Power scheduling system for charging pile groups based on distributed consensus mechanism

Abstract

This invention belongs to the field of power system and charging facility management technology. It addresses the problem that existing technologies cannot effectively handle the uncertainty caused by the lag in state data during the critical transition zone of master node switching in distributed consensus mechanisms. Specifically, it is a charging pile group power scheduling system based on a distributed consensus mechanism, including: a data acquisition module, a freshness perception module, a dynamic parameter generation module, a theoretical optimization module, an individual step size constraint module, and an instruction generation module. This invention independently calculates the power change rate and criticality weight of each charging pile, thereby generating a differentiated maximum adjustment step size. While protecting critical piles (such as low-power or high-power piles) from large-scale adjustment interference, it allows piles with stable state and fresh data to quickly respond to optimization instructions, achieving coordinated optimization of global smoothness and individual response speed.

Description

Technical Field

[0001] This invention belongs to the field of power system and charging facility management technology, specifically a charging pile group power scheduling system based on a distributed consensus mechanism. Background Technology

[0002] In distributed charging management systems, to meet the power coordination needs of large-scale charging pile groups, a cluster architecture based on consensus algorithms (such as Raft) is typically adopted to ensure global consistency and high availability of scheduling decisions. Under this architecture, the cluster elects a master node, which is responsible for collecting the operating status of each charging pile in real time, running power optimization algorithms, and issuing scheduling instructions to each pile. When the master node fails due to network fluctuations, excessive load, or planned maintenance, the cluster will trigger a re-election, and the newly elected master node will take over the scheduling task.

[0003] Existing technologies typically assume that once a new master node is elected, its locally stored state data instantly reflects the real-world situation. Based on this assumption, the new master node immediately activates a global optimization algorithm and directly generates and issues new power commands based on the delayed state data. However, engineering practices in industrial settings reveal significant blind spots in this logic: charging pile status reporting usually employs asynchronous communication, and data updates do not depend on the primary replication path of the consensus log. This results in the new master node holding state data that is often delayed by hundreds of milliseconds or even seconds at the moment of takeover. When the cluster size is large or communication links are congested, the delay will be further extended.

[0004] This kind of "instant takeover" behavior based on outdated data can trigger serious chain reactions in the real physical world. The new master control node may misjudge the charging pile that has been reduced in current as a high-load state, thereby imposing unnecessary power restrictions on other piles and wasting transformer capacity. Conversely, it may also misjudge the charging pile that has been upgraded as an idle state and assign high-power commands to it, causing local overload tripping. More seriously, there may be a huge step between the first scheduling command and the command before the switch, which will cause electrical shock to the grid equipment and even trigger a secondary switch, causing the system to fall into a long-term oscillation.

[0005] Therefore, how to effectively cope with the uncertainty caused by the lag in state data during the critical transition zone of the master node switching of the distributed consensus mechanism, and avoid the step shock and oscillation of power allocation, has become the core technical problem that must be solved to improve the robustness of the charging pile group power scheduling system. Summary of the Invention

[0006] The purpose of this invention is to provide a power scheduling system for charging pile groups based on a distributed consensus mechanism, which solves the problem that existing technologies cannot effectively cope with the uncertainty caused by the lag in state data during the critical transition zone of the master control node switching in a distributed consensus mechanism.

[0007] The technical problem to be solved by this invention is: how to provide a power scheduling system for charging pile groups based on a distributed consensus mechanism that can effectively cope with the uncertainty caused by the lag in state data during the critical transition zone of the master node switching in the distributed consensus mechanism.

[0008] The objective of this invention can be achieved through the following technical solutions:

[0009] A power scheduling system for charging pile groups based on a distributed consensus mechanism includes:

[0010] The data acquisition module responds to the newly elected master node in the distributed consensus cluster by acquiring the status data of each charging pile locally cached by the new master node of the charging pile group power scheduling system based on the distributed consensus mechanism, and obtaining the base power instruction set corresponding to the last scheduling cycle before the switch from the consensus log.

[0011] The freshness perception module uses the timestamps carried by the status data of each charging pile in the charging pile group power scheduling system based on the distributed consensus mechanism and the current system time to calculate the global freshness index, which characterizes the overall staleness of the local cache data of the new master node in the charging pile group power scheduling system based on the distributed consensus mechanism, and the rate of change of the global freshness index of the charging pile group power scheduling system based on the distributed consensus mechanism over time.

[0012] The dynamic parameter generation module generates a first control parameter and a second control parameter based on the global freshness index of the charging pile group power scheduling system based on the distributed consensus mechanism and its rate of change over time. The first control parameter of the charging pile group power scheduling system based on the distributed consensus mechanism is used to constrain the change range of the scheduling instruction relative to the reference power instruction set of the charging pile group power scheduling system based on the distributed consensus mechanism. The second control parameter of the charging pile group power scheduling system based on the distributed consensus mechanism is used to adjust the confidence level of the theoretical optimal scheduling result relative to the reference power instruction set of the charging pile group power scheduling system based on the distributed consensus mechanism.

[0013] The theoretical optimization module executes a preset power scheduling optimization algorithm based on the status data of each charging pile in the charging pile group power scheduling system based on a distributed consensus mechanism to obtain the theoretically optimal power instruction set.

[0014] The individual step size constraint module calculates the maximum allowable adjustment step size for each charging pile within the current scheduling cycle, based at least on the state data lag and power change trend of that charging pile.

[0015] The instruction generation module generates the final power instruction set for the current scheduling cycle based on the baseline power instruction set, theoretical optimal power instruction set, first control parameters, second control parameters, and the maximum adjustment step size of each charging pile in the charging pile group power scheduling system based on a distributed consensus mechanism, and sends it to the corresponding charging pile for execution.

[0016] The present invention has the following beneficial effects:

[0017] 1. The freshness perception module quantifies the data lag of each charging pile in real time and aggregates it into a global freshness index. Combined with the dynamic parameter generation module, this index is mapped to the first control parameter and the second control parameter, realizing global constraints on the magnitude of changes in scheduling instructions and dynamic adjustment of the confidence level of theoretical optimization results. This mechanism avoids the step shock of instructions caused by directly applying old data at the moment of switching to a new master control node, which is a problem in traditional methods. It ensures that the power allocation can smoothly and gradually converge to the theoretical optimal value as the data freshness is restored within the critical transition zone, thereby effectively suppressing the power oscillation and transformer overload risk caused by state lag.

[0018] 2. By using an individual step size constraint module, the power change rate and criticality weight of each charging pile are calculated independently, thereby generating a differentiated maximum adjustment step size. This step size comprehensively reflects the constraint requirements of data lag, power fluctuation intensity, and battery state of charge on the adjustment range. Compared with the conventional approach of uniform limit, this solution can protect critical piles (such as low-power or high-power piles) from large adjustment interference while allowing piles with stable state and fresh data to quickly respond to optimization instructions, achieving coordinated optimization of global smoothness and individual response speed.

[0019] 3. A linear programming model with transformer capacity as the core constraint and priority coefficient as the optimization objective is constructed through the theoretical optimization module. A dual constraint of confidence weight and individual step size truncation is introduced in the instruction generation module to make the final power instruction as close as possible to the theoretical optimal solution while meeting the safety boundary. During soft switching, this design prioritizes the charging needs of high-priority charging piles. At the same time, the step size constraint prevents local overload caused by excessive adjustment in a single step. Thus, in the critical area with high data uncertainty, the safe utilization of transformer capacity and the fair satisfaction of charging needs are achieved.

[0020] 4. By comprehensively evaluating the global freshness index, its rate of change, and the difference between the theoretically optimal command and the baseline command through the exit judgment module, the soft handover mode is terminated and the normal scheduling is restored only when the data is sufficiently fresh, the dynamics tend to be stable, and the commands converge. This mechanism avoids command shocks caused by premature exit, and at the same time, the timeout forced exit mechanism prevents the system from being stuck in the transition state indefinitely, ensuring the high availability and robustness of the distributed charging network in the master node switching scenario. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is an overall system block diagram of the present invention. Detailed Implementation

[0023] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. 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.

[0024] In a power scheduling system for charging pile groups based on a distributed consensus mechanism, ensuring the continuity of power allocation at the moment of master node switching is considered a key capability to guarantee the stability of charging services. This continuity essentially depends on the fact that the state data held by the new master node can truly reflect the real-time operating conditions of each charging pile, thereby quickly generating scheduling instructions that match the physical world after the switch.

[0025] However, existing technologies lack a quantitative perception mechanism for the degree of data lag and its dynamic evolution trend during the handover process. This makes it impossible to accurately identify scheduling decision deviations introduced by the new master node in the initial stage of takeover due to outdated data. Specifically, data lag manifests as a difference between the state timestamp cached locally by the new master node and the current physical time, and this difference can reach hundreds of milliseconds or even several seconds at the moment of handover. Meanwhile, the dynamic evolution trend of the lagging data is manifested as the rate of change of the freshness of the charging pile status data over time. This rate implies key information about the quality of communication link recovery and the progress of data synchronization. As a result, a strict time-series correspondence cannot be established between the scheduling instructions of the new master node and the actual operating status of the charging piles, causing the system to misjudge the conservatism or aggressiveness of power allocation after the handover, thereby affecting the smooth transition of scheduling instructions and the safe utilization of transformer capacity.

[0026] For example, in the daily operation of large charging stations, when the master control node is switched due to network fluctuations, the newly elected master control node can only obtain the baseline instructions before the switch through the regular consensus log recovery, but cannot perceive the real-time power change trend of each charging pile at the moment of switch. Furthermore, when some charging piles are in the process of rapid load increase or decrease due to changes in vehicle demand, the system only records their lagging static power values ​​and fails to monitor the overall recovery speed of data freshness and the dynamic response requirements of individual charging piles. Specifically, the system may misjudge a charging pile as idle based on outdated data and assign it a high-power instruction, or misjudge a charging pile that is being loaded as having a high load and impose unnecessary power restrictions, thereby causing transformer capacity waste or local overload tripping.

[0027] If the above problems are not resolved, the scheduling system will continue to lose its ability to objectively judge the uncertainty of data in the critical switching zone. Among them, the failure to quantify data lag will cause the new master node to generate instructions based on erroneous states, triggering a step shock in power allocation. At the same time, the failure to capture the trend of freshness changes will prevent the system from predicting the progress of data synchronization, causing a mismatch between the convergence speed of scheduling instructions and the data recovery speed, ultimately making the power allocation process exhibit oscillating characteristics. As a result, the inaccuracy of switching decisions will systematically hinder the system from achieving a smooth transition from the old master node to the new master node, affecting the high availability and robustness of the distributed charging network.

[0028] Example 1: As Figure 1 As shown, a power scheduling system for charging pile groups based on a distributed consensus mechanism includes:

[0029] The data acquisition module responds to the newly elected master node in the distributed consensus cluster by acquiring the status data of each charging pile cached locally by the new master node and obtaining the baseline power instruction set corresponding to the last scheduling cycle before the switch from the consensus log.

[0030] The data acquisition module serves as the data entry point for the entire power scheduling system. Its core responsibility is to reliably collect all the initial data required for subsequent calculations at critical moments when the master node switches in the distributed consensus cluster. This module's implementation is not simply data retrieval; rather, it is specifically designed to address the asynchronous communication characteristics and consensus mechanism features of industrial environments.

[0031] Specifically, the system continuously monitors the status of the current master node in the distributed consensus cluster (such as a cluster built on the Raft protocol). When a re-election is triggered due to network partitioning, excessive node load, or planned maintenance, and the original slave node is successfully elected as the new master node, the data acquisition module is activated. This activation event itself is the sign that the system has entered soft switchover mode.

[0032] After activation, the data acquisition module performs two types of data collection operations in parallel. The first type of collection operation targets the real-time state cache maintained locally by the new master node. During the normal operation of the distributed consensus cluster, all nodes (including the Leader and Followers) continuously receive operational data reported by each charging pile through independent asynchronous channels and maintain a real-time state table locally. This table contains at least the following fields for each online charging pile: charging pile identifier (used to uniquely identify the physical pile), the system timestamp of the latest report, and the instantaneous output power value at that moment. In actual engineering scenarios, to support subsequent criticality analysis, this state table can also be expanded to include parameters such as battery state of charge (SOC) and maximum available power. Since this data is transmitted through a channel independent of the consensus log, when a new master node is first elected, some data in its local cache may be hundreds of milliseconds or even several seconds behind.

[0033] The second type of data acquisition operation targets the consensus log. During steady-state system operation, the final power instruction set of each scheduling cycle is persistently stored as a consensus log entry on all nodes. The data acquisition module reads the most recently submitted scheduling instruction record from the consensus log, i.e., the power instruction set issued and successfully executed by the original master node in the last scheduling cycle before the switchover, and uses this as the benchmark for subsequent calculations. This benchmark instruction set contains the power target value corresponding to the same charging pile identifier, denoted as... , where i iterates through all charging stations.

[0034] To ensure that the subsequent step-size constraint module can accurately evaluate the dynamic characteristics of each charging pile, the data acquisition module, after completing the above two types of data collection, also needs to perform an initialization operation: creating a first-in-first-out historical power queue for each charging pile. This queue is used to store the power values ​​reported by the pile in the most recent Q times (e.g., Q is 10) and their corresponding reporting timestamps, providing a data basis for subsequent calculation of the power change rate. Upon initial activation, if historical data is insufficient, the queue can be initialized with repeated filling of the latest acquired power value, or it can be gradually filled in subsequent scheduling cycles.

[0035] For charging piles that fail to report data for multiple consecutive cycles (e.g., 3 cycles), the system marks them as offline and removes them from subsequent global freshness calculations and instruction generation. When the charging pile re-reports data, the system automatically adds it back to the online set and resets its historical queue to the current single-point data.

[0036] In summary, the output of the data acquisition module consists of three parts: first, the latest status data of each charging pile (including lag information); second, the baseline power instruction set serving as the scheduling anchor point; and third, the historical power queue initialized for each charging pile. These data collectively form the input basis for the subsequent freshness perception, theoretical optimization, and individual step size constraint modules, ensuring that the entire system has a quantifiable and traceable data starting point in the switching critical zone.

[0037] The freshness perception module calculates the global freshness index and the rate of change of the global freshness index over time based on the timestamps carried by the status data of each charging pile and the current system time.

[0038] The freshness perception module receives the latest status data of each charging pile from the data acquisition module. This data includes at least the charging pile identifier, reporting timestamp, and current output power value. The core task of this module is to quantify the lag of the locally cached data of the new master node, generate a global freshness index and its rate of change over time, and provide input for the subsequent generation of dynamic parameters.

[0039] In practice, the system triggers a freshness-aware calculation every scheduling cycle (e.g., 100 milliseconds). First, for each online charging pile, the reporting timestamp from its latest status data is obtained. and the current system time Compare and calculate the data lag time of the pile: Where i is the charging pile number. If a charging pile does not report new data in the current cycle, its most recently reported data will be used, and the lag time will be continuously accumulated based on the reporting timestamp. For example, assuming that the most recently reported time of a charging pile is 14:23:45.123, and the current system time is 14:23:45.523, then its lag time is 0.4 seconds.

[0040] Delay This reflects the freshness of the pile's state data, but directly using lag time for weighted averaging can lead to unintuitive numerical values ​​due to differences in units. Therefore, this module introduces a monotonically decreasing sigma function to map lag time to individual freshness values. Its mathematical expression is: Where α is a preset scaling factor, controlling the rate of freshness decay as lag time increases. This function... The value of α is 1, indicating that the data is completely fresh; as the lag time increases, the function value smoothly approaches 0. The value of α can be determined based on engineering experience. For example, if we set the freshness to drop to 0.5 when the lag time reaches 10 seconds, then we can solve for α. In this embodiment, we take... This ensures that the decay curve has reasonable differentiation within the common lag range; α can be adjusted within the range of [0.1, 0.5] according to the on-site communication quality, and a smaller value should be taken when the communication quality is poor to prolong the freshness decay period.

[0041] After obtaining the individual freshness value for each stake, the system aggregates them using a weighted average to obtain the global freshness index F(t), which characterizes the overall staleness of the locally cached data on the new master node: ;in Preset weights are assigned to each charging pile to reflect the differences in their impact on system stability. In this embodiment, to simplify implementation, all online charging piles are treated with equal weights, i.e. If a record is offline, it will not be included in the summation. This weighted average operation ensures that the global freshness index always ranges between (0, 1], with the value closer to 1 indicating that the overall data is fresher.

[0042] To further capture the dynamic trend of freshness, the module stores the global freshness index from multiple consecutive scheduling cycles into a sliding window. The window length L is preset to 5, meaning it retains the F(t) values ​​calculated most recently (5 times). Whenever a new F(t) is calculated, it is added to the end of the window, and the oldest value at the beginning of the window is removed. Based on the sequence within the window, a least-squares method is used for linear fitting to estimate the rate of change of the global freshness index over time. .

[0043] Let the time corresponding to each period within the window be . The corresponding global freshness index is Fitting a straight line using the least squares method. slope This is the original estimate of the rate of change. Since the time intervals between each cycle are equal (all are scheduling cycles T), the calculation can be simplified by directly using the cycle number as the independent variable. Rate of change Finally, the fitted slope is divided by the period duration to obtain a result with clear physical meaning (change per second). If the window is not yet full (e.g., when the system is just starting up), then set... To avoid unstable estimates.

[0044] For example, suppose the F values ​​for the 5 periods within the window are respectively If the corresponding time interval is 0.1 seconds, then the fitting slope is approximately 0.025 per cycle, which translates to a change rate of approximately 0.25 per second. This means that the global freshness index is increasing at a rate of 0.25 per second.

[0045] Finally, the freshness perception module outputs two key quantities in each scheduling cycle: the global freshness index F(t) and its rate of change. These two quantities are directly used as inputs to the dynamic parameter generation module to calculate the subsequent first and second control parameters. Through the above quantization process, the system can transform the fuzzy "data staleness" problem into a computable continuous index, laying a data foundation for smooth scheduling in the critical switching zone.

[0046] The dynamic parameter generation module generates a first control parameter and a second control parameter based on the global freshness index and its rate of change over time. The first control parameter is used to constrain the variation of the scheduling instruction relative to the baseline power instruction set, and the second control parameter is used to adjust the confidence level of the theoretically optimal scheduling result relative to the baseline power instruction set.

[0047] The dynamic parameter generation module receives the global freshness index F(t) and its rate of change output by the freshness perception module. Simultaneously, it indirectly obtains a difference metric between the baseline power command set and the theoretically optimal power command set from the data acquisition module or the command generation module (the specific calculation of this difference metric is completed in a subsequent module, but this module can call this value for parameter correction). The core task of this module is to generate two dimensionless control parameters: the first control parameter... Second control parameter These are used to constrain the magnitude of changes in scheduling instructions relative to the baseline and to adjust the confidence level of theoretically optimal scheduling results, respectively.

[0048] Specifically, the first control parameter The design goal is to achieve a smooth transition from conservative scheduling to open scheduling. In the initial stage of a soft switch, the global freshness index is low, and the reliability of locally cached data is questionable, so instruction changes should be strictly limited. As freshness increases, the system gradually allows for more significant adjustments. This smooth transition requirement dictates that simple threshold switching logic is unsuitable; instead, a sigmoid function with continuously differentiable characteristics should be used. This module uses the following formula to generate... : ;

[0049] Where k is the steepness coefficient, which controls the transition slope of the function near the threshold; The half-activation threshold represents the threshold value when... hour The selection of parameters should be based on practical engineering experience: decrease k for a smoother transition, increase k for a faster switch to the open state. In this embodiment, k is taken as... k=10, such that when F(t) increases from 0.3 to 0.7, The value smoothly increases from approximately 0.12 to approximately 0.88, avoiding sudden changes while ensuring sufficient response speed. For example, if currently... Then the calculation yields This indicates that the allowed range of variation is only 27% of the theoretical value.

[0050] Second control parameter The design is more sophisticated, requiring it to simultaneously reflect both the absolute level of current data freshness and the trend of its change. If freshness is rapidly increasing, the system can appropriately trust the optimization results ahead of time to accelerate convergence; if freshness stagnates or declines, caution should be exercised. Therefore, this module employs the following composite formula: ;

[0051] in The target freshness threshold is defined as the value at which the data is considered sufficiently fresh, allowing complete trust in the optimization results (i.e., the upper limit of w(t) is limited by the first term being 1). This is the gain coefficient, used to adjust the increase in confidence level caused by the rate of change. In this embodiment, we take... , The first point ensures that confidence levels are reduced proportionally when freshness is insufficient, for example, when... At that time, this item was The second term provides gain only when the rate of change is positive, and the gain value increases linearly with the rate of change. Assume the current... Then the second term is Multiplying the two together gives Ultimately This means that although the absolute value of freshness has not reached the target, the system still gives it a 94% confidence level because of the significant upward trend. If the rate of change is negative or zero, the second term degenerates to 1. That is, conservatively relying on the absolute value of freshness.

[0052] The above parameter generation logic is executed once in each scheduling cycle to ensure that the control parameters can respond in real time to the dynamic changes in data freshness.

[0053] To further enhance the system's adaptability under extreme conditions, this embodiment also provides an optional enhancement mechanism. This mechanism, after the basic parameters are generated, introduces a difference metric D(t) between the baseline instruction set and the theoretically optimal instruction set to correct the steepness coefficient of the first control parameter. The difference metric D(t) is calculated by the subsequent instruction generation module or the exit determination module and is defined as the Euclidean distance between the pile power differences corresponding to the two instruction sets. A minimum correction coefficient is set. (For example, 0.5) to prevent the steepness coefficient from decaying excessively due to an excessively small correction factor; set the maximum possible difference. (e.g., transformer rated capacity) This is used to normalize the difference measure. The system pre-defines a mapping relationship and generates correction coefficients. ,in For the maximum possible difference (e.g., total transformer capacity). This is the minimum correction factor (e.g., 0.5). This correction factor decreases as the difference metric increases, indicating that a smoother transition is needed when the theoretical optimum differs significantly from the current benchmark. Multiplying the original steepness factor k by the correction factor yields the corrected steepness factor. And resubstitute into the S-shaped function to generate For example, if D(t) is large, it leads to... ,but ,make The transition curve is smoother, thus further suppressing the risk of sudden changes when command requirements change drastically.

[0054] Through the aforementioned multi-level and adjustable parameter generation mechanism, the dynamic parameter generation module provides a physically meaningful and dynamically adaptive control benchmark for subsequent instruction generation, ensuring that the system can maintain smooth and stable scheduling behavior even in critical areas with high data uncertainty.

[0055] The theoretical optimization module executes a preset power scheduling optimization algorithm based on the status data of each charging pile to obtain the theoretically optimal power instruction set;

[0056] The theoretical optimization module receives the latest status data of each charging pile from the data acquisition module as input. The core task of this module is to run a preset power scheduling optimization algorithm based on all currently available status information, calculate the theoretically optimal power value that should be allocated to each charging pile under ideal conditions (i.e., without considering the impact of data staleness), and form a theoretically optimal power instruction set to provide a target benchmark for subsequent instruction generation.

[0057] In this embodiment, the preset power scheduling optimization algorithm is not a simple formula, but a linear programming model based on constraint satisfaction. The model's construction logic is as follows: the primary constraint of the system is that the capacity of the distribution transformer must not exceed the limit, which is the safety baseline; the secondary objective is to meet the charging needs of vehicles connected to each charging pile as much as possible, while also taking into account the principle of fairness. Based on this, the specific mathematical form of the model is defined as follows.

[0058] First, define the decision variables. Suppose there are M online charging stations, and the decision variables are... This represents the target power value allocated to the i-th charging pile, in kW. The feasible domain of this variable is limited by the physical capacity of each charging pile, i.e.: ;in The maximum available power of the charging station is obtained from its status data. For charging stations where no vehicles are currently connected, It can be considered as 0 or a very small standby power value.

[0059] Secondly, the core constraints are established. The most important constraint is that the total load on the transformer does not exceed its rated capacity. : ;

[0060] in This refers to the rated capacity of the transformer (e.g., 630kW). For other fixed loads besides charging piles (such as lighting, ventilation, etc., assumed to be a constant value of 30kW), in actual deployment, auxiliary loads can be monitored in real time by installing smart meters. The measured values ​​are replaced with real-time values ​​to improve capacity utilization; both are system-preset parameters. This constraint ensures that the total power of all charging stations remains within a safe range.

[0061] Next, we define the optimization objective. This embodiment employs a weighted maximization strategy, aiming to satisfy the charging needs of high-priority vehicles as much as possible. Priority is determined by the battery state of charge of each charging station. and current output power A joint decision was made; for newly connected charging piles whose SOC has not yet been reported, a conservative strategy was adopted: temporarily setting the SOC to 0.5 or setting its priority coefficient to the average value to avoid allocation imbalance due to missing data. Specifically, a priority coefficient for each charging pile was defined. for: ;

[0062] in This represents the battery's state of charge (value range: 0 to 1). This represents the current output power of the pile (obtained from status data). Coefficient. and As a preset weight, in this embodiment, we take... This coefficient reflects a preference for vehicles with low battery levels. A higher coefficient indicates a higher allocation priority for that charging station at that time. For example, a... And the current piles with lower power, It will be higher; while one The piles that are about to be filled The answer is lower.

[0063] Based on the priority coefficients, the optimization objective is set to maximize the weighted power sum: ;

[0064] The significance of this objective function is that, within the allowable capacity, power is preferentially allocated to piles with higher priority coefficients. If capacity is sufficient, all piles can be allocated power according to their maximum demand; if capacity is tight, piles with lower priority will be the first to have their current throttled.

[0065] The linear programming model described above can be solved using the standard simplex method or interior-point method. Since the number of charging stations is typically no more than a few hundred, the solution can be completed in milliseconds, meeting real-time scheduling requirements. The solver output is the theoretically optimal power command set. Each of them This is the theoretical optimal power value of the i-th pile in the current scheduling cycle, obtained from the solution.

[0066] For example: Assume a charging station has 3 online charging piles, and the available transformer capacity is 200kW (after deducting fixed load). The parameters of each pile are shown in the table below:

[0067] Pile Number ) calculate 1 60 0.2 10 0.7*(1-0.2)+0.3*(10 / 60)=0.56+0.05=0.61 2 60 0.5 30 0.7*0.5+0.3*0.5=0.35+0.15=0.50 3 60 0.8 20 0.7*0.2+0.3*(20 / 60)=0.14+0.10=0.24

[0068] Solve the linear programming problem with the objective of maximizing... , constraint is and Since the capacity is sufficient, the optimal solution is obviously... The total is 180kW, which is within the limit. If the transformer capacity is reduced to 150kW, a trade-off must be made. The solver will prioritize satisfying the highest priority stubs, and may output... (Total 150kW), at which point the lowest priority pile 3 is limited to 30kW.

[0069] Through the above process, the theoretical optimization module outputs a set of optimal power allocation schemes based on the current state data in each scheduling cycle. It is important to emphasize that although this set of schemes is "theoretically optimal," the data it is based on may become outdated due to master node switching. Therefore, the output of this module will serve as a reference benchmark for the subsequent instruction generation module, rather than being the final instruction issued directly. This is one of the essential characteristics that distinguishes this invention from traditional methods.

[0070] The individual step size constraint module calculates the maximum allowable adjustment step size for each charging pile within the current scheduling cycle, based at least on the state data lag and power change trend of that charging pile. In this paper, 'power change trend' specifically refers to the power change rate obtained through linear fitting. The unit is kW / s;

[0071] In Example 1, the individual step size constraint module receives the latest status data and historical power queue of each charging pile from the data acquisition module, and simultaneously obtains the lag time of each pile from the freshness perception module. The core task of this module is to calculate, individually, the maximum allowable adjustment step size for each charging station within the current scheduling cycle. This step size will serve as the basis for individualizing the power adjustment range in the subsequent instruction generation module. The design philosophy is that for charging piles with older data, more drastic power changes, and more critical impacts on system stability, the single adjustment range should be subject to stricter limitations.

[0072] The calculation process of this module is divided into three sub-steps: first, determine the power change rate of each pile; then, evaluate the criticality weight of each pile; and finally, calculate the maximum adjustment step size.

[0073] Calculation of power change rate: For each charging pile i, the system maintains a historical power queue, which is initialized by the data acquisition module and updated in each scheduling cycle. The queue stores the power values ​​reported by the pile most recently (N=10 in this embodiment) and their corresponding reporting timestamps, denoted as . , where k=1 corresponds to the latest data and k=N corresponds to the earliest data.

[0074] To estimate the power change trend of the pile at the current moment, the module uses the least squares method to perform a linear fit between the power values ​​in the queue and time. Let the fitted line be... , where the slope This represents the desired rate of power change, expressed in kW / s. Due to the short time interval (100ms per scheduling cycle, A=10 covering the most recent second), the power change can be considered approximately linear. If the time interval between two consecutive reports in the historical queue exceeds the preset maximum interval (e.g., 2 seconds), it is determined that there is data loss within that time period. In this case, the least squares method is suspended. Set the value to 0 or use the average rate of change of the effective values ​​before and after the change until the queue returns to normal. The specific formula for the least squares method is as follows: ;

[0075] in and These are the average time and power in the queue, respectively. If the queue is not full (e.g., the system has just started), then set... This indicates that the trend cannot be assessed at this time. The calculated... It can be positive (power increase), negative (power decrease), or zero. For example, if the power of a certain pile increases uniformly from 30kW to 40kW in the past second, then... kW / s; if the power is stable around 35kW, then .

[0076] Calculation of importance weight: Importance weight This is used to quantify the impact of each charging station on system stability at the current moment. A higher weight indicates that the state change of that station is more likely to cause a global problem, and therefore should be subject to stricter constraints in the step size. This embodiment uses a weighted summation method to integrate three factors: the current output power percentage, the degree of power change, and the battery state of charge. The calculation formula is as follows:

[0077]

[0078] in: This is the current output power of the pile (obtained from the latest status data); The maximum available power of the pile (obtained from static configuration or status data); if If undefined or zero, then the average maximum power of the station where the pile is located is taken; if If the value is 5kW / s, then the default value is used. Ensure the denominator is non-zero before calculation; This represents the absolute value of the rate of change of power. The preset maximum allowable rate of change is used for normalization; in this embodiment, it is taken as... kW / s (rates exceeding this value are considered extremely rapid); The value range represents the state of charge of the battery of the vehicle connected to the charging pile. ;

[0079] The preset weighting coefficients satisfy... In this embodiment, we take The most important factor is to reflect the direct impact of the current power level on the system capacity, followed by the rate of change and the state of charge.

[0080] The physical meaning of this formula is: the current high-power pile (denominator) After normalization, Larger piles occupy more transformer capacity, and their fluctuations have a greater impact on the system; piles with faster change rates ( (Large) implies a transient process with high uncertainty; the third term adopts... To characterize the urgency of charging, the lower the charge level, the more critical the charging station, and the more stringent the protection it should receive in the step size constraint (i.e., smaller allowable adjustment steps to prevent frequent adjustments from affecting charging continuity). Weighting coefficient Adjustments will be made based on business strategy.

[0081] Calculation of maximum adjustment step size: This is done after obtaining the power change rate. And the weight of importance Then, the module enters the core step size calculation stage. Maximum adjustment step size. It is obtained by multiplying the base step size by three discount factors, each of which decays the step size from a different dimension: ;

[0082] in The preset base step size represents the maximum allowable adjustment range per cycle under ideal conditions (completely fresh data, stable power, and non-critical charging piles). In this embodiment, based on the typical power level of the charging pile, the step size is taken as... This means that a maximum of 2kW can be adjusted per scheduling cycle (100ms), corresponding to a change rate of 20kW per second, which is sufficient to cover most scenarios.

[0083] First Discount Factor Based on data lag calculations, the limitations of step size on data staleness are reflected: ;

[0084] in In this embodiment, the maximum tolerable time delay is set to [value]. Seconds. The meaning of this formula is: when the delay... hour, No decay occurs; as the lag time increases, Linear decrease; when the delay reaches or exceeds hour, This means that no adjustments are allowed to be made to the pile (it can only maintain its current power). For example, if a pile has a lag of 2 seconds, then... .

[0085] Second discount factor Calculated based on the rate of power change, reflecting the limitation on the step size when power fluctuates drastically: ;

[0086] in In the calculation of importance weight The rate is consistent, 5 kW / s. The logic of this formula is: the faster the rate of change, the smaller the allowable step size. When hour, ;when near hour, Approaching 0. For example, if the power change rate of a certain pile is 2kW / s, then... If the rate of change reaches 5 kW / s or higher, then This means that adjustments are prohibited to avoid exacerbating the situation during periods of severe volatility.

[0087] Third Discount Factor The importance of the stake is reflected in the additional constraints on the step length, calculated based on the weight of its criticality. ;

[0088] The design of this formula makes the weighting of importance... When changing from 0 to 1, The weight decreases linearly from 1 to 0.5. This ensures that even the most heavily weighted stubs retain at least half of their base step size, rather than being completely locked, thus maintaining the system's responsiveness. For example, if a certain stub... ,but ;like ,but .

[0089] Finally, multiplying the three discount factors by the base step size yields the maximum adjustment step size for that stake in the current period. For example: Suppose a certain stake... , , If the base step size is 2kW, then: ; ; ; ;

[0090] This means that the power adjustment range of the pile must not exceed 0.588kW within this scheduling cycle. If the theoretical optimization result requires an increase of 2kW, the actual increase can only be 0.588kW; if it requires a decrease of 1kW, since 1kW is greater than 0.588kW, the actual decrease is 0.588kW (the truncation is completed by the subsequent instruction generation module).

[0091] Through the above multi-dimensional and individualized step size constraint calculation, the individual step size constraint module generates a dynamic step size upper limit for each charging pile that is adapted to its current state, providing key parameters for the subsequent instruction generation module to achieve fine-grained smooth control.

[0092] The instruction generation module generates the final power instruction set for the current scheduling cycle based on the baseline power instruction set, the theoretical optimal power instruction set, the first control parameter, the second control parameter, and the maximum adjustment step size of each charging pile, and sends it to the corresponding charging pile for execution.

[0093] The instruction generation module, as the final output of the entire scheduling process, has the core task of fusing theoretical optimization results with dynamic control parameters to generate a final power instruction that is both smooth and conforms to individual constraints. This module is triggered in each scheduling cycle, receiving a reference power instruction set from the data acquisition module. Theoretical optimal power instruction set from the theoretical optimization module The first control parameter from the dynamic parameter generation module Second control parameter And the maximum adjustment step size of each stake from the individual step size constraint module. Based on these inputs, the module calculates the final power command for each charging station i according to the following steps. .

[0094] First, the module calculates the expected change for each charging station. This quantity reflects the extent to which the system expects the pile to adjust from its current benchmark value to its theoretical optimal value, after considering data confidence levels. The calculation formula is as follows: ;

[0095] in This is the second control parameter calculated by the dynamic parameter generation module based on the global freshness index and its rate of change. Its value ranges from 0 to 1, and it is used to adjust the confidence level of the theoretically optimal result. When the global freshness index is low or showing a downward trend, A smaller expected change reflects a cautious approach to the optimization results; conversely, when the data is fresh and the trend is positive, A value closer to 1 allows for adjustments closer to the theoretical optimum. For example, if the base power of a certain pile... Theoretically optimal And currently Then the expected change .

[0096] After obtaining the expected change, the module truncates it to ensure that the adjustment range in a single cycle does not exceed the maximum allowable adjustment step size for that pile. The truncation operation preserves the direction of the expected change (increase or decrease) but limits its absolute value to a certain range. Within. The specific formula is: ;in This is the sign function, taking values ​​of +1, -1, or 0. The physical significance of this step is that even if the theoretical optimization results require significant adjustments, the individual step size constraint module has pre-calculated a safety threshold based on factors such as data lag, power change rate, and criticality. The truncation operation forcibly limits the adjustment range to within this threshold, preventing local oscillations or equipment shocks caused by excessive single-cycle changes. Continuing the previous example, if the maximum adjustment step size of the pile... The expected change of 8kW is truncated to 3kW (in the positive direction), that is... .

[0097] The truncated change is then multiplied by the first control parameter. To obtain the final change First control parameter It is an S-shaped function value generated by the dynamic parameter generation module based on the global freshness index, used to constrain the magnitude of changes in scheduling instructions relative to the baseline. When the global freshness index is low, Approaching 0 significantly reduces the adjustment range of all piles; as freshness increases, It gradually approaches 1, allowing for more thorough adjustments. The calculation formula is: To prevent cumulative drift caused by floating-point errors, a dead-zone threshold can be introduced. (e.g., 0.01kW), when At that time, forced This means no adjustments are made. This embodiment omits this step to maintain logical simplicity; however, it can be added as appropriate during actual deployment.

[0098] This multiplication operation, together with the previous truncation operation, constitutes a double constraint: truncation is an independent hard constraint for each stake, ensuring that the single-cycle step size does not exceed the limit. It's a global soft constraint that dynamically adjusts the convergence speed based on the overall data freshness. The combination of these two factors makes the system generally conservative when the data is stale, and gradually loosen its constraints when the data is fresh, while individual outlier stubs are always protected by the step size. Continuing the example above, if the current... Then the final change .

[0099] The module then adds the final change to the reference power to obtain the final power command for the pile: ;

[0100] For example, a base power of 30kW plus 1.5kW equals 31.5kW. Before issuing the command, further adjustments are needed. Boundary constraints are applied to the physical power range to ensure it does not exceed the maximum output capacity of the charging pile and is not lower than the minimum allowable value (usually 0). In this embodiment, the boundary constraints employ a simple truncation process: ;

[0101] in This represents the maximum power of the pile (obtained from status data or static configuration). If the calculated value exceeds the range, it will be forcibly set as a boundary value, and the out-of-limit situation can be logged for operation and maintenance analysis.

[0102] After completing the instruction calculations for all charging stations, the module transmits the final power instruction set via the communication network. The command is then sent to the corresponding charging station for execution. In soft handover mode, to maintain the continuity of progressive adjustments, the module will execute the command sent in this transaction. Stored as a new reference power instruction set, i.e. This is for use in the next scheduling cycle. After exiting soft handover mode, No more updates, the system will directly use As the final instruction.

[0103] Through the above-described hierarchical calculation process, the instruction generation module achieves fine-grained control over power scheduling: first, based on confidence weights... Adjust the target value, then adjust the individual step size. Limiting a single jump, the final result is determined by a global amplitude factor. This determines the overall convergence speed. This design ensures a smooth transition of the system within the critical region of consensus switching, effectively suppressing power oscillations and command surges.

[0104] In this embodiment, the scheduling cycle is set to 100ms. This value is designed to match the power regulation response time of current mainstream DC fast charging piles (typically in the range of 200-500ms). The 100ms step size limit (basic step size of 2kW / cycle corresponds to 20kW / s) ensures that even if there is a delay in the pile-end response, the rate of change of system commands is still much lower than the physical limit of the pile end, ensuring that commands can be smoothly tracked. For piles with faster response speeds, the cycle can be appropriately shortened or the basic step size increased.

[0105] It also includes an exit determination module: in each scheduling cycle, it determines whether the global freshness index is continuously higher than the exit threshold and whether the absolute value of its rate of change is continuously lower than the fluctuation threshold for a preset number of times; if the conditions are met, or if the soft handover mode running time exceeds the preset maximum soft handover time, then the soft handover mode is exited, and subsequent scheduling cycles will directly issue the theoretically optimal power instruction set as the final power instruction set.

[0106] The exit decision module, acting as the lifecycle manager for the soft handover mode, has the core task of assessing in real time whether the system has recovered from the critical transition zone to a stable state, thereby determining when to terminate the soft handover logic and resume normal scheduling. This module is triggered in each scheduling cycle, receiving the global freshness index F(t) and its rate of change from the freshness awareness module. Simultaneously, it retrieves the reference power instruction set from the instruction generation module or internal storage. With the theoretically optimal power instruction set Used for optional enhanced decision-making.

[0107] Basic exit decision logic: The module first maintains three key state variables: a continuous and stable counter. (Initially 0), and a soft-switching timer. (Accumulation begins when the new master node is elected). In each scheduling cycle, the module executes the following decision-making process.

[0108] The first step is to check whether the global freshness index consistently exceeds the exit threshold and whether the absolute value of its rate of change consistently falls below the fluctuation threshold. Set the exit threshold. This indicates that when the global freshness index reaches 0.95 or higher, the locally cached data is considered sufficiently fresh; a fluctuation threshold is set. Per second indicates that when the absolute value of the rate of change in freshness is less than 0.01, the data freshness is considered to have stabilized. If the current period satisfies: and Then the continuous stable counter Increase by 1; otherwise, Reset to 0.

[0109] The second step is to determine whether the exit condition has been met. A preset number of attempts, M=3, is set, meaning the stability condition must be met for three consecutive scheduling cycles. When the module determines that the system has entered a stable state, it triggers the exit from soft handover mode.

[0110] For example, suppose after a certain switch, there are 5 consecutive cycles The values ​​are (0.91, 0.03), (0.93, 0.02), (0.95, 0.008), (0.96, 0.005), and (0.96, 0.004), respectively. The calculation process is as follows: First cycle If the condition is not met (>0.95), the counter is reset to zero; the same applies to the second cycle; the condition is met starting from the third cycle, and the counter is set to 1; the condition is met in the fourth cycle, and the counter is set to 2; the condition is met in the fifth cycle, and the counter is set to 3. When M=3 is reached, the system exits.

[0111] Meanwhile, the module monitors the soft handover runtime in parallel. The maximum soft handover time is set. Seconds. If accumulated from the start of the switch... If the threshold is reached or exceeded, the system will be forced to exit the soft handover mode regardless of whether the freshness condition is met. This mechanism acts as a safety net to prevent the system from remaining in the soft handover state indefinitely due to certain abnormal situations (such as some nodes being offline for a long time, causing the freshness to fail to meet the standard).

[0112] Enhanced Exit Decision Based on Instruction Difference: Building upon the basic logic, this embodiment also provides an optional enhanced decision-making mechanism to further improve the reliability of the exit decision. This mechanism introduces a metric for the difference between the baseline power instruction set and the theoretically optimal power instruction set. It is defined as the Euclidean distance between the power differences of the corresponding piles in two instruction sets: ;

[0113] This metric reflects the overall deviation of the current theoretically optimal solution from the baseline instruction. A convergence threshold is set. In this embodiment, based on the transformer capacity and the number of charging piles, the following is taken: kW indicates that when the total power difference is less than 10kW, the system is considered to have basically converged.

[0114] As a supplement to the basic exit logic, this embodiment also provides an optional enhanced determination mechanism, which adds a difference measurement constraint to the basic conditions. That is, regardless of whether the enhanced mechanism is enabled, the timeout forced exit is always effective.

[0115] In enhanced mode, the exit condition is modified as follows: and This means that not only is fresh and stable data required, but the theoretically optimal solution must also be sufficiently close to the current benchmark. This condition prevents scenarios where, although the data is fresh, a sudden change in load demand could cause a significant gap between the theoretically optimal solution and the benchmark. In such cases, directly exiting the soft handover would result in the next cycle applying drastically changed theoretically optimal instructions, potentially causing a shock. By waiting for the difference metric to converge, a smooth transition after exiting the handover is ensured.

[0116] For example, suppose at a certain moment , It meets the freshness requirement, but If the kW is greater than 10kW, the counter will remain active but will not exit until the difference metric in subsequent cycles drops below the threshold.

[0117] State transition after exit: When the exit conditions are met (basic conditions or enhanced conditions), the module performs the following operations:

[0118] Switch the system operating mode flag from "soft switching mode" to "normal scheduling mode";

[0119] Reset continuous stable counter =0;

[0120] In subsequent scheduling cycles, the instruction generation module no longer uses soft handover logic, but directly uses the theoretically optimal power instruction set. As the final instruction is issued, that is .

[0121] Through the aforementioned hierarchical exit determination mechanism, the exit determination module ensures that the system can only end the smooth transition of the critical section under the premise that the data is fresh enough, the dynamics are stable enough, and (optionally) the instructions are converged enough, thereby minimizing the scheduling impact caused by premature exit.

[0122] As a supplementary safety mechanism, when the exit conditions cannot be met for Z consecutive cycles (e.g., 100 cycles), the system will trigger a forced exit and record an alarm to avoid indefinite stay.

[0123] In summary, this embodiment addresses the power scheduling deviation caused by state data lag during master node switching under a distributed consensus mechanism by constructing a dynamic smooth scheduling mechanism based on data freshness awareness. This mechanism quantifies the data lag of each charging pile and aggregates it into a global freshness index to capture the degree of data staleness and its evolution trend in real time. Then, it uses this index to dynamically generate global control parameters to constrain the scheduling amplitude and weight parameters to adjust the confidence level of the optimization results. Simultaneously, it calculates the upper limit of the differentiated step size by combining the power change rate and criticality of individual piles, ultimately achieving gradual convergence from the baseline command to the theoretically optimal command in the command generation stage. Through the above multi-dimensional adaptive control, the system effectively suppresses power step shocks and capacity allocation oscillations caused by data uncertainty within the switching critical region, ensuring that transformer capacity is safely and fully utilized, and providing a smooth transition scheme with engineering robustness for high-availability switching of distributed charging networks.

[0124] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

[0125] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0126] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A charging pile group power scheduling system based on a distributed consensus mechanism, characterized in that, include: The data acquisition module, in response to the newly elected master node in the distributed consensus cluster, acquires the status data of each charging pile cached locally by the new master node and obtains the reference power instruction set corresponding to the last scheduling cycle before the switch from the consensus log; The freshness perception module calculates a global freshness index, which characterizes the overall staleness of the local cached data of the new master node, and the rate of change of the global freshness index over time, based on the timestamp carried by the status data of each charging pile and the current system time. The dynamic parameter generation module generates a first control parameter and a second control parameter based on the global freshness index and its rate of change over time. The first control parameter is used to constrain the variation of the scheduling instruction relative to the baseline power instruction set, and the second control parameter is used to adjust the confidence level of the theoretically optimal scheduling result relative to the baseline power instruction set. The theoretical optimization module executes a preset power scheduling optimization algorithm based on the status data of each charging pile to obtain the theoretically optimal power instruction set; The individual step size constraint module calculates the maximum allowable adjustment step size for each charging pile within the current scheduling cycle, based at least on the state data lag and power change trend of that charging pile. The instruction generation module generates the final power instruction set for the current scheduling cycle based on the reference power instruction set, the theoretical optimal power instruction set, the first control parameter, the second control parameter, and the maximum adjustment step size of each charging pile, and sends it to the corresponding charging pile for execution.

2. The charging pile group power scheduling system based on a distributed consensus mechanism according to claim 1, characterized in that, The status data of each charging pile includes at least the charging pile identifier, the reporting timestamp, and the current output power value; the reference power instruction set includes the power target value issued to each charging pile in the last scheduling cycle before the switchover.

3. The charging pile group power scheduling system based on a distributed consensus mechanism according to claim 1, characterized in that, The specific calculation process for the global freshness index and its rate of change over time includes: For each charging station, calculate its status data lag time. ,in The current system time. This is the timestamp for the latest status data of this pile. The lag time is mapped to the individual freshness value of the pile using a monotonically decreasing S-shaped function. Where α is a preset scaling factor; The global freshness index is obtained by weighted averaging of the individual freshness values ​​of all online charging piles. The global freshness index for multiple consecutive scheduling cycles is stored in a sliding window, and a linear fit is performed on the sequence within the window to obtain the rate of change of the global freshness index. .

4. The charging pile group power scheduling system based on a distributed consensus mechanism according to claim 1, characterized in that, The dynamic parameter generation step specifically includes: Based on the global freshness index F(t), the first control parameter is generated using a sigmoid function. Where k is the steepness coefficient, The threshold for half-activation points; Based on the global freshness index F(t) and its rate of change Generate the second control parameter ,in The target freshness threshold, This is the gain coefficient.

5. The charging pile group power scheduling system based on a distributed consensus mechanism according to claim 1, characterized in that, The individual step size constraint process also includes: The power change rate of each charging station is obtained by performing linear fitting based on its historical power sequence. The historical power sequence consists of the power values ​​reported by the charging pile the most recently (previously a preset number of times) and their reporting timestamps. Based on the current output power of the charging station Power change rate and battery state of charge Calculate its importance weight ,in This is the maximum power of the charging station. The preset maximum allowable rate of change, , , These are the preset weighting coefficients.

6. The charging pile group power scheduling system based on a distributed consensus mechanism according to claim 5, characterized in that, Maximum adjustment step size for each charging station The calculation process includes: Based on the status data lag of the charging pile Calculate the first discount factor ,in The preset maximum tolerable delay; Based on its power change trend Calculate the second discount factor ; Weighted according to their importance Calculate the third discount factor ; Based on the preset base step size The maximum adjustment step size of the charging pile is obtained by multiplying the first, second, and third discount factors. .

7. The charging pile group power scheduling system based on a distributed consensus mechanism according to claim 1, characterized in that, The specific process of generating the instruction includes: Calculate the expected change of each charging station. ;in The power value of stub i in the theoretically optimal power command set. The power value of stake i in the reference power command set; The desired change is truncated using the maximum adjustment step size of the pile to obtain the truncated change. ; Multiply the truncated change by the first control parameter to obtain the final change. ; Calculate the final power command And constrain it within the physical power range of the charging station.

8. The charging pile group power scheduling system based on a distributed consensus mechanism according to claim 1, characterized in that, It also includes an exit determination module: In each scheduling cycle, it is determined whether the global freshness index is continuously higher than the exit threshold and whether the absolute value of its rate of change is continuously lower than the fluctuation threshold for a preset number of times. If the conditions are met, or if the soft handover mode running time exceeds the preset maximum soft handover time, the soft handover mode is exited and subsequent scheduling cycles directly issue the theoretically optimal power instruction set as the final power instruction set.

9. The charging pile group power scheduling system based on a distributed consensus mechanism according to claim 8, characterized in that, The exit determination process also includes: obtaining a difference measure between the reference power instruction set and the theoretical optimal power instruction set in the current scheduling period; Determine whether the difference metric is less than a preset convergence threshold; The soft switching mode will exit only when the difference metric is less than the convergence threshold, and the global freshness index is consistently higher than the exit threshold and its absolute value of change rate is consistently lower than the fluctuation threshold for a preset number of times.

10. The charging pile group power scheduling system based on a distributed consensus mechanism according to claim 4, characterized in that, The dynamic parameter generation process also includes: Obtain a measure of the difference between the baseline power instruction set and the theoretical optimal power instruction set in the current scheduling period; A correction coefficient is generated based on the difference measure and a preset mapping relationship, and the correction coefficient decreases as the difference measure increases; The preset steepness coefficient is multiplied by the correction coefficient to obtain the corrected steepness coefficient; the first control parameter is generated based on the corrected steepness coefficient.

Images