to rise, providing a large amount of controllable resources for microgrid dispatch. Electric vehicles offer flexibility in charging and discharging, and through vehicle-to-grid (V2G) interaction technology, they can participate in microgrid dispatch, effectively mitigating fluctuations in wind and solar power generation and improving the stability and economy of microgrid operation. However, existing electric vehicle dispatch methods still have many shortcomings: First, electric vehicle travel behavior is characterized by randomness and volatility. Existing methods often rely on simple statistical approaches to process travel data, resulting in insufficient representativeness of the sampling samples. This leads to a disconnect between control strategies and actual travel demand, making it difficult to protect the rights of car owners. Second, wind and solar power generation is significantly affected by meteorological factors, resulting in large prediction errors. Single prediction models are insufficient to meet the accuracy requirements of basic control strategies, thus impacting control effectiveness. Third, existing control strategies often employ single-time-scale scheduling, lacking dynamic correction mechanisms. This fails to effectively offset the impact of short-term prediction errors and load fluctuations, resulting in poor control flexibility and adaptability. Fourth, control objectives often focus on the economic efficiency of microgrid operation, neglecting the interests of car owners such as charging costs and battery degradation. This leads to low participation from car owners and makes it difficult to achieve a balance of interests among all parties.

[0003] Therefore, in view of the shortcomings of the existing technology, how to provide a method for multi-timescale elastic response and rolling optimization control of electric vehicles is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] In view of this, the present invention provides a method for multi-timescale elastic response and rolling optimization control of electric vehicles, which takes into account data reliability, prediction accuracy, multi-timescale adaptability and balance of interests of all parties, reduces the impact of prediction error and improves the operational stability of microgrids and the wind and solar power absorption rate.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a method for multi-timescale elastic response and rolling optimization control of electric vehicles, comprising: Historical travel data of electric vehicles were obtained using the Monte Carlo random sampling method to generate a historical travel dataset. Wind and solar power prediction results are generated using a wind and solar power prediction model. Based on the historical travel dataset, a dual-objective model based on microgrid operating costs and vehicle owner charging costs is established, and a basic control strategy is obtained by combining wind and solar power prediction results. A rolling time window is set up, and the energy storage charging and discharging power boundary and the electric vehicle charging and discharging range are dynamically adjusted based on ultra-short-term load forecast data to make real-time corrections to the basic control strategy. Parallel optimization algorithms are used to solve the corrected optimization problem, and a multi-timescale control strategy is formed by minimizing the deviation caused by the prediction error.

[0006] Preferably, a time-series decomposition algorithm is integrated with an improved LSTM neural network wind and solar power prediction model; Historical wind and solar power data are collected to generate a prediction base dataset, and the prediction base dataset is preprocessed. The preprocessed wind and solar power data is decomposed into trend, periodic and random components by empirical mode decomposition, thus separating the fluctuation characteristics at different scales. Predict the wind and solar power sequences for the trend term, periodic term, and random term respectively; A weighted fusion strategy is adopted, which assigns weights based on the reciprocal of the prediction error of each component, and merges the prediction results of each component to obtain the wind and solar power prediction curve.

[0007] Preferably, the objective function for the microgrid operating cost is constructed as follows: ; in, The formula for calculating the cost of wind and solar power curtailment is as follows: ; Cost per unit of power curtailment loss Let be the predicted photovoltaic power at time t. Let t be the predicted wind power. Let t be the power that the microgrid can absorb. The formula for calculating the cost of energy storage charging and discharging losses is as follows: ; The unit cost of battery loss. Let t be the energy storage charging power. Let be the energy storage discharge power at time t. For energy storage charging efficiency, For energy storage discharge efficiency; The formula for calculating the grid interaction price cost is as follows: ; Let t be the power purchased by the power grid. Let t be the electricity purchase price. Let t be the power output of the power grid at time t. Let t be the electricity price at time t; The formula for calculating the cost of standby capacity is as follows: ; As a unit of reserve cost, (t) represents the microgrid's reserve power at time t.

[0008] Preferably, the objective function for vehicle owner charging costs is constructed, including: ; in, The formula for calculating the electricity cost for charging is as follows: ; N represents the number of electric vehicles involved in the regulation. (t) represents the charging power of the i-th vehicle at time t. (t) represents the charging electricity price at time t. For time intervals; The battery wear and tear cost is calculated based on the depth of charge / discharge, cycle life, and initial battery cost. The formula is as follows: ; Let i be the battery cycle life of the i-th vehicle after discharge via vehicle-to-grid interaction. Let be the battery cycle life of the i-th vehicle during only one trip. Let $\frac{i}{i}$ be the initial battery cost for the $i$-th vehicle. Let be the discharge power of the i-th vehicle at time t. Let be the discharge duration of the i-th vehicle; The formula for calculating the equivalent cost of charging waiting time is as follows: ; The charging wait time for the i-th vehicle. Let be the time value coefficient for the owner of the i-th vehicle.

[0009] Preferably, based on the objective function and constraints, a genetic algorithm is used to solve the optimization problem and generate basic control strategies, including: Set algorithm parameters, including population size, number of iterations, crossover probability, and mutation probability; Using real-number encoding, each chromosome corresponds to a set of electric vehicle charging and discharging power and energy storage charging and discharging power control schemes; The fitness function is the reciprocal of the transformed single objective function value. Using the roulette wheel selection method, individuals with higher fitness have a greater probability of being selected; A single-point crossover method was used, where crossover points were randomly selected to exchange partial genes between two individuals. Gaussian mutation is used to randomly perturb the genes of an individual; After iterative convergence, the optimal control scheme is output as the basic control strategy.

[0010] Preferably, the gradient boosting tree algorithm is used for ultra-short-term load forecasting; Historical load data, real-time load data, meteorological data, date type, and time period characteristics are selected as input features, and redundant features are eliminated through Pearson correlation coefficient analysis. The model hyperparameters were optimized using a grid search method. The load forecast value within the rolling window is output based on the forecast period; The model parameters are updated using actual load data to optimize the forecasting results.

[0011] Based on the deviation between ultra-short-term load forecast data and day-ahead forecast data, the energy storage charging and discharging power boundaries and electric vehicle charging and discharging ranges are dynamically adjusted.

[0012] Preferably, a parallel optimization algorithm is used to solve the corrected optimization problem. By minimizing the deviation caused by the prediction error, a multi-time-scale control strategy is formed, including: The revised optimization problem is decomposed into two sub-problems: the microgrid side and the electric vehicle user side. An improved particle swarm optimization algorithm is adopted, and an adaptive adjustment strategy for inertial weights is introduced to solve the microgrid side. A genetic algorithm is used to solve the problem on the user side of electric vehicles through adaptive crossover and mutation probability; After the iteration is complete, the optimization results from both sides are merged to obtain the global optimal solution; By integrating basic control strategies with dynamic optimization results, a multi-timescale control strategy is obtained.

[0013] Preferably, the microgrid side sub-problem aims to minimize operating costs and prediction errors by optimizing energy storage charging and discharging power and grid interaction power; The electric vehicle user-side problem aims to minimize the owner's charging costs and meet travel constraints by optimizing the charging and discharging time and power of a single vehicle.

[0014] As can be seen from the above technical solution, compared with the prior art, this invention discloses a multi-timescale elastic response and rolling optimization control method for electric vehicles. Based on Monte Carlo random sampling combined with Latin hypercube sampling optimization, it generates a historical travel dataset of electric vehicles that conforms to actual operating patterns. It employs a combined model integrating time-series decomposition and an improved LSTM neural network to generate high-precision wind and solar power prediction results. It constructs a dual-objective model of microgrid operating costs and vehicle owner charging costs, and combines the wind and solar power prediction results to generate a 24-hour basic control strategy. It dynamically adjusts control parameters through a rolling time window to correct deviations in the basic strategy. A hybrid parallel optimization algorithm is used to solve the corrected optimization problem, forming a three-level multi-timescale control strategy of "day-to-day-real-time". This invention can effectively reduce the impact of prediction errors, improve the operational stability of microgrids and the wind and solar power absorption rate, and has strong practicality and promotional value. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0016] Figure 1 This invention provides a schematic flowchart of a multi-timescale elastic response and rolling optimization control method for electric vehicles. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] This invention discloses a method for multi-timescale elastic response and rolling optimization control of electric vehicles, comprising: Historical travel data of electric vehicles were obtained using the Monte Carlo random sampling method to generate a historical travel dataset. Wind and solar power prediction results are generated using a wind and solar power prediction model. Based on the historical travel dataset, a dual-objective model based on microgrid operating costs and vehicle owner charging costs is established, and a basic control strategy is obtained by combining wind and solar power prediction results. A rolling time window is set up, and the energy storage charging and discharging power boundary and the electric vehicle charging and discharging range are dynamically adjusted based on ultra-short-term load forecast data to make real-time corrections to the basic control strategy. Parallel optimization algorithms are used to solve the corrected optimization problem, and a multi-timescale control strategy is formed by minimizing the deviation caused by the prediction error.

[0019] Specifically, historical travel data of electric vehicles is obtained based on the Monte Carlo random sampling method. By simulating travel and charging behaviors in multiple scenarios and removing abnormal data, a historical travel dataset that conforms to actual operating patterns is generated.

[0020] The historical travel dataset covers three dimensions: scenario characteristics, vehicle parameters, and data attributes, ensuring the comprehensiveness and usability of the data. Scenario data: Identify city types (first-tier cities / second-tier cities / third-tier cities and below), quantify road network density (unit: km / km²), statistically analyze charging facility coverage (ratio of public charging piles to private charging piles), divide climate zones (cold temperate / subtropical / tropical, etc.), and differentiate time periods (weekdays / weekends / holidays) to provide a basis for differentiated regulation of scenarios.

[0021] Vehicle basic parameters: Record electric vehicle model (small car / mid-size car / SUV / new energy truck, etc.), indicate battery capacity (unit: kWh), nominal range (unit: km), charging power level (fast charging ≥30kW / slow charging ≤7kW), define the initial battery range (0-20% / 20%-50% / 50%-80% / 80%-100%), and match the adjustment potential of different vehicles.

[0022] Basic data attributes: The dataset sample size is set to be no less than 10,000 groups, and the data time granularity supports 1 minute or 5 minutes. Data fields include key information such as travel time, origin latitude and longitude, destination latitude and longitude, driving distance (unit: km), driving speed (unit: km / h), charging time (unit: min), charging amount (unit: kWh), remaining power (unit: kWh), and charging facility type to meet multi-dimensional analysis needs.

[0023] Determining the probability distribution of key parameters

[0024] Based on the statistical characteristics of survey data, public datasets, and road test data, the probability distribution type of each travel parameter is verified through the KS test, and the mean μ and coefficient of variation CV are clarified to ensure that the sampling data conforms to actual operating patterns. Trip triggering parameters: The number of daily trips follows a normal distribution N(2.3, 0.5²) with a coefficient of variation of 0.22; the time of the first trip follows a Weibull distribution W(6.2, 2.1) with a coefficient of variation of 0.34; the time of the last trip follows a normal distribution N(18.5, 1.2²) with a coefficient of variation of 0.06; the duration of the trip interval follows a log-normal distribution lnN(2.1, 0.6²) with a coefficient of variation of 0.29.

[0025] Trip status parameters: Single trip distance follows a log-normal distribution lnN(3.2,0.8²), with a coefficient of variation of 0.25; the speed distribution is divided into congested (≤20km / h), normal (20-60km / h), and smooth (≥60km / h) conditions, with proportions of 30%, 55%, and 15%, respectively; the trip duration follows a normal distribution N(25,8²), with a coefficient of variation of 0.32; the number of stops follows a Poisson distribution P(0.8), with a coefficient of variation of 1.12; and the stop duration follows an exponential distribution E(15), with a coefficient of variation of 1.0.

[0026] Monte Carlo random sampling implementation

[0027] Based on the probability distribution model and statistical characteristics of each parameter, Monte Carlo random sampling is used to generate a specific sample size of travel behavior parameters. The sampling process optimizes the uniformity of sample distribution through Latin hypercube sampling to ensure coverage of travel characteristics across multiple scenarios. 1. Initialize sampling parameters: Set the sample size to 10,000 groups, and the sampling dimensions include all the above-mentioned travel trigger parameters and travel status parameters; 2. Distribution Sampling: For parameters of different distribution types, corresponding sampling methods are adopted. For normal distribution, the inverse transformation method is used; for log-normal distribution, sampling is performed after logarithmic transformation; for Weibull distribution, the inverse operation of the distribution function is used; and for Poisson distribution, the recursive method is used. 3. Sample optimization: The distribution range of each parameter is divided into 10,000 equally probable intervals by Latin hypercube sampling, and one sample point is drawn from each interval to ensure that the sample is evenly covered across the entire distribution range. 4. Data cleaning: Remove abnormal samples (such as driving distance exceeding twice the range or charging time being negative), and finally retain no less than 9,800 valid samples to form a historical travel dataset that conforms to actual operating patterns.

[0028] Specifically, based on the obtained historical travel dataset, a dual-objective model based on microgrid operating costs and vehicle owner charging costs is established. Combined with wind and solar power prediction results, and taking into account constraints such as electric vehicle charging and discharging power limits, battery SOC safety margin, and travel electricity demand, a basic control strategy with a 24-hour control cycle is generated, taking into account both economy and feasibility.

[0029] To offset the impact of wind and solar power and load forecasting errors on the basic control strategy, a rolling time window is set with a window length of 1-4 hours and a rolling step of 15-30 minutes. Based on ultra-short-term load forecasting data, the boundary of energy storage charging and discharging power and the charging and discharging range of electric vehicles are dynamically adjusted, with a prediction accuracy of no less than 95%, so as to realize the real-time adaptation of control parameters and correct the deviation of the basic strategy.

[0030] Parallel optimization algorithms are used to solve the corrected optimization problem. By minimizing the deviation caused by prediction error, the matching degree between the system's equivalent load and the day-ahead plan is maximized. Finally, the basic control strategy and dynamic optimization results are integrated to form a multi-time-scale control strategy of basic and rolling, ensuring the stable and efficient operation of the microgrid.

[0031] Specifically, the time-series decomposition algorithm is integrated with an improved LSTM neural network model for wind and solar power prediction. Historical wind and solar power data are collected to generate a prediction base dataset, and the prediction base dataset is preprocessed. The preprocessed wind and solar power data is decomposed into trend, periodic and random components by empirical mode decomposition, thus separating the fluctuation characteristics at different scales. Predict the wind and solar power sequences for the trend term, periodic term, and random term respectively; A weighted fusion strategy is adopted, which assigns weights based on the reciprocal of the prediction error of each component, and merges the prediction results of each component to obtain the wind and solar power prediction curve.

[0032] Specifically, the wind and solar power prediction model is used to generate wind and solar power prediction results. The specific implementation steps are as follows: Data preprocessing 1. Data collection: Collect historical wind and solar power data (15-minute time granularity) and corresponding meteorological data (irradiance, wind speed, wind direction, temperature, humidity, and air pressure) over the past 3 years to build a basic dataset for prediction; 2. Noise Removal: Random noise in the data is removed using the moving average method (window size set to 5), and missing values ​​are filled using linear interpolation to ensure data continuity; 3. Data normalization: Z-score normalization is used to map wind and solar power data and meteorological data to a unified normalized space to avoid the impact of differences in units on model training.

[0033] Empirical Mode Decomposition (EMD) is used to decompose the wind and solar power series into trend, periodic, and stochastic components, separating the fluctuation characteristics at different scales. 1. Trend Items: Reflect the long-term changing patterns of wind and solar power, such as seasonal trends; 2. Periodic items: These reflect the periodic fluctuations within a day or week, such as the daily peak-valley changes in photovoltaic power and the weekly stability fluctuations in wind power. 3. Random term: Characterizes power fluctuations caused by random meteorological fluctuations, such as power changes caused by sudden cloud cover, gusts, etc.

[0034] Itemized Forecasts

[0035] 1. Trend prediction: Multinomial fitting is used for prediction. The trend prediction curve is obtained by fitting the trend data through the least squares method. The fitting order is adaptively selected according to the data characteristics (2nd-4th order). 2. Periodic Term Prediction: The seasonal autoregressive integral moving average (SARIMA) model is adopted. The model parameters are set according to the time period characteristics of the periodic term (24 hours in a day / 7 days in a week) to capture the periodic fluctuation pattern. 3. Random Term Prediction: An improved LSTM neural network is adopted, and an attention mechanism is introduced to enhance the model's ability to capture data in key time periods. The network structure is set as follows: input layer dimension = number of meteorological features + number of historical power features, hidden layers are set to 3 layers (the number of neurons in each layer is 128, 64 and 32 respectively), output layer dimension = 1, ReLU is used as the activation function, Adam is used as the optimizer, the learning rate is set to 0.001, and the number of iterations is 100.

[0036] Results fusion

[0037] A weighted fusion strategy is adopted, and weights are assigned according to the reciprocals of the prediction errors of each component. The prediction results of each component are then fused to obtain the 24-hour wind and solar power prediction curve. 1. Error Calculation: The root mean square error (RMSE) is used to calculate the prediction errors of the trend term, periodic term, and random term respectively; 2. Weighting: The weighting coefficients are inversely proportional to the prediction error; that is, the smaller the error, the greater the weight. The weighting calculation formula is as follows: The formula for calculating the weight is: ; in, Let be the weight of the i-th component. Let be the prediction error of the i-th component; The combined wind and solar power prediction values ​​are: P pred =w1P trend +w2P cycle +w3P random ; Among them, P trend P is the predicted value of the trend term. cycle P is the predicted value of the periodic term. random These are the predicted values ​​for the random term, with the prediction error controlled within ±8%.

[0038] Specifically, a wind and solar power prediction model is used to generate wind and solar power prediction results, providing dual data support for subsequent strategy formulation and improving prediction accuracy and stability.

[0039] Specifically, the objective function for the operating cost of the microgrid is constructed as follows: ; in, The formula for calculating the cost of wind and solar power curtailment is as follows: ; Cost per unit of power curtailment loss Let be the predicted photovoltaic power at time t. Let t be the predicted wind power. Let t be the power that the microgrid can absorb. The formula for calculating the cost of energy storage charging and discharging losses is as follows: ; The unit cost of battery loss. Let t be the energy storage charging power. Let be the energy storage discharge power at time t. For energy storage charging efficiency, For energy storage discharge efficiency; The formula for calculating the grid interaction price cost is as follows: ; Let t be the power purchased by the power grid. Let t be the electricity purchase price. Let t be the power output of the power grid at time t. Let t be the electricity price at time t; The formula for calculating the cost of standby capacity is as follows: ; As a unit of reserve cost, (t) represents the microgrid's reserve power at time t.

[0040] Specifically, construct the objective function for vehicle owner charging costs, including: ; in, The formula for calculating the electricity cost for charging is as follows: ; N represents the number of electric vehicles involved in the regulation. (t) represents the charging power of the i-th vehicle at time t. (t) represents the charging electricity price at time t. For time intervals; The battery wear and tear cost is calculated based on the depth of charge / discharge, cycle life, and initial battery cost. The formula is as follows: ; Let i be the battery cycle life of the i-th vehicle after discharge via vehicle-to-grid interaction. Let be the battery cycle life of the i-th vehicle during only one trip. Let $\frac{i}{i}$ be the initial battery cost for the $i$-th vehicle. Let be the discharge power of the i-th vehicle at time t. Let be the discharge duration of the i-th vehicle; The formula for calculating the equivalent cost of charging waiting time is as follows: ; The charging wait time for the i-th vehicle. Let be the time value coefficient for the owner of the i-th vehicle.

[0041] Specifically, dual-objective transformation

[0042] The weights of the two objectives, namely the microgrid operating cost, are determined using the analytic hierarchy process (AHP). Weighting of charging costs for car owners satisfy The weight adjustment range is 0.3-0.7, dynamically adapted according to the microgrid operation scenario (such as peak grid electricity prices). When the value is 0.7, and car owner participation is low. Taking 0.7), the bi-objective problem is transformed into a single-objective optimization problem: ; Constraint Setting Microgrid power balance constraints ; in, Let be the total charging power of the electric vehicle at time t. Let t be the total discharge power of the electric vehicle. Let t be the total load power of the microgrid at time t.

[0043] Distribution network security constraints

[0044] Node voltage constraints: All node voltages are controlled within the range of 0.95-1.05 times the rated voltage; Line transmission power constraint: The transmission power of each line shall not exceed 80% of the rated capacity to avoid line overload.

[0045] Electric vehicle operating constraints

[0046] State of charge (SOC) constraints: , This represents the minimum permissible state of charge for the i-th vehicle (usually taken as 20%). This represents the maximum permissible state of charge (usually taken as 100%) for the i-th vehicle. Charge and discharge power constraints: , , The maximum charging power for the i-th vehicle. The maximum discharge power of the i-th vehicle; Travel demand constraint: Predict the travel time of vehicle i based on historical travel dataset. and driving distance Ensure that SOC is met before departure: , Let be the energy consumption per unit mileage of vehicle i (kWh / km).

[0047] Energy storage operation constraints

[0048] Capacity constraints: , This represents the maximum energy storage capacity (kWh). Let t be the energy storage capacity; Charge and discharge power constraints: , , This is the maximum charging power for energy storage. This represents the maximum discharge power of the energy storage. Power change rate constraint: , To avoid power surges affecting system stability.

[0049] Specifically, based on the objective function and constraints, a genetic algorithm is used to solve the optimization problem, generating a basic regulatory strategy with a 24-hour control cycle, including: 1. Set algorithm parameters, including population size, number of iterations, crossover probability, and mutation probability; The population size was set to 100, the number of iterations was 200, the crossover probability was 0.8, and the mutation probability was 0.05. 2. Encoding method: Real number encoding is used, with each chromosome corresponding to a set of 24-hour electric vehicle charging and discharging power and energy storage charging and discharging power control schemes; 3. Fitness function: The fitness function is the reciprocal of the transformed single-objective function value. ; 4. Selection operation: A roulette wheel selection method is used, and individuals with higher fitness have a greater probability of being selected; 5. Crossover operation: Single-point crossover is used, and a crossover point is randomly selected to exchange part of the genes of two individuals; 6. Mutation operation: Gaussian mutation is used to perform small-amplitude random perturbations on the individual genes; 7. Strategy Output: After iterative convergence, the optimal control scheme is output as the basic control strategy. The number of electric vehicles participating in the control, the charging / discharging power allocation, and the charging / discharging state of the energy storage are clearly defined at each time step.

[0050] Specifically, the gradient boosting tree algorithm is used for ultra-short-term load forecasting; Historical load data, real-time load data, meteorological data, date type, and time period characteristics are selected as input features, and redundant features are eliminated through Pearson correlation coefficient analysis. The model hyperparameters were optimized using a grid search method. The load forecast value within the rolling window is output based on the forecast period; The model parameters are updated using actual load data to optimize the forecasting results.

[0051] Based on the deviation between ultra-short-term load forecast data and day-ahead forecast data, the energy storage charging and discharging power boundaries and electric vehicle charging and discharging ranges are dynamically adjusted.

[0052] Specifically, to offset the impact of wind and solar power and load forecasting errors on the basic control strategy, a rolling time window is set up, and the control parameters are dynamically adjusted based on ultra-short-term load forecasting data to achieve real-time correction of the basic strategy.

[0053] Scrolling time window settings

[0054] Window length: Adjustable from 1 to 4 hours based on microgrid response characteristics, with a default value of 2 hours; Rolling step: Set to 15-30 minutes to ensure a rapid response to prediction errors; the default value is 15 minutes. Window update mechanism: After each rolling step, the window slides forward, incorporating the latest ultra-short-term load forecast data and removing outdated data.

[0055] Ultra-short-term load forecasting

[0056] The gradient boosting tree algorithm is used to achieve ultra-short-term load forecasting with a forecasting accuracy of no less than 95%. Feature engineering: Select historical load data (load at the same time in the past 7 days), real-time load data (load in the past 1 hour), meteorological data (temperature, humidity, wind speed), date type (weekday / weekend / holiday), and time period characteristics (peak / slow / valley) as input features, and remove redundant features (features with an absolute value of correlation coefficient < 0.3) through Pearson correlation coefficient analysis. Model training: The model hyperparameters are optimized using a grid search method, with the learning rate ranging from 0.01 to 0.1, the tree depth ranging from 3 to 10, and the number of leaf nodes ranging from 10 to 100. Five-fold cross-validation is used to improve the model's generalization ability. Forecast output: The load forecast value is output within a rolling window with a forecast period of 15 minutes; Model update: The model parameters are updated every 24 hours using actual load data to continuously optimize the prediction results.

[0057] Dynamic adjustment strategy

[0058] Based on the deviation between ultra-short-term load forecast data and day-ahead forecast data, dynamically adjust the energy storage charging and discharging power boundaries and electric vehicle charging and discharging ranges: Peak load scenario (forecast value is more than 10% higher than the planned load of the previous day): Electric vehicle adjustments: Reduce the upper limit of charging power to 50% of the rated power, expand the upper limit of discharging power to 100% of the rated power, and prioritize the use of vehicles with high charge status for discharging; Energy storage adjustment: Increase the discharge power boundary to 100% of the rated power, shorten the charging time, extend the discharge time, smooth out peak loads, and ensure that the equivalent load fluctuation does not exceed ±5%; Low load scenario (forecast value is more than 10% lower than the current day's planned load): Electric vehicle adjustments: The upper limit of charging power is increased to 100% of the rated power, while the upper limit of discharging power is reduced to 30% of the rated power, encouraging vehicles with low charge status to be charged first; Energy storage adjustment: reduce the discharge power boundary to 30% of the rated power, extend the charging time, absorb excess energy, and avoid wind and solar power curtailment; Normal load scenario (forecast value deviates from the day-ahead planned load within ±10%): The charging and discharging power range of electric vehicles and the boundary of charging and discharging power of energy storage remain unchanged, and the basic control strategy is implemented, with only minor adjustments made to small deviations.

[0059] Specifically, a parallel optimization algorithm is used to solve the corrected optimization problem. By minimizing the bias caused by the prediction error, a multi-time-scale control strategy is formed, including: The revised optimization problem is decomposed into two sub-problems: the microgrid side and the electric vehicle user side. An improved particle swarm optimization algorithm is adopted, and an adaptive adjustment strategy for inertial weights is introduced to solve the microgrid side. A genetic algorithm is used to solve the problem on the user side of electric vehicles through adaptive crossover and mutation probability; After the iteration is complete, the optimization results from both sides are merged to obtain the global optimal solution; By integrating basic control strategies with dynamic optimization results, a multi-timescale control strategy is obtained.

[0060] Specifically, a hybrid parallel optimization algorithm combining an improved particle swarm optimization algorithm and a genetic algorithm is used to solve the corrected optimization problem, minimize the deviation caused by the prediction error, and maximize the matching degree between the system's equivalent load and the day-ahead plan.

[0061] Problem Breakdown

[0062] The modified optimization problem is decomposed into two sub-problems: one on the microgrid side and the other on the electric vehicle user side, enabling parallel solution.

[0063] Algorithm Initialization

[0064] Population initialization: Generate the initial populations for both subproblems, with the population size set to 50-100. The initialization range is determined based on the constraints to ensure that the initial individuals all meet the feasibility requirements. Parallel environment setup: A multi-threaded parallel computing framework is adopted, with independent computing threads allocated to the two sub-problems to improve solution efficiency; Communication mechanism setup: Establish an inter-thread communication interface to share key information during the optimization process (such as microgrid power gap, available capacity of electric vehicles, etc.).

[0065] Parallel solution process

[0066] Microgrid-side solution: An improved particle swarm optimization algorithm is adopted, introducing an adaptive adjustment strategy for inertia weight (the inertia weight decreases linearly from 0.9 to 0.4 with the number of iterations) to improve the convergence speed. The objective function is: ; in, The cost of the deviation between the equivalent load and the current day plan. For deviation weights; User-side solution: A genetic algorithm is used, employing adaptive crossover and mutation probabilities (the crossover probability is linearly adjusted from 0.9 to 0.6 with fitness, and the mutation probability is adjusted from 0.05 to 0.1 with fitness) to avoid local optima. The objective function is: ; in, To impose penalties on travel restrictions, To constrain weights; Information exchange: Information exchange is performed once every 5 iterations to share the optimization results on both sides and adjust the constraint boundaries (e.g., the microgrid adjusts the reserve capacity requirement based on the available discharge power of electric vehicles). Global optimal fusion: After the iteration, the optimization results on both sides are merged to obtain the global optimal solution, ensuring a balance between the economic efficiency of microgrid operation and the interests of users.

[0067] Formation of multi-timescale regulation strategies

[0068] Integrating basic control strategies with dynamic optimization results, a three-level, multi-timescale control strategy is formed: "day-to-day-real-time". Day-ahead scale (24 hours): Based on the basic control strategy, the scheduling framework for electric vehicles and energy storage in each time period is clarified to provide a basis for intraday control; Intra-day scale (1-4 hours): Based on the rolling window optimization results, adjust the scheduling scheme to adapt to medium-term changes in load and wind and solar power; Real-time scale (15 minutes): Based on the real-time solution results of the parallel optimization algorithm, the operating status of each device is precisely controlled to offset short-term prediction errors.

[0069] Specifically, the microgrid side sub-problem aims to minimize operating costs and prediction errors by optimizing energy storage charging and discharging power and grid interaction power; The electric vehicle user-side problem aims to minimize the owner's charging costs and meet travel constraints by optimizing the charging and discharging time and power of a single vehicle.

[0070] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0071] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for multi-timescale elastic response and rolling optimization control of electric vehicles, characterized in that, include: Historical travel data of electric vehicles were obtained using the Monte Carlo random sampling method to generate a historical travel dataset. Wind and solar power prediction results are generated using a wind and solar power prediction model. Based on the historical travel dataset, a dual-objective model based on microgrid operating costs and vehicle owner charging costs is established, and a basic control strategy is obtained by combining wind and solar power prediction results. A rolling time window is set up, and the energy storage charging and discharging power boundary and the electric vehicle charging and discharging range are dynamically adjusted based on ultra-short-term load forecast data to make real-time corrections to the basic control strategy. Parallel optimization algorithms are used to solve the corrected optimization problem, and a multi-timescale control strategy is formed by minimizing the deviation caused by the prediction error.

2. The method for multi-timescale elastic response and rolling optimization control of electric vehicles according to claim 1, characterized in that, A wind and solar power prediction model that integrates temporal decomposition algorithm and improved LSTM neural network; Historical wind and solar power data are collected to generate a prediction base dataset, and the prediction base dataset is preprocessed. The preprocessed wind and solar power data is decomposed into trend, periodic and random components by empirical mode decomposition, thus separating the fluctuation characteristics at different scales. Predict the wind and solar power sequences for the trend term, periodic term, and random term respectively; A weighted fusion strategy is adopted, which assigns weights based on the reciprocal of the prediction error of each component, and merges the prediction results of each component to obtain the wind and solar power prediction curve.

3. The method for multi-timescale elastic response and rolling optimization control of electric vehicles according to claim 1, characterized in that, The objective function for the operating cost of the microgrid is constructed as follows: ； in, The formula for calculating the cost of wind and solar power curtailment is as follows: ； Cost per unit of power curtailment loss Let be the predicted photovoltaic power at time t. Let t be the predicted wind power. Let t be the power that the microgrid can absorb. The formula for calculating the cost of energy storage charging and discharging losses is as follows: ； The unit cost of battery loss. Let t be the energy storage charging power. Let be the energy storage discharge power at time t. For energy storage charging efficiency, For energy storage discharge efficiency; The formula for calculating the grid interaction price cost is as follows: ； Let t be the power purchased by the power grid. Let t be the electricity purchase price. Let t be the power output of the power grid at time t. Let t be the electricity price at time t; The formula for calculating the cost of standby capacity is as follows: ； As a unit of reserve cost, (t) represents the microgrid's reserve power at time t.

4. The method for multi-timescale elastic response and rolling optimization control of electric vehicles according to claim 3, characterized in that, Construct the objective function for vehicle owner charging costs, including: ； in, The formula for calculating the electricity cost for charging is as follows: ； N represents the number of electric vehicles involved in the regulation. (t) represents the charging power of the i-th vehicle at time t. (t) represents the charging electricity price at time t. For time intervals; The battery wear and tear cost is calculated based on the depth of charge / discharge, cycle life, and initial battery cost. The formula is as follows: ； Let i be the battery cycle life of the i-th vehicle after discharge via vehicle-to-grid interaction. Let i be the battery cycle life of the i-th vehicle during its sole operation. Let $\frac{i}{i}$ be the initial battery cost for the $i$-th vehicle. Let be the discharge power of the i-th vehicle at time t. Let be the discharge duration of the i-th vehicle; The formula for calculating the equivalent cost of charging waiting time is as follows: ； The charging wait time for the i-th vehicle. Let be the time value coefficient for the owner of the i-th vehicle.

5. The method for multi-timescale elastic response and rolling optimization control of electric vehicles according to claim 4, characterized in that, Based on the objective function and constraints, a genetic algorithm is used to solve the optimization problem and generate basic regulatory strategies, including: Set algorithm parameters, including population size, number of iterations, crossover probability, and mutation probability; Using real-number encoding, each chromosome corresponds to a set of electric vehicle charging and discharging power and energy storage charging and discharging power control schemes; The fitness function is the reciprocal of the transformed single objective function value. Using the roulette wheel selection method, individuals with higher fitness have a greater probability of being selected; A single-point crossover method was used, where crossover points were randomly selected to exchange partial genes between two individuals. Gaussian mutation is used to randomly perturb the genes of an individual; After iterative convergence, the optimal control scheme is output as the basic control strategy.

6. The method for multi-timescale elastic response and rolling optimization control of electric vehicles according to claim 1, characterized in that, Gradient boosting tree algorithm is used for ultra-short-term load forecasting; Historical load data, real-time load data, meteorological data, date type, and time period characteristics are selected as input features, and redundant features are eliminated through Pearson correlation coefficient analysis. The model hyperparameters were optimized using a grid search method. Output the load forecast value within the rolling window based on the forecast period; The model parameters are updated using actual load data to optimize the forecasting results; Based on the deviation between ultra-short-term load forecast data and day-ahead forecast data, the energy storage charging and discharging power boundaries and electric vehicle charging and discharging ranges are dynamically adjusted.

7. The method for multi-timescale elastic response and rolling optimization control of electric vehicles according to claim 1, characterized in that, Parallel optimization algorithms are used to solve the corrected optimization problem. By minimizing the bias caused by prediction errors, a multi-timescale control strategy is formed, including: The revised optimization problem is decomposed into two sub-problems: the microgrid side and the electric vehicle user side. An improved particle swarm optimization algorithm is adopted, and an adaptive adjustment strategy for inertial weights is introduced to solve the microgrid side. A genetic algorithm is used to solve the problem on the user side of electric vehicles through adaptive crossover and mutation probability; After the iteration is complete, the optimization results from both sides are merged to obtain the global optimal solution; By integrating basic control strategies with dynamic optimization results, a multi-timescale control strategy is obtained.

8. The method for multi-timescale elastic response and rolling optimization control of electric vehicles according to claim 7, characterized in that, The microgrid side sub-problem aims to minimize operating costs and prediction errors by optimizing the energy storage charging and discharging power and grid interaction power. The electric vehicle user-side problem aims to minimize the owner's charging costs and meet travel constraints by optimizing the charging and discharging time and power of a single vehicle.

Images