A peak-valley charging and discharging ordered control method for a dispersed charging pile inverter

Abstract

The present application belongs to the technical field of power electronic control and power distribution network load optimization, and provides a peak-valley charging and discharging ordered control method for decentralized charging piles, which comprises historical charging behavior fitting, charging time sequence probability refinement, vehicle demand energy distribution, inverter operation parameter optimization, dispatch execution and load feedback. Through the density refinement of skew correction and quartile reference position constraint, the skew charging behavior fitting deviation is reduced, and the local density feature representation capability is improved. By distributing the vehicle demand energy according to the charging time sequence probability, the expected power and the demand energy are kept cumulative consistent, and the coherence of behavior fitting and power scheduling is improved. By introducing the GNDO optimization of dynamic convergence factor, the early search and late convergence are considered. Through load feedback, forward power reduction and reverse feedback regulation, the remaining demand rolling correction and charging pile cluster peak clipping and valley filling are realized, and the power distribution transformer load fluctuation is suppressed.

Description

Technical Field

[0001] This invention relates to the field of power electronic control and distribution network load optimization technology, and in particular to a method for orderly peak-valley charging and discharging control of distributed charging pile inverters. Background Technology

[0002] With the large-scale integration of electric vehicles into residential communities, industrial parks, and public charging stations, distributed charging piles are gradually participating in the orderly charging and load regulation of distribution substations. Existing control systems typically rely on historical charging records to statistically analyze user access times, durations, and energy demands. A probability distribution model is then used to estimate charging demand for different time periods, combined with constraints such as vehicle charging time, energy demand, inverter power boundaries, and distribution transformer capacity to form a charging power plan. For scenarios with a large number of charging piles and high scheduling dimensionality, swarm intelligence optimization methods can be used to search for operating power and adjust the generated scheduling plan based on the substation load status.

[0003] The methods described above can achieve charging demand prediction and power scheduling, but they still have limitations in distributed charging scenarios. The start time of user charging is affected by factors such as travel arrangements, remaining battery power, and parking duration, and historical samples are prone to skewness, long tails, and local density differences. Models using symmetrical distributions or those relying solely on limited location and scale parameters have limited ability to characterize high-probability periods and local variations near quartiles, and fitting biases can be transmitted to the desired power formation process. Insufficient alignment between charging probability and vehicle energy demand, available time windows, and equipment power limits can easily lead to local power concentration or incomplete energy allocation. High-dimensional power optimization using a fixed search scale also struggles to balance the initial search range with later convergence stability. If scheduling relies primarily on offline planning, its adaptability to temporary vehicle access, early user departures, and fluctuations in base load is limited, further impacting the stability of peak-valley difference control in distribution areas. Summary of the Invention

[0004] To overcome the shortcomings of the existing technology, the purpose of this invention is to provide a peak-valley charging and discharging orderly control method for distributed charging pile inverters, which solves the problems of insufficient characterization of the skewed characteristics and local density characteristics of distributed charging pile users' charging behavior, which makes it difficult to accurately quantify the expected output power of the inverter and insufficient stability of the peak-valley difference control in the distribution area.

[0005] To achieve the above objectives, the present invention provides the following solution: A method for orderly peak-valley charging and discharging control of distributed charging pile inverters includes: The charging behavior skew coefficient is determined based on the historical charging data of the collected distributed charging piles. The charging behavior skew coefficient is then introduced into a three-parameter generalized normal distribution model. The GNDO optimization algorithm is used to determine the fitting parameters of the three-parameter generalized normal distribution model, thus obtaining the user charging behavior fitting model. Based on the empirical cumulative probability of the historical charging data or the cumulative distribution of the user charging behavior fitting model, the quantile reference parameter is determined. A single-parameter polynomial correction function with the quantile reference parameter as the zero value constraint position is constructed to refine the probability density of the user charging behavior fitting model and obtain the charging time sequence probability of each scheduling time period. The vehicle charging time window is determined based on the vehicle's expected access time and vehicle's expected departure time. The energy demand allocation weight for each scheduling time period is determined based on the vehicle charging time window and the charging sequence probability. The expected output power of each distributed charging pile inverter is determined based on the vehicle's energy demand and the energy demand allocation weight. The active load of the distribution area is determined based on the basic load of the distribution area and the candidate operating power of each distributed charging pile inverter. The peak-valley difference rate of the distribution area is determined based on the active load of the distribution area. Under the constraint of optimization parameters, the GNDO optimization algorithm with a dynamic convergence factor that decreases with the number of iterations is used to search the operating control parameters of each distributed charging pile inverter to obtain the optimized power command. Power scheduling is executed according to the optimized power command. The real-time load rate of the distribution transformer is determined according to the transformer load parameters. When the real-time load rate exceeds a preset threshold, the power reduction amount is determined according to the transformer load parameters. The forward charging power of each of the distributed charging piles is reduced according to the power reduction amount to obtain the cumulative power reduction amount. When the cumulative power reduction is less than the power reduction, a bidirectional inverter that meets the reverse power supply condition is called to perform reverse power supply adjustment, and the vehicle's remaining demand is updated according to the reduction in forward charging power and the change in vehicle battery energy generated by reverse power supply. Based on the updated remaining demand of the vehicles, the expected output power of the inverters for subsequent scheduling periods within the vehicle charging time window is re-determined, and the GNDO optimization algorithm is used to re-search the operating control parameters of each distributed charging pile inverter to obtain the updated optimized power command.

[0006] The present invention discloses the following technical effects: This invention provides a method for orderly peak-valley charging and discharging control of distributed charging pile inverters. By modeling charging behavior skew and refining probability density, it addresses the problem of insufficient characterization of skewed charging behavior and local density changes, thus improving the representation of user charging time-series distribution. Through time-series allocation of vehicle energy demand, it solves the problem of insufficient connection between charging probability and actual vehicle demand and charging time windows, achieving reasonable allocation of demand energy within the available charging time period. By introducing GNDO optimization with a dynamic convergence factor, it solves the problem of fixed search scales failing to balance early exploration and later convergence, achieving stable convergence of operating control parameter search. Through load feedback and rolling adjustment of the distribution transformer, it addresses the problem of insufficient adaptability of offline scheduling to load fluctuations, achieving closed-loop regulation of forward charging reduction, reverse power feedback, and re-optimization of remaining demand. Attached Figure Description

[0007] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 This is a schematic diagram of the peak-valley charging and discharging orderly control process of the distributed charging pile inverter provided in an embodiment of the present invention; Figure 2 A general flowchart provided for embodiments of the present invention; Figure 3 This is a schematic diagram of user charging behavior fitting and location feature extraction provided in an embodiment of the present invention; Figure 4 A schematic diagram illustrating the principle of allocating required energy to scheduling time periods, as provided in an embodiment of the present invention; Figure 5 An improved GNDO search flowchart with the introduction of a dynamic convergence factor is provided for embodiments of the present invention; Figure 6 A flowchart of distribution transformer load rate feedback and bidirectional inverter regulation provided in an embodiment of the present invention; Figure 7 This is a line graph showing the load comparison of different transformer substations, provided in an embodiment of the present invention. Detailed Implementation

[0009] 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.

[0010] The purpose of this invention is to provide a method for orderly peak-valley charging and discharging control of distributed charging pile inverters, which solves the problems of insufficient characterization of the skewed characteristics and local density characteristics of distributed charging pile users, resulting in difficulty in accurately quantifying the expected output power of the inverter and insufficient stability of peak-valley difference control in the distribution area.

[0011] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0012] Figure 1 This is a schematic diagram of the peak-valley charging and discharging orderly control process of the distributed charging pile inverter provided in an embodiment of the present invention, as shown below. Figure 1 As shown, this invention provides a method for orderly peak-valley charging and discharging control of distributed charging pile inverters, comprising: Step 100: Determine the charging behavior skew coefficient based on the collected historical charging data of the distributed charging piles, introduce the charging behavior skew coefficient into the three-parameter generalized normal distribution model, and use the GNDO optimization algorithm to determine the fitting parameters of the three-parameter generalized normal distribution model to obtain the user charging behavior fitting model. Step 200: Determine the quantile reference parameter based on the empirical cumulative probability of the historical charging data or the cumulative distribution of the user charging behavior fitting model, construct a single-parameter polynomial correction function with the quantile reference parameter as the zero value constraint position, refine the probability density of the user charging behavior fitting model, and obtain the charging time sequence probability of each scheduling time period. Step 300: Determine the vehicle charging time window based on the vehicle's expected access time and vehicle's expected departure time; determine the energy demand allocation weight for each scheduling time period based on the vehicle charging time window and the charging sequence probability; and determine the expected output power of each of the distributed charging pile inverters based on the vehicle's energy demand and the energy demand allocation weight. Step 400: Determine the active load of the distribution area based on the basic load of the distribution area and the candidate operating power of each distributed charging pile inverter. Determine the peak-valley difference rate of the distribution area based on the active load of the distribution area. Under the constraint of optimization parameters, use the GNDO optimization algorithm, which introduces a dynamic convergence factor that decreases with the number of iterations, to search for the operating control parameters of each distributed charging pile inverter and obtain the optimized power command. Step 500: Execute power scheduling according to the optimized power command, determine the real-time load rate of the distribution transformer according to the transformer load parameters, when the real-time load rate exceeds the preset threshold, determine the power reduction amount according to the transformer load parameters, reduce the forward charging power of each of the distributed charging piles according to the power reduction amount, and obtain the cumulative power reduction amount; Step 600: When the cumulative power reduction is less than the power reduction, call the bidirectional inverter that meets the reverse power supply condition to perform reverse power supply adjustment, and update the vehicle's remaining demand based on the decrease in forward charging power and the change in vehicle battery energy generated by reverse power supply. Step 700: Based on the updated remaining demand of the vehicle, redetermine the expected output power of the inverter for the subsequent scheduling period within the vehicle charging time window, and use the GNDO optimization algorithm to re-search the operating control parameters of each distributed charging pile inverter to obtain the updated optimized power command.

[0013] refer to Figure 2 This solution starts with historical charging behavior modeling, and sequentially completes probability density refinement and location feature extraction, vehicle energy demand allocation, inverter operating parameter optimization, and introduces distribution transformer load feedback during the scheduling execution phase. Each step uses data generated in the previous step as input, enabling historical behavior characteristics to be continuously transmitted to the power planning and real-time control processes.

[0014] Optionally, in the following description, the three-parameter probability distribution describing the user charging timing is uniformly referred to as the three-parameter generalized normal distribution model, and the group optimization method for searching the model fitting parameters and inverter operating parameters is abbreviated as the GNDO optimization algorithm. Skewed GNDO fitting introduces historical sample skew coefficients into the three-parameter generalized normal distribution model and uses GNDO to determine the model parameters; improved GNDO is used for searching inverter operating control parameters.

[0015] Specifically, after historical charging data enters the control process, the charging start times of different dates are first mapped to the same control cycle, and skew information is extracted from the sample distribution. Using a three-parameter generalized normal distribution model as the basic fitting structure, skew correction is used to describe the asymmetry of the charging timing on both sides of the central position. Based on this, using the first quartile, the median, and the third quartile as reference positions, a polynomial correction function with only one parameter to be adjusted is introduced to further refine the probability density of different time intervals, forming the charging timing probability and location characteristics required for subsequent power allocation.

[0016] Furthermore, during the power generation phase, the energy demand of each vehicle is limited to the corresponding charging time window and allocated to each scheduling time period according to the refined charging sequence probability, thus obtaining the expected output power of each charging pile inverter. Operating parameter optimization uses this expected power as a reference, while also being constrained by the inverter's physical power boundary, the platform's allowable power range, and the operating constraints of the distribution transformer. The actual operating power is searched using GNDO with a dynamic convergence factor. After the scheduling command is issued, the load rate of the distribution transformer is continuously monitored; if an over-limit occurs, forward charging power is preferentially reduced. If the bidirectional inverter and vehicle conditions allow and the reduction is still insufficient, the reverse feed-in capability is then invoked.

[0017] Optionally, a control cycle can be divided into... There are several scheduling time periods, and the length of each time period is denoted as . When using intraday control, the control period can be one day, and each scheduling time period continuously covers this period; It can be determined based on the charging controller refresh cycle, communication cycle, and transformer area load sampling cycle.

[0018] Multiple decentralized charging stations are set up within the area for orderly control. The charging pile is at the first The active power of the inverter during each scheduling period is denoted as . This embodiment specifies... This indicates that the power distribution network supplies charging power to the vehicle. This indicates that the vehicle feeds power back to the distribution network via a bidirectional inverter. The permissible power range of each charging station is When only unidirectional charging is supported, Take zero; when bidirectional adjustment is supported, Take the negative value.

[0019] Historical charging data should include at least the necessary fields to determine the charging sequence and energy demand. The basic records can include the charging start time, charging end time, single charge amount, and charging station number. When the platform can read the vehicle's battery capacity, the initial state of charge (SOC), and the target SOC, this information can be used as supplementary data to determine energy demand. Records lacking necessary fields, with contradictory time relationships, or with significantly abnormal charge levels will not be included in subsequent distribution fitting.

[0020] Specifically, historical charging data from scattered charging piles within the area is collected. For each valid record, the charging start time is mapped to a unified intraday time coordinate, and normalized according to its position relative to the start of the control cycle. The normalized time position is denoted as... By adopting a unified time scale, charging records from different dates can be directly compared statistically. For continuous charging records spanning multiple days, the start time of charging is still represented by the location on the day of access, and the duration is handled in subsequent energy demand allocation and charging time constraints.

[0021] Preferably, the normalized time position reflects the relative occurrence time of historical charging events within the control period, and subsequent skew statistics and probability distribution fitting are based on this time position.

[0022] Furthermore, the charging behavior skew coefficient is determined based on valid historical samples. This is used to characterize the degree of asymmetry in the distribution of charging start times relative to the center position. When the sample time positions lack statistically discreteness, the corresponding fixed time position can be used as the subsequent assignment center, or the historical sampling window can be expanded and refitted.

[0023] The basic fitting uses position parameters Scale parameters and shape parameters The three-parameter generalized normal distribution model is constructed, in which the scale parameter and shape parameter take positive values ​​to maintain the statistical meaning of the model parameters.

[0024] Preferably, an exponential weight formed by the sample skewness coefficient is introduced into the basic three-parameter generalized normal distribution, so that the asymmetric features of historical samples directly participate in the probability density correction. The correction relationship is as follows: .

[0025] in, This represents the probability density of user charging behavior after introducing a skewness coefficient. This represents the generalized normal probability density function with three basic parameters. This represents the skewness coefficient determined from historical samples. Indicates the normalized time position. This represents the normalization coefficient. Because... Determined directly from historical data, the fitting process only requires searching for positional parameters. Scale parameters and shape parameters .

[0026] Considering that historical charging records exist as discrete events, the control period is divided into several statistical intervals, and the empirical probability is determined by the proportion of the sample size in each interval to the total effective sample size. The model fitting error is represented by the sum of the squared deviations between the model probability and the empirical probability in each statistical interval, so that the fitting process is constrained by the sample distribution of multiple time intervals simultaneously.

[0027] Furthermore, using position parameters Scale parameters and shape parameters As parameters to be optimized, the GNDO optimization algorithm is used to search for parameter combinations that minimize the fitting error, resulting in a skew-corrected fitting model of user charging behavior. The search boundary is set based on the historical sample range, with the location parameter limited to the normalized time interval, and the scale and shape parameters both limited to positive values.

[0028] The skewness coefficient was obtained through the above fitting. Three-parameter fitting results , , and the probability density after skew correction These results are then used as input for probability density refinement.

[0029] refer to Figure 3 The first quartile is determined based on the empirical cumulative probability of historical samples or the cumulative distribution corresponding to the fitted model. , midpoint and third and fourth quartiles These three correspond to intraday time positions with cumulative probabilities of 25%, 50%, and 75%, respectively. , and As the quartile reference positions for the basic fitted distribution, the subsequent single-parameter polynomials use these three positions as zero-value constraint points.

[0030] To maintain zero correction at the three quartile reference positions and adjust the density shape in the remaining time regions, a cubic polynomial with only one parameter to be determined is introduced: .

[0031] in, It is the only parameter to be determined for the polynomial. The polynomial contains... , and The corresponding three zero factors mean that the corrections at the three quartile reference positions are all zero.

[0032] The three zero factors only restrict the polynomial correction to zero at the corresponding reference positions; they do not imply that the cumulative distribution after density correction still uses these positions as the 25%, 50%, and 75% quantiles. To avoid negative probability densities after correction, the polynomial is converted into density weights using exponential form, resulting in: .

[0033] in, This represents the probability density of charging behavior after refinement by a single-parameter polynomial. This represents the second normalization coefficient. Parameter The search can be conducted using GNDO within a preset range, with the goal of minimizing the deviation between the model probability and the empirical probability; when When the value is zero, no additional polynomial correction is introduced.

[0034] Furthermore, the refined probability density is used to extract the positional features of the user charging behavior allocation curve. , and The quartile reference positions of the basic distribution are retained; when it is necessary to describe the main concentrated period, the intraday time position corresponding to the maximum value of the refined probability density is determined as the density peak position.

[0035] When you need to output the first quartile, median, and third quartile positions of the refined probability distribution, you should redetermine the 25th, 50th, and 75th quartile positions based on the refined cumulative distribution, and not directly use the values ​​from the original distribution. , and As quartiles of the refined distribution, the baseline reference position, the redefined quartile positions, and the density peak position can all serve as positional features of the user charging behavior distribution curve.

[0036] Preferably, the refined probability density is integrated or discretely statistically analyzed according to the corresponding range of each scheduling time period within the normalized time interval to obtain the charging time sequence probability of each time period. Then, the probabilities of all time periods are uniformly normalized so that their sum is 1, so that they can be directly used for the time-series allocation of vehicle energy demand. The final refined charging time sequence probability distribution is obtained. And the corresponding location feature information.

[0037] refer to Figure 4 , for the For each charging station and the vehicles waiting to be served, the energy demand within the current scheduling cycle is first determined. If the charging platform or vehicle management system can directly provide the order's energy demand, that demand is used; otherwise, if only battery capacity, connected state of charge (SOC), and target SOC are available, the incremental energy on the battery side is determined based on the difference between the target SOC and the connected SOC, and then converted into the distribution network's energy demand by combining this with the charging link efficiency. Relevant vehicle parameters are read by the vehicle management system or charging platform.

[0038] Furthermore, the scheduling time period during which vehicles can accept charging is determined based on the expected vehicle arrival and departure times. The previously obtained charging timing probabilities are then restricted to this time window, and normalized again according to the relative proportion of the original probabilities for each time period within the window, forming the allocation weight of vehicle energy demand across different time periods.

[0039] The energy demand allocation weight only describes the relative proportion of the vehicle's total energy demand across different charging time periods. If the historical probability distribution does not overlap with the current vehicle charging time window, a proportional allocation can be used within the current charging time period, or the historical probability weight can be omitted, and the real-time scheduling calculation can proceed directly.

[0040] Preferably, the first energy demand of the vehicle and the corresponding allocation weight are determined. The inverter of the charging pile is in the first Expected output power over a time period: .

[0041] in, Indicates the first The charging pile is at the first Expected charging power over a time period Indicates the first Each charging station corresponds to the energy demand of a vehicle within the current scheduling cycle. This indicates that the vehicle was in the... The energy demand is allocated according to the weight of each time period. After allocation according to this weight, the cumulative input energy corresponding to the expected power of each charging time period is consistent with the energy demand of the vehicle.

[0042] Furthermore, this energy consistency relationship is used to constrain the scope of the temporal probability, that is, only the distribution of vehicle energy demand on the time axis is adjusted, without changing the total energy demand that needs to be met in the current scheduling cycle.

[0043] If a certain time period is directly obtained by probability allocation Exceeding the inverter's maximum allowable charging power Then the expected power for that time period is limited to The remaining energy demand corresponding to the excess will be charged according to the remaining charging time period. The proportional allocation continues. Power limiting and energy reallocation are repeated until all required energy is allocated, or the current vehicle charging time window has reached the maximum available energy. If the maximum available energy is still less than [a certain value], then [the remaining energy allocation continues]. Record the demand gap and the upper-level platform will indicate that the target demand cannot be fully met within the current time window.

[0044] After determining the expected power of a single vehicle, the expected output power of all charging piles participating in the control within the same scheduling time period is aggregated to obtain the expected aggregate power of the charging pile cluster during that time period. Through the above processing, the expected power sequence of each charging pile and the aggregate expected power sequence of the charging pile cluster are formed.

[0045] refer to Figure 5The inverter operating power of each charging pile during each scheduling time period is used as the optimization variable. Each variable is first constrained by the upper and lower limits of the corresponding equipment's physical power; when the platform sets the allowable offset range based on the expected power, the intersection of the equipment's physical power range and the allowable range of the expected power is taken as the actual search boundary of GNDO.

[0046] Optionally, if the platform does not set a separate allowable range for the desired power, the GNDO directly searches within the physical power boundary of the inverter. Charging piles without reverse power feeding capability have their lower operating power limit set to zero; inverters with bidirectional power regulation capability have a negative allowable power limit to retain reverse power feeding regulation space.

[0047] Preferably, within each scheduling time period, the basic load of the distribution transformer excluding the controlled charging piles is first obtained, and then superimposed with the candidate operating power of all controlled charging piles in the same time period, thereby obtaining the active load sequence of the distribution area corresponding to the candidate scheduling scheme.

[0048] Optionally, the peak-to-valley difference rate can be determined based on the peak and valley values ​​of the active power load in a dispatching cycle, and the degree of improvement of the peak-to-valley difference of the dispatching scheme in the dispatching area can be evaluated based on this.

[0049] The specific statistical caliber of peak-valley difference rate can follow the existing transformer area load evaluation standard of the project, but the calculation caliber should be kept consistent within the same optimization cycle to ensure comparability between different candidate scheduling schemes.

[0050] The 28.5% peak-to-valley difference rate is the result obtained from example data under the set operating conditions of this embodiment, and is not used as a fixed control threshold for other operating scenarios. When an engineering project needs to set an upper limit for the peak-to-valley difference rate, the corresponding target value should be determined according to the actual operating conditions.

[0051] Preferably, using only the peak-valley difference rate as the optimization objective may cause the scheduled power to deviate excessively from the formed demand sequence. Therefore, the normalized deviation of each charging pile's candidate operating power relative to its corresponding expected power can be introduced into the objective function as the expected power deviation term.

[0052] The expected power deviation term measures the degree to which the candidate scheduling scheme deviates from the demand timing given by the historical behavior model. When normalizing, the rated power of the corresponding inverter or the maximum allowable charging power can be used as a reference scale to avoid unnecessary influence of different equipment capacities on the magnitude of the deviation.

[0053] Specifically, the operating constraints of distribution transformers are characterized by the load factor. The load factor for each scheduling period is determined by the ratio of the active load of the distribution area to the reference value of the allowable active load of the distribution transformer.

[0054] When the platform sets a load rate threshold, the time period exceeding the threshold will be penalized for transformer over-limit according to the actual degree of over-limit; the time period not exceeding the threshold will not be included in this penalty item.

[0055] Furthermore, in achieving the goal of simultaneously balancing peak shaving and valley filling, demand timing maintenance, and transformer load constraints, the peak-valley difference rate, expected power deviation term, and transformer over-limit penalty term can be weighted and combined to form a comprehensive objective function: .

[0056] in, This represents the comprehensive evaluation value of the candidate scheduling scheme. This indicates the peak-to-valley difference rate of the distribution area. Represents the expected power deviation term. This indicates the penalty for exceeding transformer limits. , and These are the corresponding non-negative weights. Prioritizing peak smoothing and valley filling can improve [the effectiveness / efficiency]. The relative weight; when the distribution transformer operates closer to the load threshold, it can be increased The relative weights are determined. Each weight remains constant within the same optimization cycle.

[0057] Preferably, in addition to the upper and lower limits of operating power, the candidate scheduling scheme must also meet the energy demand constraints of the vehicles. Since the bidirectional inverter may perform reverse feeding during certain time periods, the vehicle energy balance is statistically calculated based on the net energy on the battery side; when the scheduling ends and the vehicle is required to reach the target state of charge, the following should be satisfied: .

[0058] in, Indicates the first The vehicle's available battery capacity, and These represent the vehicle's initial state of charge and the target state of charge, respectively. Indicates charging link efficiency. Indicates the efficiency of the reverse feeder link; For corresponding vehicle charging, This corresponds to the vehicle feeding power back to the power distribution network. The energy reduction on the battery side caused by the reverse feeding should be made up within the subsequent charging period; if the charging time window is insufficient to fully meet the demand, the minimum guaranteed energy set by the platform will be used, and the unmet energy will be recorded.

[0059] Specifically, the GNDO population size and the dimension of the optimization variables are set based on the number of charging piles involved in the control, the number of scheduling time periods, and the available solution resources. Candidate solutions are randomly initialized within each operating power search boundary, and after initialization, feasibility corrections are made according to vehicle charging time, equipment power boundaries, and energy demand constraints.

[0060] Furthermore, local development uses the current candidate solution, the current best candidate solution, and the population average position as basic information to determine the update direction of the candidate solution near the current better region.

[0061] A local search center is constructed from the current candidate solution, the current best candidate solution, and the population average position, and the local search scale is determined based on the degree of dispersion of the three relative to the search center.

[0062] The local search center determines the baseline position for updating candidate solutions, while the discrete scale controls the amplitude of local perturbations. Together, they limit the search scope in the local development phase.

[0063] Preferably, to allow the search scale to adjust as the optimization process progresses, a dynamic convergence factor that decreases with the number of iterations is set in the improved GNDO: .

[0064] in, Indicates the first The dynamic convergence factor of the next iteration. and These represent the search scales at the beginning and end of the iteration, respectively. Indicates the maximum number of iterations. Represents the decay exponent and satisfies .when When the value is 1, the dynamic convergence factor decreases linearly. When taking other positive values, the decay rate of the search scale with the number of iterations can be adjusted.

[0065] During local development, the local search center is used as a benchmark, and a dynamic convergence factor is applied to the local search scale and the amplitude of random perturbation to obtain new local candidate locations.

[0066] In the early stages of iteration, the dynamic convergence factor is relatively large, and candidate solutions can be searched over a wide range. As the iteration progresses, the perturbation amplitude gradually shrinks, causing the later search to focus more on the vicinity of the current optimal region.

[0067] Furthermore, when entering global exploration, multiple individuals that are different from each other and different from the current candidate solutions are randomly selected from the current population, and the positional differences between the candidate individuals are used to construct the global exploration direction.

[0068] The current candidate solution, the global exploration direction, and the dynamic convergence factor are used together to update the candidate position, resulting in a new global exploration candidate solution.

[0069] The positional differences between candidate individuals provide cross-regional search directions, while the dynamic convergence factor limits the update range, allowing global exploration to retain greater jump capabilities in the early stages and reducing ineffective large-scale movements in the later stages.

[0070] Specifically, each iteration generates new candidate solutions by selecting either local development or global exploration according to a preset probability. Variables exceeding the power boundary are first truncated, and candidate solutions that do not meet the vehicle charging time and energy demand constraints are then corrected through energy redistribution. Subsequently, the corrected candidate solutions are determined. If the objective function value of a new candidate solution is better than that of the original candidate solution, the new candidate solution is replaced; otherwise, the original candidate solution is retained. The search ends when the maximum number of iterations is reached, or when the change in the optimal objective function after a preset number of iterations is below a stopping threshold, and the optimal operating control parameters are output. After the search, the optimized power commands for each charging pile inverter in each scheduling time period are obtained. .

[0071] refer to Figure 6 Based on the previous results Generate scheduling instructions for each distributed charging pile inverter. For bidirectional AC / DC inverters, Perform charging power control at the same time; Power supply control is executed when both the vehicle and the inverter meet the reverse power supply conditions. When a device that does not support reverse power supply receives a negative power command, the command is limited to zero, and the adjustment amount that needs to be compensated is reallocated.

[0072] During the scheduling execution, the actual active load of the distribution transformer is read according to the sampling interval matching the control cycle, and the real-time load rate is determined by the ratio of the measured load to the reference value of the transformer's allowable active load.

[0073] Preferably, when the real-time load rate does not exceed the preset threshold, the current optimized power command is maintained; when it exceeds the threshold, the power to be reduced is determined according to the actual load corresponding to the over-limit, and the load priority adjustment is switched to the load priority adjustment.

[0074] Specifically, load priority regulation first utilizes the available reduction space of existing forward charging power. A first-level regulation margin is determined based on the current forward charging power of each participating charging station. This margin represents the load reduction capacity that each charging station can provide by reducing its current charging power to zero without entering reverse feeding mode.

[0075] The forward charging power is adjusted from large to small according to the first-level adjustment margin, and the first-level adjustment is stopped after the cumulative reduction reaches the required reduction power. If the over-limit cannot be eliminated even after all the forward charging power involved in the adjustment has been reduced to zero, the second-level reverse feed-in margin is further determined for bidirectional equipment that meets the reverse feed-in conditions of the vehicle and inverter.

[0076] The secondary reverse feed margin represents the maximum feed space that the bidirectional equipment can provide when it continues to adjust from zero power to negative power to the physical lower limit of the equipment; when the vehicle or inverter does not have reverse feed conditions, this margin is zero. Subsequently, the power is further reduced according to the secondary margin until the sum of the primary reduction and the secondary feed amount reaches the required reduction power, or all adjustable capacity has been used.

[0077] Furthermore, after the real-time power adjustment is completed, the reduction in forward charging and the change in battery-side energy caused by reverse power feedback are included in the vehicle's remaining demand, and the adjustment is re-executed in subsequent charging periods based on battery-side energy constraints. Figure 4 and Figure 5 The corresponding steps, or rolling corrections in the next control cycle. Thus, the historical behavior model is responsible for forming the initial demand sequence, the improved GNDO is responsible for optimizing the power distribution of the distribution area under constraints, and the real-time load rate of the distribution transformer is used to correct the load deviation in the scheduling execution. The three work continuously within the same control link.

[0078] Example 1: This example uses a residential community distribution substation with multiple distributed charging piles as the implementation object, illustrating the specific process from historical charging behavior modeling to scheduling execution and load feedback. The charging piles are connected via power electronic interfaces capable of receiving external active power setpoints, with some devices employing bidirectional AC / DC inverters that support charging and reverse power feedback. The substation's operation and maintenance platform can read historical charging records, current vehicle demand information, inverter power boundaries, and active load measurements of the distribution transformer. Key implementation parameters and data conditions are shown in Table 1.

[0079] Table 1

[0080] After collecting historical charging records for three consecutive months, the samples were first organized according to charging pile number and intraday time location, and records lacking essential fields such as charging start time and charging amount were removed. The charging start times of valid records were normalized to a left-closed, right-open intraday time interval of 0 to 1, and the empirical probability of each time interval was calculated to determine the sample skewness coefficient. A three-parameter generalized normal density function was used as the base function, with a skewness exponential correction weight added to it, and then a GNDO search was employed. , and The fitting objective is to minimize the squared error between the model probability and the empirical probability in each statistical interval.

[0081] After completing the skew fitting, determine the skew based on historical samples or the cumulative distribution of the fitted data. , and and construct This makes the polynomial correction zero at the three quartile reference positions. Subsequently, exponential weights are used to refine the probability density of the first stage, with parameters... The search is still based on the deviation between the corrected statistical interval probabilities and the empirical probabilities. The final probability density determines the scheduling time period. When the interquartile positions of the refined distribution are needed, they should be re-determined based on the refined cumulative distribution, rather than directly using the unrefined distribution. , and .

[0082] After entering intraday scheduling, the current energy demand of each vehicle corresponding to a participating charging pile is read. When the platform can directly provide the order's energy demand, the order value is used; when only battery capacity and SOC information are available, the energy demand on the distribution network side is calculated based on battery capacity, access SOC, target SOC, and charging link efficiency. Then, a charging time window is determined based on the vehicle's access time and expected departure time. The obtained charging sequence probability is limited to this window and re-normalized to obtain the energy demand allocation weight for each time period, thereby forming the expected output power. If the expected power for a certain time period exceeds the inverter's maximum charging power, the excess energy is further allocated to other available charging time periods, ensuring that the cumulative expected input energy matches the vehicle's energy demand.

[0083] Read the basic load forecast of the distribution area or the basic load sequence of the current dispatch cycle. Subsequently, a candidate vector consisting of the operating power of all controlled inverters is established. Each variable is constrained to the intersection of the physical power boundary of the equipment and the allowable range of the desired power, while simultaneously satisfying the constraints of vehicle charging time and net energy demand on the battery side. Optimization uses the peak-valley difference rate of the transformer substation as the main evaluation metric, and incorporates the desired power deviation and transformer over-limit penalty according to the project's operational requirements, forming a comprehensive objective function. .

[0084] After completing the improved GNDO population initialization, the current optimal candidate solution and the average population position are determined in each iteration. Local exploration determines the local search center and discrete scale based on the current candidate solution, the current optimal candidate solution, and the average population position; global exploration randomly selects different candidate individuals to form a differential search direction. Both types of updates are subject to a dynamic convergence factor. Control. As the number of iterations increases, from Towards The search process decreases gradually, maintaining a large range of candidate movement in the early stages and then gradually reducing the update magnitude in the later stages.

[0085] After generating new candidate solutions, power boundary adjustments are performed, followed by checks on vehicle charging time and cumulative energy demand constraints. Once the constraints are met, the objective function value is determined, and the population is updated according to the best-of-breed principle. After iteration, the optimal power for each charging station in each scheduling time period is obtained. It converts the active power setpoint that the inverter controller can recognize and sends it through the substation operation and maintenance platform or the charging pile communication link.

[0086] After the control command is executed, the actual active load of the distribution transformer is continuously collected and the real-time load rate is determined. When the load rate is within the threshold, the current power command is maintained; if it exceeds the limit, the power reduction required to bring the load rate back to the threshold is first determined, and then the forward charging power is reduced first. If the forward reduction capacity is insufficient and a bidirectional inverter that meets the reverse feed-in conditions exists, the adjustment continues from zero power to negative power. The energy changes in the vehicle caused by reducing charging or reverse feed-in are recalculated into the remaining demand and participate in the next round of optimization during subsequent charging periods.

[0087] Furthermore, the following data illustrates the specific execution method of the aforementioned calculation relationship. Within a scheduling cycle, a vehicle has four charging time periods, each lasting 1 hour; the vehicle's available battery capacity is 48 kWh, the initial state of charge (SOC) is 50%, the target SOC is 75%, and the charging link efficiency is 0.96. Therefore, an additional 12 kWh of charging energy is needed on the battery side, corresponding to approximately 12.5 kWh of charging energy being provided by the distribution network.

[0088] If the normalized time-series weights obtained within the vehicle's charging time window are 0.10, 0.20, 0.30, and 0.40 respectively, then the expected charging powers for the four time periods are 1.25kW, 2.50kW, 3.75kW, and 5.00kW respectively, with a cumulative input energy of 12.5kWh. When the inverter's maximum charging power is not less than 5.00kW, all time periods meet the power boundary requirements, and power cutoff and residual energy redistribution are not required.

[0089] Taking load feedback as an example, the allowable active load reference value for the distribution transformer is taken as 100kW, and the load rate threshold is taken as 0.90. When the measured active load of the distribution area is 96kW during a certain period, it is necessary to reduce 6kW to bring the load rate back to the threshold. If the forward power of all adjustable charging piles can only be reduced by a maximum of 3kW, then the 3kW forward charging power is first reduced to zero, and then the remaining 3kW regulation is provided by a bidirectional inverter that meets the reverse feedback conditions and has sufficient negative power margin. After adjustment, the load of the distribution area is 90kW, the real-time load rate is 0.90, and the energy reduction on the battery side caused by reverse feedback continues to be included in the subsequent charging demand.

[0090] Under the operating conditions of the residential distribution transformer area set in this embodiment, skew fitting and single-parameter polynomial density refinement were performed using three consecutive months of historical charging data. Demand energy allocation, improved GNDO optimization, and scheduling execution were then completed. The resulting peak-to-valley difference rate of the active power load in the transformer area was 28.5%, demonstrating a smoothing effect on the load fluctuations of the distribution transformer. This result corresponds to the specific transformer area's base load, number of vehicles, access time period, adjustable power range, and proportion of bidirectional feeder equipment, and is not a fixed indicator that must be achieved in other implementation scenarios.

[0091] Example 2: Using the same distribution area equipment, vehicle demand, scheduling time period, inverter power boundary, and distribution transformer load rate threshold as in Example 1, and setting up a conventional comparison method, this example observes the impact of changes in user behavior modeling and optimization search methods on the distribution area load curve. The following representative time period data is used to verify the correspondence between calculation relationships and technical effects, and is not considered as actual on-site testing data.

[0092] The conventional comparison method uses a three-parameter basic distribution to fit the historical charging start time, without introducing a historical sample skew coefficient, and does not employ a single-parameter polynomial with the quartile reference position as the zero value for density refinement. Both methods maintain consistency in terms of vehicle energy demand, charging time window, and inverter power boundary. During the operating parameter search phase, the conventional method uses a fixed search scale, while this scheme uses a dynamic convergence factor that decreases with the number of iterations. Apart from the above differences, both methods aim to reduce the peak-to-valley difference in the transformer substation area while meeting the vehicle's energy demand.

[0093] The load of the transformer substations in six representative time periods within a scheduling cycle was analyzed, and the results are shown in Table 2. Both sets of data used the same base load and vehicle access conditions. Conventional comparison methods resulted in higher aggregated charging power during high-demand periods. This scheme, however, allocates demand energy after refining the charging time sequence probability and adjusts the operating power through dynamic convergence GNDO, thereby shifting some charging demand to other available charging time periods.

[0094] Table 2

[0095] Table 2 and Figure 7 In the example data provided, the peak load of the transformer area using the conventional comparison method is 110.0 kW, and the valley load is 65.0 kW, corresponding to a peak-valley difference rate of 40.9%. In this scheme, the peak load is 100.0 kW, and the valley load is 71.5 kW, corresponding to a peak-valley difference rate of 28.5%. Comparing using the same statistical caliber, the peak-valley difference rate decreases by 12.4 percentage points, the representative peak load decreases by 10.0 kW, and some demand shifts to lower load periods. This set of data is only used to illustrate the relationship between time-series probability refinement, demand energy redistribution, and dynamic convergence optimization.

[0096] Furthermore, from the perspective of the control link, conventional three-parameter fitting tends to assign concentrated weights to high-probability time periods when historical samples exhibit skewness and local density differences. This scheme first refines the temporal probability using skewness coefficients and single-parameter polynomials, then redistributes the vehicle's energy demand within the charging time window, resulting in a more reasonable temporal distribution of expected power while meeting total energy demand. The improved GNDO adjusts the search scale through a dynamic convergence factor, retaining a wider search range in the early stages and narrowing the update amplitude in the later stages. Therefore, it can obtain a smoother exemplary transformer load sequence under the same equipment power boundary.

[0097] Specifically, the above comparison results correspond to the comprehensive effect of the complete control link. The skewed modeling of user charging behavior and the density refinement of quartile reference position constraints are used to improve the temporal probability description; vehicle demand energy allocation is used to connect the probabilistic model with the current vehicle charging conditions; dynamic convergence GNDO is used to optimize operating power within the constraints; and distribution transformer load rate feedback is used to handle real-time load deviations during the execution phase.

[0098] The beneficial effects of this invention are as follows: (1) After refining the description of the skewed features and local density changes of user charging behavior, the charging time sequence probability can be closer to the actual distribution of historical samples, reducing the transmission of behavior fitting deviation to the subsequent power scheduling process.

[0099] (2) The energy demand of the vehicle is allocated according to the temporal probability within the actual charging time window, taking into account the historical charging pattern and the current available charging time of the vehicle, improving the problem of excessive power concentration in some periods, and improving the integrity of the energy demand allocation.

[0100] (3) The GNDO search scale gradually shrinks during the iteration process. In the early stage, a large parameter exploration space is retained, and in the later stage, it can continue to refine around the better area, thus improving the stability of the operation control parameter search and helping to smooth the load curve of the substation area.

[0101] (4) After introducing real-time load feedback from the distribution transformer during the optimized power command execution process, the charging power can be adjusted when the load of the distribution area changes, and the vehicle energy changes caused by the adjustment are included in the subsequent scheduling, which enhances the adaptability of the power plan to changes in the actual operating status.

[0102] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0103] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for orderly peak-valley charging and discharging control of distributed charging pile inverters, characterized in that, include: The charging behavior skew coefficient is determined based on the historical charging data of the collected distributed charging piles. The charging behavior skew coefficient is then introduced into a three-parameter generalized normal distribution model. The GNDO optimization algorithm is used to determine the fitting parameters of the three-parameter generalized normal distribution model, thus obtaining the user charging behavior fitting model. Based on the empirical cumulative probability of the historical charging data or the cumulative distribution of the user charging behavior fitting model, the quantile reference parameter is determined. A single-parameter polynomial correction function with the quantile reference parameter as the zero value constraint position is constructed to refine the probability density of the user charging behavior fitting model and obtain the charging time sequence probability of each scheduling time period. The vehicle charging time window is determined based on the vehicle's expected access time and vehicle's expected departure time. The energy demand allocation weight for each scheduling time period is determined based on the vehicle charging time window and the charging sequence probability. The expected output power of each distributed charging pile inverter is determined based on the vehicle's energy demand and the energy demand allocation weight. The active load of the distribution area is determined based on the basic load of the distribution area and the candidate operating power of each distributed charging pile inverter. The peak-valley difference rate of the distribution area is determined based on the active load of the distribution area. Under the constraint of optimization parameters, the GNDO optimization algorithm with a dynamic convergence factor that decreases with the number of iterations is used to search the operating control parameters of each distributed charging pile inverter to obtain the optimized power command. Power scheduling is executed according to the optimized power command. The real-time load rate of the distribution transformer is determined according to the transformer load parameters. When the real-time load rate exceeds a preset threshold, the power reduction amount is determined according to the transformer load parameters. The forward charging power of each of the distributed charging piles is reduced according to the power reduction amount to obtain the cumulative power reduction amount. When the cumulative power reduction is less than the power reduction, a bidirectional inverter that meets the reverse power supply condition is called to perform reverse power supply adjustment, and the vehicle's remaining demand is updated according to the reduction in forward charging power and the change in vehicle battery energy generated by reverse power supply. Based on the updated remaining demand of the vehicles, the expected output power of the inverters for subsequent scheduling periods within the vehicle charging time window is re-determined, and the GNDO optimization algorithm is used to re-search the operating control parameters of each distributed charging pile inverter to obtain the updated optimized power command.

2. The method for orderly peak-valley charging and discharging control of a distributed charging pile inverter according to claim 1, characterized in that, The quartile reference parameters include: the first quartile, the median, and the third quartile; the optimization constraint parameters include: the inverter power boundary, the vehicle charging time window, and the vehicle's energy demand; the transformer load parameters include: the actual active load of the distribution transformer, the allowable active load reference value, and the preset threshold.

3. The method for orderly peak-valley charging and discharging control of a distributed charging pile inverter according to claim 1, characterized in that, Based on the collected historical charging data of distributed charging piles, a charging behavior skew coefficient is determined. This skew coefficient is then introduced into a three-parameter generalized normal distribution model. The GNDO optimization algorithm is used to determine the fitting parameters of the three-parameter generalized normal distribution model, resulting in a user charging behavior fitting model, including: The charging start time in each valid historical charging record is mapped to a unified intraday time coordinate, and normalized according to the position of the charging start time relative to the start of the control cycle to obtain the normalized time position. The charging behavior skew coefficient is determined based on each of the normalized time positions; The three-parameter generalized normal distribution model is constructed using position parameters, scale parameters, and shape parameters. The probability density of the three-parameter generalized normal distribution model is then skewed based on the charging behavior skew coefficient. The control period is divided into multiple statistical intervals. The empirical probability is determined based on the number of historical samples in each statistical interval. The fitting error is determined based on the model probability and the empirical probability in each statistical interval. The position parameter, the scale parameter, and the shape parameter are determined using the GNDO optimization algorithm to obtain the user charging behavior fitting model.

4. The method for orderly peak-valley charging and discharging control of a distributed charging pile inverter according to claim 2, characterized in that, Based on the empirical cumulative probability of the historical charging data or the cumulative distribution of the user charging behavior fitting model, quantile reference parameters are determined. A single-parameter polynomial correction function is constructed with the quantile reference parameters as the zero-value constraint position. The probability density of the user charging behavior fitting model is refined to obtain the charging time-series probability for each scheduling time period, including: Based on the empirical cumulative probability or the cumulative distribution, determine the first quartile, median, and third quartile corresponding to the cumulative probabilities of 25%, 50%, and 75%, respectively. Construct the single-parameter polynomial correction function; the single-parameter polynomial correction function is: ;in, This refers to the single-parameter polynomial correction function; Parameters to be determined; To normalize the time position; , , These are the first quartile, the median, and the third quartile, respectively. The single-parameter polynomial correction function is converted into density weights in exponential form. The probability density of the user charging behavior fitting model is corrected according to the density weights, and the corrected probability density is normalized to obtain the refined charging behavior probability density. Based on the time range corresponding to each scheduling time period, the refined charging behavior probability density is segmented and normalized to obtain the charging time sequence probability for each scheduling time period.

5. The method for orderly peak-valley charging and discharging control of a distributed charging pile inverter according to claim 1, characterized in that, The vehicle charging time window is determined based on the vehicle's expected access time and expected departure time. The energy demand allocation weight for each scheduling time period is determined based on the vehicle charging time window and the charging timing probability. The expected output power of each distributed charging pile inverter is determined based on the vehicle's energy demand and the energy demand allocation weight, including: Based on the vehicle's expected access time and the vehicle's expected departure time, the scheduling time period within the vehicle's charging time window is filtered; The charging sequence probability is limited to the vehicle charging time window, and the charging sequence probability of each scheduling time period within the vehicle charging time window is normalized to obtain the demand energy allocation weight. The expected output power for each scheduling time period is determined based on the vehicle's energy demand and the energy demand allocation weight; the expression for the expected output power is: ;in, For the first The charging pile is at the first Expected charging power over a time period; For the first Each charging station corresponds to the energy demand of the vehicle within the current scheduling cycle; For the vehicle in the Weighting of energy demand allocation for each time period; The duration of the scheduling period; When the expected output power during any scheduling time period exceeds the maximum charging power of the corresponding distributed charging pile inverter, the expected output power is limited to the maximum charging power, and the remaining required energy is allocated to the remaining scheduling time periods within the vehicle charging time window.

6. The method for orderly peak-valley charging and discharging control of a distributed charging pile inverter according to claim 2, characterized in that, The active power load of the distribution area is determined based on the base load of the distribution area and the candidate operating power of each distributed charging pile inverter. The peak-valley difference rate of the distribution area is determined based on the active power load. Under the constraint of optimization parameters, the GNDO optimization algorithm, which introduces a dynamic convergence factor that decreases with the number of iterations, is used to search for the operating control parameters of each distributed charging pile inverter to obtain the optimized power command, including: The active load of the distribution area is obtained by superimposing the basic load of the distribution area with the candidate operating power of each of the distributed charging pile inverters within the same scheduling time period. The peak-valley difference rate of the transformer area is determined based on the maximum and minimum values ​​of the active load of the transformer area within a control cycle; the expected power deviation item is determined based on each candidate operating power and the corresponding expected output power; and the transformer over-limit penalty item is determined based on the active load of the transformer area, the allowable active load reference value, and the preset threshold. The peak-valley difference rate of the transformer area, the expected power deviation term, and the transformer over-limit penalty term are weighted to obtain the comprehensive evaluation value of the candidate scheduling scheme; Under the constraints of the inverter power boundary, the vehicle charging time window, and the vehicle energy demand, the comprehensive evaluation value is used as the objective function value. The GNDO optimization algorithm, which incorporates the dynamic convergence factor, is used to search for the operating control parameters of each distributed charging pile inverter to obtain the optimized power command.

7. The method for orderly peak-valley charging and discharging control of a distributed charging pile inverter according to claim 6, characterized in that, Under the constraints of the inverter power boundary, the vehicle charging time window, and the vehicle's energy demand, the comprehensive evaluation value is used as the objective function value. The GNDO optimization algorithm, incorporating the dynamic convergence factor, is employed to search for the operating control parameters of each distributed charging pile inverter, resulting in the optimized power command, including: Candidate solutions are generated within the operating power search range corresponding to each of the distributed charging pile inverters, and the candidate solutions are constrained and corrected according to the vehicle charging time window, the inverter power boundary, and the vehicle's energy demand. The local search center is determined based on the current candidate solution, the current best candidate solution, and the average position of the population; the local search scale is determined based on the degree of dispersion of the current candidate solution, the current best candidate solution, and the average position of the population relative to the local search center. The dynamic convergence factor is determined based on the number of iterations; the expression for the dynamic convergence factor is: ;in, For the first The dynamic convergence factor of the next iteration; The search scale at the end of the iteration; This is the search scale used in the initial iteration. This represents the maximum number of iterations. The decay index; The local search scale and random perturbation amplitude are adjusted according to the dynamic convergence factor to obtain local candidate positions, or the global exploration direction is determined according to the positional differences between candidate individuals, and global exploration candidate positions are generated according to the global exploration direction and the dynamic convergence factor. A new candidate solution is generated based on the local candidate position or the global exploration candidate position. The new candidate solution is then constrained and corrected based on the vehicle charging time window, the inverter power boundary, and the vehicle energy demand to obtain a corrected candidate solution. The objective function value of the corrected candidate solution is determined, and the candidate solution is updated according to the objective function value of the corrected candidate solution until a preset stopping condition is met, thereby obtaining the optimized power command.

8. The method for orderly peak-valley charging and discharging control of a distributed charging pile inverter according to claim 2, characterized in that, Power scheduling is executed according to the optimized power command. The real-time load rate of the distribution transformer is determined according to the transformer load parameters. When the real-time load rate exceeds a preset threshold, the power reduction amount is determined according to the transformer load parameters. The forward charging power of each of the distributed charging piles is reduced according to the power reduction amount to obtain the cumulative power reduction amount, including: The real-time load rate of the distribution transformer is determined based on the actual active load of the distribution transformer and the allowable active load reference value. When the real-time load rate exceeds the preset threshold, the power reduction amount is determined based on the actual active load of the distribution transformer, the allowable active load reference value, and the preset threshold. The adjustment margin corresponding to the reduction of the current forward charging power of each of the distributed charging piles to zero is determined based on the current forward charging power of each of the distributed charging piles. The forward charging power of each of the distributed charging piles is reduced from large to small according to the adjustment margin until the cumulative power reduction reaches the power reduction amount or the forward charging power of each of the distributed charging piles involved in the adjustment is reduced to zero, thus obtaining the cumulative power reduction amount.

9. The method for orderly peak-valley charging and discharging control of a distributed charging pile inverter according to claim 1, characterized in that, When the cumulative power reduction is less than the power reduction amount, a bidirectional inverter that meets the reverse feed-in condition is activated for reverse feed-in adjustment. The remaining vehicle demand is updated based on the decrease in forward charging power and the change in vehicle battery energy caused by reverse feed-in, including: Select bidirectional inverters in which both the vehicle and the inverter meet the reverse feed conditions. The reverse feed-in regulation margin is determined based on the power range corresponding to the lower limit of inverter power when each bidirectional inverter is adjusted from zero power to negative power. The operating power of each bidirectional inverter is reduced according to the reverse feed adjustment margin until the sum of the cumulative power reduction and the reverse feed power reaches the power reduction amount or the reverse feed adjustment margin of each bidirectional inverter is activated. The vehicle's remaining demand is updated based on the reduction in charging energy corresponding to the decrease in forward charging power and the reduction in vehicle battery energy corresponding to the reverse power feedback.

10. The method for orderly peak-valley charging and discharging control of a distributed charging pile inverter according to claim 1, characterized in that, Based on the updated remaining vehicle demand, the expected inverter output power for subsequent scheduling periods within the vehicle charging time window is re-determined, and the GNDO optimization algorithm is used to re-search the operating control parameters of each distributed charging pile inverter to obtain the updated optimized power command, including: Filter the scheduling time periods within the vehicle charging time window that are after the current scheduling time period to obtain the subsequent scheduling time periods; The charging timing probability corresponding to the subsequent scheduling time period is normalized again to obtain the updated demand energy allocation weight. The inverter's expected output power for the subsequent scheduling period is re-determined based on the updated vehicle remaining demand and the updated demand energy allocation weights. Using the redefined expected output power of the inverter as a power scheduling reference, the GNDO optimization algorithm is used to re-search the operating control parameters of each of the distributed charging pile inverters under the constraints of the optimization parameters, so as to obtain the updated optimized power command.

Images