Vehicle-pile-net cluster collaborative regulation method and system considering demand response

Abstract

The present application relates to the technical field of intelligent power distribution, in particular to a method and system for coordinated regulation of a cluster of vehicles, piles and power grids considering demand response. The method first compares the change value of the demand response power instruction at adjacent time to determine the instruction change degree; determines the prediction deviation degree according to the error between the rated power on the power supply area side, the actual load power and the predicted load power; iteratively updates the penalty parameter of the ADMM standard solvable model based on the instruction change degree; iteratively updates the convergence threshold of the ADMM standard solvable model in combination with the instruction change degree and the prediction deviation degree; when the original residual and the dual residual both satisfy the convergence condition or reach the maximum number of iterations, the iteration is stopped, the charging and discharging power regulation instruction of the cluster of vehicles, piles and power grids is determined based on the current local variable and is executed. The present application improves the fast and accurate tracking ability of the demand response instruction under the scenario of large-scale electric vehicles access, effectively guarantees the safe and stable operation of the distribution network.

Description

Technical Field

[0001] This invention relates to the field of intelligent power distribution technology, and more specifically to a method and system for coordinated control of vehicle-charging-network clusters that takes demand response into account. Background Technology

[0002] With the continuous advancement of smart power systems and the acceleration of electrification in the transportation sector, the number of electric vehicles has reached an unprecedented scale. The disorderly charging behavior of large-scale electric vehicles poses a severe challenge to the safe and stable operation of the power system, especially at the distribution network level. However, as mobile energy storage units, the massive number of electric vehicles also provide valuable flexibility resources for the power grid due to their flexible charging and discharging characteristics.

[0003] Demand response, as a power load management mechanism, is based on the following core idea: when the reliability of the power system is threatened, the power load side receives instructions from the power grid and then proactively changes its inherent electricity consumption habits, temporarily reducing or transferring some of the electricity load, thereby responding to the tight power supply situation and transforming rigid, unschedulable user loads into flexible, interactive power grid resources to ensure the stable operation of the power grid. When using the Alternating Directional Multiplier Method (ADMM) to achieve coordinated control of vehicle-charging-network clusters, the global control problem can be decomposed into multiple local sub-problems by cluster. Each cluster only needs to solve its own charging and discharging optimization sub-problem, avoiding the exponential complexity of centralized optimization. This is not only suitable for large-scale electric vehicle access, but also does not rely on a single central controller, avoiding the drawback of a single point of failure causing the entire network to lose control. However, when using the ADMM to achieve coordinated control of vehicle-charging-network clusters that take demand response into account, the penalty parameter and convergence threshold in the traditional ADMM lack objective selection criteria. When the demand response commands of the vehicle-charging-network cluster change frequently, it is difficult to perform accurate dynamic tracking. After each response command update, it is necessary to re-iterate convergence, resulting in serious lag in control commands. Summary of the Invention

[0004] To address the technical problem of adaptively adjusting the penalty parameter and convergence threshold in traditional ADMM models, this invention aims to provide a vehicle-charging-network cluster collaborative control method and system that considers demand response. The specific technical solution adopted is as follows: Construct a standard solvable ADMM model that includes local variables, global variables, and consistency constraints, and construct an augmented Lagrangian function; Obtain operational data of the vehicle-charging network cluster, including: demand response power command, actual load power, and rated power of the power supply station area; By comparing the changes in demand response power commands at adjacent time points, the degree of command change is determined; based on the error between the rated power, actual load power, and predicted load power at the power supply station, the prediction deviation is determined; based on the degree of command change, the penalty parameters of the ADMM standard solvable model are iteratively updated to obtain the penalty parameters corresponding to the current iteration. Combining the degree of instruction change and the prediction bias, the convergence threshold of the ADMM standard solvable model is iteratively updated to obtain the convergence threshold corresponding to the current iteration; during the iterative update, the original residual and the dual residual are updated simultaneously. When both the original residual and the dual residual satisfy the convergence condition or reach the maximum number of iterations, the iteration stops, and the charging and discharging power control command of the vehicle-charging network cluster is determined and executed based on the current local variables.

[0005] The embodiments of the present invention have at least the following beneficial effects: This invention dynamically adjusts the penalty parameters and convergence threshold of the ADMM standard solvable model based on the statistical characteristics of the operational data of the vehicle-pile network cluster. It compares the changes in demand response power commands at adjacent time points to determine the degree of command change; and determines the prediction deviation based on the error between the rated power, actual load power, and predicted load power at the power supply substation. Based on the degree of command change, the penalty parameters of the ADMM standard solvable model are iteratively updated. When the demand response power command changes drastically, the penalty parameters are increased forward to accelerate the recovery of the original feasibility and the convergence threshold is relaxed to allow for early shutdown. When the command is stable, the threshold is tightened to ensure accuracy, achieving a dynamic trade-off between response speed and solution accuracy. This avoids the defects of static parameters, where re-iteration convergence is required after each command update and the control command is severely lagging. Using low-complexity statistical characteristics as the adjustment basis, the adjustment can be completed online in real time at the substation edge gateway. While retaining the distributed solution architecture of ADMM, which decomposes the problem into local subproblems by cluster, and avoiding the dimensionality curse and single point of failure risks of centralized optimization, the dynamic adjustment of the vehicle-pile-network cluster operation characteristics is embedded into a closed loop of alternating iterations of local subproblems, global subproblems, and dual variables. This allows the penalty parameters and convergence thresholds to be updated in a rolling manner with the vehicle-pile-network operation status, thereby improving the ability to quickly and accurately track demand response commands in large-scale electric vehicle access scenarios and effectively ensuring the safe and stable operation of the power distribution network. Attached Figure Description

[0006] Figure 1 This is a flowchart of a vehicle-charging network cluster collaborative control method considering demand response provided in an embodiment of the present invention; Figure 2 This is a flowchart illustrating the steps involved in providing iterative penalty parameters according to an embodiment of the present invention. Detailed Implementation

[0007] The following description, in conjunction with the accompanying drawings, details the specific scheme of the vehicle-charging network cluster collaborative control method and system that takes into account demand response provided by the present invention.

[0008] Please see Figure 1 The diagram illustrates a flowchart of a vehicle-charging station network cluster collaborative control method considering demand response, according to an embodiment of the present invention. The method includes the following steps: Step S100: Construct a standard solvable ADMM model containing local variables, global variables, and consistency constraints, and construct the augmented Lagrangian function.

[0009] The centralized vehicle-charging-network cluster collaborative regulation and optimization problem is decoupled and decomposed according to the vehicle-side local decision-making and network-side global coordination, so that each charging pile / electric vehicle and its corresponding distribution station area can be solved in a distributed manner, avoiding the computational bottleneck and data centralization problem of centralized optimization under large-scale vehicle-charging-network clusters.

[0010] The objective function of the vehicle-charging-network cluster coordinated regulation includes: charging / discharging power plan tracking deviation and battery state of charge (SOC) balance; the constraints include: charging power upper and lower limits, discharging power upper and lower limits, battery SOC, and demand response power command.

[0011] The power planning vector of each charging pile is used as a local variable, and the aggregated power of the vehicle-charging-network cluster is used as a global variable. The local objective function is to minimize the mean square deviation of the SOC of the electric vehicle corresponding to each charging pile, and the global objective function is to minimize the total power fluctuation of the vehicle-charging-network cluster.

[0012] Construct a standard solvable ADMM model that includes local variables, global variables, and consistency constraints.

[0013] As an embodiment of the present invention, the ADMM standard solvable model is as follows: ; Wherein, the local variable x is the power plan vector of each charging pile under any charging pile cluster, and takes the charging and discharging power sequence of each charging pile at all times during the power dispatch period; the global variable z is the coupling amount of the total power of all vehicle-charging network clusters, which in this invention is the aggregated power of the vehicle-charging network cluster in the area to be coordinated and regulated, that is, the algebraic sum of the charging and discharging power of all charging piles in the area vehicle-charging network cluster at each time during the dispatch period. The objective function is a local objective function used to describe the technical indicators of the local sub-problem on the vehicle side. In this invention, it is set as the degree of consistency of the SOC of electric vehicles corresponding to each charging pile, which can be achieved by minimizing the mean square deviation. Here, g(z) is the global objective function, used to describe the technical indicators of the global sub-problem on the network side. In this invention, it is the total power fluctuation of the vehicle-pile network cluster, which can be obtained by minimizing the first-order difference. Specifically: g(z) is the sum of squares of the aggregated power differences between adjacent times in the global variable z, used to limit the impact of drastic power changes on the distribution network; A is the local coefficient matrix, which maps the local variable x to the coefficient matrix of the consistency constraint, used to aggregate the local power of each pile into a coupling quantity; B is the global coefficient matrix, which maps the global variable z to the coefficient matrix of the consistency constraint, used to express the contribution of the global variable in the coupling relationship; c is a constant vector, which is the right-hand side term of the consistency constraint, and is the zero vector in this invention. For consistency constraints, used to describe the coupling relationship between local variables and global variables, in this invention, the sum of the power of each charging pile is equal to the total power of the charging pile cluster to which it belongs; Wherein, the state of charge of the electric vehicle battery at time t for the i-th charging pile Obtained by integrating charge and discharge power: ,in, Let represent the state of charge of the electric vehicle battery at time t-1 for the i-th charging pile. The charging and discharging efficiency is obtained from the hardware calibration test of the charging pile or taken as an empirical constant. In this embodiment of the invention, it is taken as 0.95. The time interval is set as the scheduling interval, and the value is strictly aligned with the issuance cycle of the power grid scheduling demand response. In this embodiment of the invention, 15 minutes is used, which is converted to 0.25 hours when substituted into the calculation. The rated capacity of the battery is reported and obtained by the battery management system (BMS) during the vehicle plug-in handshake phase. For example, a typical value is 50kWh to 100kWh. Let t be the charging and discharging power in the local variables at time t, where charging is a positive value and discharging is a negative value; it should be noted that the charging power used here is the actual active power before normalization. This is the preset charging state limit. To prevent battery lifespan degradation caused by deep charging and discharging, a preset upper limit for the charging state is set in the embodiments of the present invention. and The value of is in the range [0,1]. As an example of the present invention, The value is set to 0.2. The value is set to 0.9. Local objective function. The expression is: ,in Let t be the average State of Charge (SOC) of each electric vehicle connected to the cluster at time t. The initial state of charge (SOC) of the electric vehicle when it connects to the i-th charging pile, i.e., the SOC value at time t=0, is measured by the battery management system (BMS) of the electric vehicle and read through the charging pile communication protocol (such as GB / T 27930).

[0014] Then, the dual variables corresponding to the consistency constraints are introduced. Construct the augmented Lagrangian function: ; in, To augment the Lagrangian function, which is used to transform the original optimization problem with coupling constraints into an augmented form of a saddle point problem; For unscaled dual variables, i.e., Lagrange multipliers, and for cumulative consistency constraint violations; This is the penalty parameter, also known as the coefficient of the quadratic penalty term. This parameter controls the weight of the penalty term in the augmented Lagrange function, and it affects the convergence speed and numerical stability of the algorithm. The squared L2 norm is used to calculate the sum of squares of all elements in the vector, and is used to measure the degree of violation of consistency constraints. It should be noted that the unscaled dual variable is also referred to as the dual variable in subsequent steps.

[0015] By using the solvable form of the standard ADMM and the augmented Lagrangian function, we can obtain the decomposed local subproblems, global subproblems, and dual variable update formulas, as well as the initial values ​​of variables required for subsequent iterations. , , Initial penalty parameters With the initial convergence threshold The local subproblem is to independently optimize the charging power plan of each charging pile under the given coordination information issued by the network side; the global subproblem is to uniformly optimize the total power coupling of the charging pile network cluster in the area to which the vehicle-charging pile network cluster belongs in this scheme.

[0016] In this embodiment of the invention, the initial penalty parameter is set to 0.1, and can be taken within the range of 0.01 to 1 according to the requirements for controlling the convergence speed of the dual variable. The initial convergence threshold includes an initial absolute convergence threshold and an initial relative convergence threshold, both of which are taken as 10% of the rated power on the power supply station side. -3 times.

[0017] In each iteration of the ADMM standard solvable model, the decomposition idea of ​​alternating directions is followed. The remaining variables are fixed, and the minimization is only performed on the current variable. The three types of variables are updated in the order of local subproblem - global subproblem - dual variable, which constitutes a complete iteration. The updates of local and global subproblems optimize the objective function along the original direction, while the updates of dual variables accumulate the violation of consistency constraints along the dual direction. The two are promoted alternately, so that the iteration sequence gradually approaches the original-dual optimal solution, which is also the global optimal solution of the original coupled optimization problem. The local variable x is the charging and discharging power plan of each charging pile in any charging pile cluster, the global variable z is the coupling amount of the total power of the vehicle-charging-network cluster, and the dual variable u is the Lagrange multiplier reflecting the degree of violation of consistency constraints.

[0018] Iterate alternately using the following well-known update formula: ; ; ; In the above formulas, k is the iteration number; the argmin function represents the value of the independent variable that minimizes the objective function; the objective function of the local subproblem adopts the quadratic form of the SOC mean square deviation constructed above, and can be solved in a closed form; when there are SOC coupling constraints, the projection gradient method or interior point method is used; the global subproblem is a convex quadratic programming problem with network loss and penalty for breach of contract, which can be solved using analytical projection or dual methods. The original residual and dual residual are calculated simultaneously. ; ; The original residual is used to measure the amount of violation of consistency constraints; The dual residual is used to measure the deviation from the feasibility of duality. Output ADMM standard solvable model number The next iteration and residuals The significance of updating local subproblems lies in decoupling the coordinated regulation and optimization problem of the global vehicle-charging-pile network cluster response demand into local subproblems that each charging pile can solve independently, thus achieving distributed computing and avoiding the curse of dimensionality in centralized optimization. The role of updating global subproblems is to uniformly optimize the total power coupling of the cluster from the power grid side of the power supply area to which the charging pile cluster belongs, setting global coupling constraints and demand response instructions, and thereby coordinating the local plans reported by each charging pile. The updating of dual variables reflects the cumulative feedback mechanism for constraint violations. It should be noted that constructing the ADMM model is a well-known technique in the art and will not be elaborated upon here.

[0019] Step S200: Obtain the operation data of the vehicle-charging network cluster. The operation data includes: demand response power command, actual load power, and rated power of the power supply station area.

[0020] After constructing the ADMM standard solvable model, relevant operational data of the vehicle-charging-network cluster are obtained, including: charging load data, demand response command data, predicted load power and prediction error data, and power supply station side operational data. This charging load data is also the actual load power at the charging pile level.

[0021] This demand response command data, also known as the demand response power command, is the total power tracking target for the power supply area of ​​the vehicle-charging network cluster, issued by the dispatch center of the power supply area. The operating data for this power supply area includes: transformer load rate, three-phase voltage / current, active / reactive power, and rated power of the power supply area.

[0022] The actual load power is collected through the built-in metering unit of smart energy meters or charging piles, with a sampling interval of 1 second and an aggregation upload setting of 15 minutes. This actual load power is active power. After obtaining the actual load power, data preprocessing is performed, including: clock synchronization (NTP / GPS), marking and removing outliers using the interquartile range (IQR) method, and filling missing values ​​using linear interpolation. The power data is uniformly divided by the rated power of the power supply area to facilitate cross-time period comparison.

[0023] Demand response power commands are collected through the power grid dispatch channel and updated every 15 minutes. Emergency response can be set to minute level. After receiving the demand response power command, data preprocessing is performed, including: clock synchronization (NTP / GPS), marking and removing outliers using the interquartile range (IQR) method, and filling missing values ​​using linear interpolation. Power data is uniformly divided by the rated power of the power supply area to facilitate cross-time period comparison.

[0024] The predicted load power is obtained through the energy management system of the distribution area to which the vehicle-charging network cluster belongs. Specifically, the predicted load power is calculated using the EMA exponential moving average method based on the actual load power in the historical dataset. After obtaining the predicted load power, data preprocessing is performed, including: clock synchronization (NTP / GPS), outlier marking and removal using the interquartile range (IQR) method, and filling missing values ​​using linear interpolation; the power data is uniformly divided by the rated power of the power supply area to facilitate cross-time period comparison. In this embodiment of the invention, when calculating the predicted load power using the EMA exponential moving average method, the smoothing coefficient is set between 0.1 and 0.3. As a reference example, the smoothing coefficient is set to 0.2 in this invention.

[0025] The rated power of the power supply area is a pre-set basic parameter of the equipment. The nameplate capacity value of the distribution transformer is directly retrieved through the cloud-based energy management system (EMS) as the rated power of the power supply area.

[0026] In this invention, the sensors and acquisition devices adopt a hierarchical distributed deployment of "pile-area-platform level": at the pile level, the charging power is collected locally by the built-in metering unit and communication module of the charging pile; the communication module can be 4G / 5G, Ethernet, or HPLC broadband carrier; at the area level, a distribution transformer monitoring terminal (TTU) and an edge computing gateway are deployed on the low-voltage side of the distribution room / substation to aggregate pile-level data, access demand response commands and predicted values, and complete feature extraction and update calculations of penalty parameters and convergence thresholds locally at the edge; at the platform level, the cloud-based energy management system (EMS) is responsible for coordinating multiple areas and unifying the interaction of demand response commands. This deployment makes the feature extraction and parameter update hierarchy consistent with the ADMM "local subproblem - global subproblem" distributed solution hierarchy, avoiding the latency of centralized data transmission.

[0027] Step S300: Compare the changes in demand response power commands at adjacent time points to determine the degree of command change; determine the prediction deviation based on the error between the rated power, actual load power, and predicted load power at the power supply station; and iteratively update the penalty parameters of the ADMM standard solvable model based on the degree of command change.

[0028] To address the shortcomings of traditional ADMM standard solvable models in the alternating direction multiplier method, where static penalty parameters and fixed convergence thresholds are difficult to adapt to the dynamic changes in vehicle-pile network cluster constraints in response to demand commands, leading to lags in coordinated regulation, this invention utilizes real-time operational data of the vehicle-pile network cluster to quantify the degree of dynamic change in constraints and the level of data uncertainty. Based on this, the penalty parameters and convergence thresholds are adaptively adjusted in each iteration.

[0029] The adjustment strategy for the penalty parameter and convergence threshold of the ADMM standard solvable model is as follows: when the external demand response changes drastically, the penalty parameter is increased in a feedforward manner to accelerate the recovery of the original feasibility, and the convergence threshold is appropriately relaxed to allow for early shutdown response; conversely, when the external demand response changes smoothly and the prediction error is small, the penalty parameter is kept small and the convergence threshold is tightened to ensure the solution accuracy.

[0030] Please see Figure 2 , Figure 2 The flowchart for the iterative penalty parameter method includes the following steps: Step S310: Compare the changes in demand response power commands at adjacent time points to determine the degree of command change; determine the prediction deviation based on the error between the rated power, actual load power, and predicted load power at the power supply station.

[0031] The more drastic the changes in the demand response power command issued by the power grid in the power supply area, the more timely the adjustment of the penalty parameters in the ADMM standard solvable model should be.

[0032] A sliding window is set, and the number of sampling points within the sliding window is set to a range of 8 to 12. In this embodiment of the invention, 8 is used, which corresponds to a sliding time range of 2 to 3 hours.

[0033] For the sliding window corresponding to the current time, calculate the average of the absolute values ​​of the differences between every two adjacent demand response power commands as the average power change. More specifically: take the current time as the time corresponding to the last data in the sliding window, take the sequence of demand response power commands in the sliding window, calculate the absolute value of the difference between two adjacent values ​​in the sequence, and then average the absolute values ​​of the differences between all two adjacent values ​​in the sequence to obtain the average power change.

[0034] The 95th percentile of the historical one-day average power change at the current moment is used as the reference average change. The degree of command change is determined by comparing the average power change with the reference average change. More specifically, the ratio of the average power change (numerator) to the reference average change (denominator) is used as the degree of command change. It should be noted that if the calculated reference average change is zero, then the reference average change should be directly assigned the value of 0.1 to avoid the problem of the denominator being zero.

[0035] The greater the degree of change in the command, the more drastic the change in the demand response power command issued by the power grid in the power supply area within the neighborhood time range. In this case, the degree of adjustment of the penalty parameter in ADMM during the iteration process should be increased in a timely manner, and the convergence threshold should be appropriately relaxed to enable more timely shutdown response.

[0036] Therefore, further, based on the error between the rated power, actual load power, and predicted load power at the power supply substation, the prediction deviation is determined: First, based on the error between the actual load power and the predicted load power, the root mean square error of the load prediction for the distribution area is determined; then, by combining the root mean square error of the load prediction and the rated power on the power supply distribution area side, the prediction deviation is determined.

[0037] As a specific embodiment of the present invention, taking any sliding window t as an example, the formulas for calculating the root mean square error of the load prediction and the prediction deviation are as follows: ; ; in, Let be the root mean square error of the load forecast for the sliding window t; This represents the number of sampling points within the sliding window t; , These represent the actual load power and the predicted load power at the j-th sampling point within the sliding window t, respectively. The normalization ratio is set to the 95th percentile of the root mean square error of the normalized load forecast within the historical 1-day range in this invention. denoted as the prediction bias of the sliding window t; min is the function for finding the minimum value.

[0038] When prediction deviation The larger the value, the larger the relative normalization coefficient of the root mean square error (RMSE) of the load forecast error of the transformer area within the sliding window t, which means that the load forecast accuracy of the current period is worse and the forecast uncertainty is higher.

[0039] Step S320: Based on the degree of instruction change, iteratively update the penalty parameters of the ADMM standard solvable model.

[0040] The physical reliability and accuracy of the solution to the optimization problem are limited by the noise level of the data / model. If the convergence threshold is maintained for iteration, it will lead to a waste of computing resources and increase the delay in the issuance of control commands. In this case, the convergence threshold should be relaxed appropriately to stop the machine in time, output control commands in time, and improve the timeliness of demand response.

[0041] First, based on the comparison of the magnitudes of the original residuals and dual residuals of the ADMM standard solvable model, the direction for adjusting the penalty parameter is determined.

[0042] when When the penalty parameter is set to be adjusted in the direction of increase, that is, the penalty parameter adjustment index is set to a positive value; when When the penalty parameter is set to decrease, the adjustment direction is set to negative.

[0043] After determining the adjustment direction, a penalty parameter adjustment index is constructed based on the degree of change in the instruction. Based on the penalty parameter adjustment index, the penalty parameter is iteratively updated to obtain the penalty parameter corresponding to the current iteration. More specifically, the product of the penalty parameter adjustment index raised to the power of the basic adjustment step size and the penalty parameter is used as the penalty parameter obtained after the current iteration. The penalty parameter obtained after iteration is limited to the lower limit and the upper limit of the penalty parameter.

[0044] In a specific embodiment of the present invention, the formula for iteratively updating the penalty parameter is: ; ; In the formula, Adjust the exponent for the penalty parameter after the k-th iteration; The penalty parameter is the value after the k-th iteration. This is the penalty parameter for the (k+1)th iteration; The original residual for the k-th iteration; Let be the dual residual of the k-th iteration; The step size is adjusted based on the basic principle, and in this embodiment of the invention, it is set to 2. As the dynamic gain factor, when When the value is 0, it degenerates into the known residual balance method, with the value range set to [0,2]. In this embodiment of the invention, it is set to 1.5. Implementers can adjust it themselves according to simulation modeling or on-site tuning. Value; μ is the residual imbalance judgment threshold. μ is greater than 1. In order to avoid frequent invalid adjustments caused by noise, it is set to 10 in this embodiment of the invention. , These are the lower limit and upper limit of the penalty parameter, respectively, which need to be obtained according to the regulation of the vehicle-pile network control in the substation area. In this embodiment of the invention, the lower limit of the penalty parameter is taken as 1 / 5 of the initial penalty parameter, and the upper limit of the penalty parameter is taken as 5 times the initial penalty parameter.

[0045] Step S400: Combining the degree of instruction change and the prediction deviation, the convergence threshold of the ADMM standard solvable model is iteratively updated; during the iterative update, the original residual and the dual residual are updated simultaneously.

[0046] The convergence thresholds include: absolute convergence thresholds and relative convergence thresholds.

[0047] The degree of instruction change and the degree of prediction deviation are weighted based on a preset relaxation coefficient, and the convergence threshold is iterated. The iteratively updated convergence threshold is then limited to a base value and a preset upper limit of convergence.

[0048] In a specific embodiment of the present invention, the formula for iteratively updating the convergence threshold is: ; ; in, It is the absolute convergence threshold after the (k+1)th iteration update. Its dimension is the same as that of the residual. It is used to limit the absolute magnitude of the residual operation and is the basic component for the final shutdown decision. The larger the value, the more lenient the shutdown condition. The relative convergence threshold after the (k+1)th iteration update, and the absolute convergence threshold. Together, they constitute the shutdown judgment condition, so that the threshold takes into account both the magnitude and the relative size of the variables; clip() is a cutoff (limiting) function that restricts the convergence threshold after iterative update to between the baseline value and the preset convergence upper limit; The reference value for the absolute convergence threshold is [value], which is [value] relative to the rated power of the power supply substation. In the embodiments of the present invention, the multiple is taken as... , This is the baseline value for the relative convergence threshold, and its range is [value missing]. In the embodiments of the present invention, ; , These are the demand response fluctuation slack coefficient and the load forecasting error slack coefficient, respectively. , Both are greater than or equal to 0, reflecting the contribution of the intensity of the change in the control command and the level of prediction error to the relaxation of the convergence threshold, respectively. In the embodiments of this invention... and The value range is limited to [0,5], and is obtained through offline simulation or calibration of historical operating data of the transformer area. Under the premise that the total power scheduling deviation of the transformer area cluster does not exceed 5% of the rated power of the transformer area, the set of parameter combinations that minimizes the average number of convergence iterations of ADMM is selected as an example of the present invention. The value is 2.0. The value is 1.5; in this embodiment of the invention, the demand response fluctuation relaxation coefficient and the load forecast error relaxation coefficient are collectively referred to as the preset relaxation coefficient; , These are the upper limits of the absolute convergence threshold and the upper limit of the relative convergence threshold, respectively, to prevent excessive relaxation. In this embodiment of the invention, they are set to 5% of the acceptable proportion of the rated power of the transformer area aggregation. The preset convergence upper limits include: the upper limit of the absolute convergence threshold and the upper limit of the relative convergence threshold; The degree of change in instructions; To predict the deviation.

[0049] Each time the penalty parameter and convergence threshold are iteratively updated, the original residual and the dual residual are updated synchronously.

[0050] Based on the convergence threshold after iterative updates, and combining the dimension of local variables, the dimension of consistency constraints, and the dual variable in the augmented Lagrangian function, the original residual stopping threshold and the dual residual stopping threshold are calculated.

[0051] Original residuals: Used to measure the amount of consistency constraint violation; dual residual: It is used to measure the deviation of dual feasibility.

[0052] ; ; The stopping threshold for the original residual in the (k+1)th iteration is the maximum value that the original residual r is allowed to reach. is the stopping threshold for the dual residual in the (k+1)th iteration, i.e., the maximum value that the dual residual s is allowed to reach; p is the dimension of the consistency constraint, i.e., the number of elements in the original residual vector r, which is equal to the number of constraint equations Ax+Bz=c. Used to convert the magnitude of a single-element residual to the norm of the entire residual vector; n is the dimension of the local variable x, that is, the total number of elements in the power plan vector of each charging pile; Used for order-of-magnitude conversion of dual residual thresholds; is the unscaled dual variable of the (k+1)th iteration, i.e., the standard Lagrange multiplier, used for dual threshold magnitude estimation; max is the maximum value function.

[0053] Based on the above steps, the updated penalty parameters, absolute convergence threshold, relative convergence threshold, and parameters within the sliding window t for the (k+1)th iteration can be obtained. , The penalty parameter in the (k+1)th iteration serves as the coefficient of the quadratic penalty term in the subproblems of the local variable x and the global variable z in the next round, as well as the penalty parameter in the calculation of the dual residual. The updated absolute convergence threshold and relative convergence threshold are used to generate the shutdown threshold.

[0054] Step S500: When both the original residual and the dual residual meet the convergence condition or reach the maximum number of iterations, stop the iteration, determine the charging and discharging power control command of the vehicle-charging network cluster based on the current local variables, and issue it for execution.

[0055] The stopping criterion is whether both the original residual and the dual residual simultaneously satisfy the convergence condition. If they do, a control command is output and executed; otherwise, iteration continues. Simultaneously, the iteration count is set to force termination when the maximum number of iterations is reached. More specifically: Determine whether the norm of the original residual after iterative update is less than or equal to the original residual stopping threshold, and whether the norm of the dual residual is less than or equal to the dual residual stopping threshold. If satisfied, terminate the iteration, multiply the current local variable by the rated power of the power supply area, and use the result as the charging and discharging power control command for the vehicle-charging network cluster, then issue and execute it; when At this time, the convergence condition must be met.

[0056] If the condition is not met, return to continue iterating and updating until the convergence condition is met or the maximum number of iterations is reached.

[0057] When the above shutdown criteria are met, output , that is, the local power plan vector obtained by the i-th charging pile cluster after the k+1th iteration convergence, that is, the active power sequence of each charging pile in the cluster during the scheduling period is used as the cluster charging and discharging power control instruction and issued for execution, otherwise the iteration continues until the convergence condition is met or the maximum number of iterations is reached, and the iteration stops; finally, the output vehicle-charging-network cluster charging and discharging power control instruction is realized, and this round of coordinated control is completed.

[0058] An embodiment of the present invention provides a vehicle-charging-network cluster collaborative control system that takes into account demand response, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the vehicle-charging-network cluster collaborative control method that takes into account demand response.

Claims

1. A method for coordinated control of vehicle-charging network clusters considering demand response, characterized in that, The method includes the following steps: Construct a standard solvable ADMM model that includes local variables, global variables, and consistency constraints, and construct an augmented Lagrangian function; Obtain operational data of the vehicle-charging network cluster, including: demand response power command, actual load power, and rated power of the power supply station area; By comparing the changes in demand response power commands at adjacent time points, the degree of command change is determined; based on the error between the rated power, actual load power, and predicted load power at the power supply station, the prediction deviation is determined; and based on the degree of command change, the penalty parameters of the ADMM standard solvable model are iteratively updated. The convergence threshold of the ADMM standard solvable model is iteratively updated by combining the degree of instruction change and the prediction deviation; during the iterative update, the original residual and the dual residual are updated simultaneously. When both the original residual and the dual residual satisfy the convergence condition or reach the maximum number of iterations, the iteration stops, and the charging and discharging power control command of the vehicle-charging network cluster is determined and executed based on the current local variables.

2. The vehicle-charging-network cluster collaborative control method considering demand response as described in claim 1, characterized in that, The construction of the standard solvable ADMM model, which includes local variables, global variables, and consistency constraints, and the construction of the augmented Lagrangian function, includes: The problem of centralized vehicle-charging-network cluster coordinated regulation and optimization is decoupled, with the power plan vector of each charging pile as the local variable and the aggregated power of the vehicle-charging-network cluster as the global variable. The local objective function is to minimize the mean square deviation of the SOC of electric vehicles corresponding to each charging pile, and the global objective function is to minimize the total power fluctuation of the vehicle-charging-network cluster. A consistency constraint is established that the sum of the power of each charging pile equals the total power of its cluster, and an augmented Lagrangian function containing a quadratic penalty term is constructed by introducing dual variables.

3. The vehicle-charging-network cluster collaborative control method considering demand response as described in claim 1, characterized in that, The step of comparing the changes in demand response power commands at adjacent time points to determine the degree of command change includes: Set a sliding window. For the sliding window corresponding to the current moment, calculate the average of the absolute values ​​of the differences between every two adjacent demand response power commands, and use it as the average power change. By comparing the average power change with the reference average power change, the degree of command change is determined.

4. The vehicle-charging-network cluster collaborative control method considering demand response as described in claim 1, characterized in that, The determination of prediction deviation based on the error between the rated power, actual load power, and predicted load power of the power supply area includes: Based on the error between the actual load power and the predicted load power, the root mean square error of the load prediction for the transformer area is determined. The prediction deviation is determined by combining the root mean square of the load prediction error and the rated power of the power supply area.

5. The vehicle-charging-network cluster collaborative control method considering demand response as described in claim 1, characterized in that, The iterative update of the penalty parameters of the ADMM standard solvable model based on the degree of instruction change includes: Based on the comparison of the magnitudes of the original residuals and dual residuals of the ADMM standard solvable model, the adjustment direction of the penalty parameter is determined; after determining the adjustment direction, the penalty parameter adjustment index is constructed based on the degree of change of the instruction. The penalty parameter is adjusted based on the penalty parameter adjustment index, and the penalty parameter is iteratively updated to obtain the penalty parameter corresponding to the current iteration.

6. The vehicle-charging-network cluster collaborative control method considering demand response as described in claim 5, characterized in that, The step of adjusting the exponent based on the penalty parameter and iteratively updating the penalty parameter to obtain the penalty parameter corresponding to the current iteration includes: The product of the penalty parameter adjusted to the power of the basic adjustment step size and the penalty parameter is used as the penalty parameter obtained after the current iteration; wherein, the penalty parameter obtained after the iteration is restricted to the lower limit and the upper limit of the penalty parameter.

7. The vehicle-charging-network cluster collaborative control method considering demand response as described in claim 1, characterized in that, The iterative update of the convergence threshold of the standard solvable ADMM model, combining the degree of instruction variation and the prediction bias, includes: The convergence thresholds include: absolute convergence thresholds and relative convergence thresholds; The degree of instruction change and the degree of prediction deviation are weighted based on a preset relaxation coefficient, and the convergence threshold is iterated. The iteratively updated convergence threshold is then limited to a base value and a preset upper limit of convergence.

8. The vehicle-charging-network cluster collaborative control method considering demand response according to claim 1, characterized in that, After iteratively updating the convergence threshold of the standard solvable ADMM model by combining the degree of instruction change and the prediction bias, the method further includes: Based on the convergence threshold after iterative updates, and combining the dimension of local variables, the dimension of consistency constraints, and the dual variable in the augmented Lagrangian function, the original residual stopping threshold and the dual residual stopping threshold are calculated.

9. The vehicle-charging-network cluster collaborative control method considering demand response as described in claim 8, characterized in that, When both the original residual and the dual residual satisfy the convergence condition or reach the maximum number of iterations, the iteration stops, and the charging and discharging power control command of the vehicle-charging network cluster is determined and executed based on the current local variables, including: Determine whether the norm of the original residual after iterative update is less than or equal to the original residual stopping threshold, and whether the norm of the dual residual is less than or equal to the dual residual stopping threshold. If satisfied, terminate the iteration, multiply the current local variable by the rated power of the power supply area, and use the result as the charging and discharging power control command for the vehicle-charging network cluster and issue it for execution. If the condition is not met, return to continue iterating and updating until the convergence condition is met or the maximum number of iterations is reached.

10. A vehicle-charging-network cluster collaborative control system considering demand response, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the vehicle-charging network cluster collaborative control method that takes into account demand response as described in any one of claims 1 to 9.

Images