An electric vehicle charging guidance method and system, electronic device and medium

Abstract

The present application relates to the technical field of power system information processing, and particularly relates to an electric vehicle charging guidance method and system, an electronic device and a medium; the method comprises the following steps: constructing a master-slave game decision model with the maximum aggregation merchant revenue as an objective function and the minimum user charging cost as a response constraint. The master-slave game decision model is reconstructed in a single layer based on the Kalai-Kuhn-Tucker condition to generate a mixed integer linear programming problem. The solution vector of the mixed integer linear programming problem is extracted to obtain an optimal response power benchmark dataset. A distribution network scheduling optimization model considering the optimal response power benchmark dataset is constructed. The distribution network scheduling optimization model is subjected to distributed robust optimization processing to output a pricing guidance strategy. In this way, the technical problem of insufficient power grid operation safety caused by the difficulty of existing electric vehicle charging pricing decision methods in adapting to source-load bilateral uncertainty scenarios is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system information processing, and in particular, relates to an electric vehicle charging guidance method, system, electronic device and medium. Background Art

[0002] As the environment for distributed resources on the distribution and power consumption side to participate in market transactions is increasingly improved, restricted by market access rules, scattered electric vehicle users usually cannot directly participate in the electricity market, and need to participate in transactions through intermediary entities such as power sales companies and load aggregators as agents. The core operation logic of such new market entities is to aggregate scattered user-side flexible resources, guide users to optimize power consumption behaviors by formulating refined charging pricing strategies, and maximize their own aggregation benefits on the premise of ensuring users' energy consumption demands; while charging users on the response end make differentiated responses to pricing signals based on the principle of minimizing charging costs, forming a typical leader-follower game decision-making relationship between the two. Current related researches generally adopt the leader-follower game as the model construction paradigm, and coordinate the interests of multiple parties through the game mechanism to realize the benefit equilibrium of market participants.

[0003] The prior art (Application Publication No. CN114862460A) discloses a pricing method for electric vehicle virtual power plant operators based on a leader-follower game, which constructs a two-layer game model with the upper layer maximizing the revenue of the virtual power plant operator and the lower layer minimizing the charging cost of electric vehicle users, and converts the nonlinear game model into a mixed integer linear programming model for solution through KKT (Karush-Kuhn-Tucker, a core solving tool in the field of nonlinear programming) conditions and the big M method, achieving a win-win situation for the interests of operators and users. However, this prior art still has obvious limitations: most of the existing leader-follower game pricing models focus on the economic benefit game of a single node, only consider the influence of the pricing strategy on user response behaviors and the corresponding benefit changes, do not incorporate pricing decisions and the physical constraints of the actual operation of the distribution network into a unified decision-making framework, and do not consider the dual-side uncertainty impacts brought by the fluctuation of new energy output on the source side and the randomness of load on the load side under the new power system. The pricing strategy formulated purely based on the economic game is very likely to fail to be implemented in actual execution because it breaks through the grid safety boundary, and it is difficult to realize the collaborative closed loop of economic incentive and safe operation. Therefore, the existing electric vehicle charging pricing decision-making method is difficult to adapt to the scenario of dual-side uncertainty of source and load, resulting in the disconnection between the scheduling instruction generated based on the pure economic game and the actual physical operation constraints of the distribution network, and further leading to the technical problem of insufficient power grid operation safety. Summary of the Invention

[0004] In view of the above deficiencies or shortcomings, the present invention provides an electric vehicle charging guidance method, system, electronic device and medium, which can solve the technical problem of insufficient power grid operation safety caused by the existing electric vehicle charging pricing decision-making method being difficult to adapt to the scenario of dual-side uncertainty of source and load.

[0005] This invention provides a method for guiding electric vehicle charging, comprising: constructing a master-slave game decision model based on bilateral market interaction data obtained from aggregators and electric vehicle users, with the objective function of maximizing aggregator revenue and the response constraint of minimizing user charging costs; reconstructing the master-slave game decision model into a single layer based on the Caro-Kuhn-Tucker conditions, linearizing the complementary relaxation conditions, and generating a mixed-integer linear programming problem; iteratively optimizing the mixed-integer linear programming problem, extracting the solution vector, and obtaining the optimal response power benchmark dataset; incorporating the power deviation penalty term into the objective function, and embedding the grid operation state constraints as the feasible region boundary into the initial distribution network scheduling optimization model to construct a distribution network scheduling optimization model that takes into account the optimal response power benchmark dataset; and performing bibliometric optimization on the distribution network scheduling optimization model using a risk metric model based on probability distribution distance, outputting a pricing guidance strategy.

[0006] According to a second aspect, this invention provides a market and operational collaborative electric vehicle charging guidance system that considers source-load uncertainty. This system is deployed on a cloud server cluster. The system includes a processing unit and a storage unit. The storage unit stores executable instructions, which, when invoked by the processing unit, implement any of the electric vehicle charging guidance methods described in this invention.

[0007] According to a third aspect, the present invention provides an electronic device comprising: at least one processing unit; and a storage unit communicatively connected to the at least one processing unit. The storage unit stores executable instructions that can be invoked by the at least one processing unit. When the executable instructions are executed by the at least one processing unit, the at least one processing unit causes it to execute any electric vehicle charging guidance method according to an embodiment of the present invention.

[0008] According to another aspect of the present invention, a medium for storing executable instructions is provided, wherein the computer instructions are used to cause a computer to execute any of the electric vehicle charging guidance methods in the embodiments of the present invention.

[0009] The present invention provides a market and operational coordination method for guiding electric vehicle charging that considers source-load uncertainty. This method is achieved through six core collaborative steps: acquiring bilateral market interaction data, constructing a master-slave game decision model, reconstructing and linearizing the model into a single layer, iteratively optimizing a mixed-integer linear programming problem, constructing a distribution network scheduling optimization model, and performing distributed bar optimization. Specifically, based on bilateral market interaction data obtained from aggregators and electric vehicle users, a master-slave game decision model is constructed with maximizing aggregator revenue as the objective function and minimizing user charging costs as the response constraint, providing a data foundation and game framework for market-based pricing decisions. The master-slave game decision model is reconstructed into a single layer based on the Caro-Kuhn-Tucker conditions, and the complementary relaxation conditions are linearized to generate a mixed-integer linear programming problem, transforming the two-layer game model into a single-layer solvable optimization problem and reducing the solution complexity of the decision model. Iterative optimization of the mixed-integer linear programming problem is performed, and the solution vector is extracted to obtain the optimal solution. The response power benchmark dataset provides a benchmark reference that conforms to market game equilibrium for subsequent scheduling optimization. By incorporating the power deviation penalty term into the objective function and embedding the grid operation state constraints as the feasible domain boundary into the initial distribution network scheduling optimization model, a distribution network scheduling optimization model that takes into account the optimal response power benchmark dataset is constructed, achieving deep coupling between market decision objectives and grid physical operation constraints. Through a risk measurement model based on probability distribution distance, the distribution network scheduling optimization model is subjected to bibliometric optimization, outputting a pricing guidance strategy. This effectively quantifies the impact of source and load uncertainties on the decision results and improves the adaptability of the pricing strategy in complex operating environments.

[0010] In this technical solution, the present invention addresses the technical problem described in the background art: existing master-slave game pricing models often focus on the economic gains of a single node, failing to incorporate pricing decisions and the actual physical constraints of the distribution network into a unified decision-making framework, and neglecting the impact of uncertainties on both the source and load sides. This leads to the technical problem that purely economically oriented pricing strategies are prone to exceeding the grid safety boundary and thus cannot be implemented. By constructing an integrated decision-making mechanism of "market game-physical scheduling-uncertainty optimization," the invention embeds grid operating state constraints as the feasible domain boundary into the optimization model, fundamentally solving the core defect of the disconnect between market decisions and physical operation. Addressing the consequences of this defect, such as pricing strategy execution failure, increased grid safety operation risks, and insufficient renewable energy absorption capacity, the invention establishes a power deviation penalty mechanism and a risk measurement model based on probability distribution distance. On the one hand, it quantitatively penalizes the deviation between market response results and scheduling execution results, forcing the pricing strategy to proactively adapt to grid operation requirements. On the other hand, it comprehensively considers the fluctuation range of source and load prediction errors through bibliometric optimization, achieving a balance between aggregator revenue and user utility while ensuring the safe and stable operation of the grid. Therefore, the technical solution of the present invention solves the technical problem that the existing electric vehicle charging pricing decision-making method is difficult to adapt to the uncertainty of the source and load sides, resulting in insufficient grid operation security. It improves the feasibility of pricing guidance strategy, the safety and stability of grid operation and the efficiency of power market resource allocation, and provides reliable information processing and decision support for the market-oriented operation of distribution and consumption side resources. Attached Figure Description

[0011] Figure 1 This is a flowchart of an electric vehicle charging guidance method according to an embodiment of the present invention; Figure 2 This is an example graph showing the comparison of electric vehicle charging load before and after the implementation of the solution based on the present invention; Figure 3 This is an example curve comparing the dynamic charging pricing strategy of aggregators and the linkage between spot prices in the electricity market after the implementation of the solution based on the present invention; Figure 4 This is an example graph showing the comparison between the day-ahead forecast and measured curves of the distribution network load and photovoltaic power after the implementation of the scheme based on the present invention; Figure 5 This is a block diagram of an electronic device used to implement embodiments of the present invention. Detailed Implementation

[0012] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of the present invention, including various details to aid understanding. These details should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of the invention. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0013] During the development of this invention, researchers conducted numerous experiments and data analysis, discovering a strong negative correlation between aggregator pricing strategies and the voltage stability levels of distribution network nodes: pricing strategies solely driven by maximizing economic returns can lead to imbalances in the spatiotemporal distribution of electric vehicle charging loads, resulting in voltage overruns at local nodes and line overloads. Based on this relationship, this invention innovatively proposes this technical solution, utilizing the inherent connection between market game theory decision-making and the physical operation of the power grid. By embedding power grid operating state constraints as the feasible domain boundary into the optimization model, combined with a risk measurement method based on probability distribution distance, it embodies the core concept of "economic decision-making based on physical security and physical scheduling based on market game theory."

[0014] Specifically, the invention team discovered through experiments that traditional electric vehicle charging pricing methods based on master-slave game theory have technical flaws: they only focus on the economic benefit equilibrium between aggregators and users, without considering the feasibility of pricing decisions at the physical level of the distribution network. This leads to the frequent failure of purely economic-oriented pricing strategies in high-penetration distributed power access scenarios due to exceeding the grid safety boundary. By constructing a distribution network scheduling optimization model that takes into account the optimal response power benchmark dataset, dynamic matching between market game results and actual grid operation needs can be achieved. By performing distributed bar optimization on the distribution network scheduling optimization model and quantifying the fluctuation boundary of the source-load two-sided prediction error based on probability distribution distance, the decision vulnerability of traditional deterministic scheduling methods caused by ignoring uncertainty in complex operating environments is eliminated.

[0015] Therefore, this invention provides an electric vehicle charging guidance method based on the first aspect. This method can be applied to a market and operational coordination electric vehicle charging guidance system (hereinafter referred to as the "system") that considers source-load uncertainty. This system can, by calling pre-set bilateral market interaction datasets and source-load historical datasets in the storage unit, complete the construction of a master-slave game decision-making model, solve mixed-integer linear programming problems, and transform and calculate a distributed bar optimization model, thereby achieving coordinated optimization of the market equilibrium of electric vehicle charging pricing strategies and the safety of power grid operation. Specifically, this system can be deployed in various hardware environments, including but not limited to: cloud server clusters, edge computing nodes, and local data centers with high-performance computing capabilities. Figure 1 As shown, the method may include: Step S110: Based on the bilateral market interaction data obtained from aggregators and electric vehicle users, construct a master-slave game decision model with the aggregator's revenue maximization as the objective function and the user's charging cost minimization as the response constraint.

[0016] Among them, bilateral market interaction data refers to the structured data set generated by aggregators and electric vehicle users in the electricity retail market to support pricing decisions and response behaviors. It includes at least the nodal marginal electricity price in the day-ahead spot market (in yuan per kilowatt-hour), the charging power limit of each controlled node (in kilowatts), and the user charging cost coefficient (dimensionless). The master-slave game decision model is a two-layer optimization framework that describes aggregators as leaders who first set charging service prices, and electric vehicle users as followers who optimize their charging plans accordingly. The upper layer takes maximizing the aggregator's revenue from the electricity purchase and sale price difference as the objective function, and the lower layer takes minimizing the user's charging cost as the response constraint.

[0017] Specifically, the system can periodically retrieve bilateral market interaction data from the power trading platform and user energy management system through the data acquisition interface deployed on the cloud server cluster, and store the cleaned data in a structured data table in the distributed persistent storage area.

[0018] For example, at midnight each day, the system obtains the node marginal electricity price data for the next day's 24 scheduling periods from the provincial power trading center based on the preset application programming interface (API). At the same time, it obtains the charging power upper limit parameters (ranging from 7 kW to 120 kW) of the 50 managed nodes under its jurisdiction and the charging cost coefficients of 1,000 users (ranging from 0.8 to 1.2) from the aggregator's operation platform, and constructs a two-level decision model input matrix with a dimension of 1,128 rows by 3 columns.

[0019] Step S120: Based on the Caro-Kun-Tucker conditions, the master-slave game decision model is reconstructed into a single layer, and the complementary relaxation conditions are linearized to generate a mixed integer linear programming problem.

[0020] Among them, the Caro-Kuhn-Tucker condition refers to the set of equations used in nonlinear programming problems to determine whether the optimal solution satisfies the first-order necessary condition. In this scheme, it is used to transform the optimality condition of the lower-level user response model into the algebraic constraint of the upper-level problem. The complementary relaxation condition refers to the mathematical relationship that the product of the Lagrange multiplier and the constraint inequality is zero, which is manifested by the characteristic that either "the variable is non-negative" or "the constraint is ineffective" must be true. Single-layer reconstruction refers to transforming a two-layer optimization model containing nested logic into an equivalent optimization problem containing only a single-layer constraint set through mathematical transformation. The Big M method refers to introducing a sufficiently large positive number (usually taken as a value of 1). This involves a modeling technique that transforms the nonlinear logic "if A>0 then B=0" in complementary relaxation conditions into linear inequality constraints, along with binary decision variables.

[0021] Specifically, the system can first list the Lagrangian function of the lower-level user response model through the built-in optimization solution engine, derive the corresponding Caro-Kun-Tucker conditions and complementary relaxation conditions, then introduce binary decision variables and sufficiently large positive numbers, use the Big M method to transform the complementary relaxation conditions into linear inequality constraints, and incorporate all the transformed constraints into the constraint set of the upper-level optimization problem to complete the single-layer reconstruction.

[0022] For example, the system introduces a binary decision variable with a value of 0 or 1 and a variable with a value of 1 to address the constraint of "charging power not exceeding the upper limit" in the lower-level model. A sufficiently large positive number transforms the nonlinear complementary relationship "0 ≤ charging power ≤ upper limit and 0 ≤ Lagrange multiplier" into two linear inequalities: charging power ≤ upper limit × binary decision variable, and Lagrange multiplier ≤ ×(1-binary decision variables) thus generates a mixed-integer linear programming problem containing approximately 5000 continuous variables, 2000 binary decision variables, and 8000 constraints.

[0023] Step S130: Iteratively optimize the mixed-integer linear programming problem, extract the solution vector of the mixed-integer linear programming problem, and obtain the optimal response power benchmark dataset.

[0024] Iterative optimization refers to the process by which a mathematical programming solver uses algorithms such as Branch and Bound (B&B) to iteratively search for the optimal value of the objective function within the feasible region. The solution vector is a one-dimensional array containing the values ​​of all decision variables (including continuous and discrete variables) after the optimization problem is solved. The optimal response power benchmark dataset is a data set organized in a spatiotemporal dimension that represents the optimal charging behavior of electric vehicle users under pure market game equilibrium. It is usually stored as a key-value mapping (KVM) structure, where the key contains node identifiers and time identifiers, and the value contains power values.

[0025] Specifically, the system can call a high-performance mathematical programming solver (such as Gurobi or CPLEX) to perform iterative optimization calculations based on the branch and bound method on the generated mixed integer linear programming problem. After the solution status is displayed as optimal, the system extracts the charging power variable values ​​corresponding to each control node and each scheduling period from the high-dimensional solution vector and encapsulates them into a multi-dimensional index structure.

[0026] For example, the system calls the Gurobi 10.0 solver and sets the relative gap tolerance to... After approximately 15 seconds of iterative calculation, the system outputs a solution vector containing over 7,000 variable values. From this vector, the system extracts a total of 1,200 charging power values ​​(in kilowatts) for 24 time periods and 50 nodes. It then constructs a multidimensional index structure with "node ID - time period ID" as the key and "power value" as the value, and persistently stores this structure in the global data plane.

[0027] Step S140: Incorporate the power deviation penalty term into the objective function and embed the grid operation state constraint as the feasible domain boundary into the initial distribution network scheduling optimization model to construct a distribution network scheduling optimization model that takes into account the optimal response power benchmark dataset.

[0028] Among them, the power deviation penalty term refers to the cost function that quantifies the degree of deviation between the real-time dispatch power and the aforementioned optimal response power benchmark dataset, which is usually expressed as the product of the absolute value of the deviation and the preset penalty coefficient (in yuan); the power grid operation state constraint refers to the set of boundary conditions that guarantee the physical safety of the distribution network, described in the form of mathematical equations or inequalities, including at least node voltage amplitude constraints (in kilovolts), branch current thermal stability constraints (in kiloamperes), and node power balance constraints; the feasible domain boundary refers to the multidimensional spatial geometric boundary that is enclosed by all constraints and contains all feasible solutions in the optimization problem.

[0029] Specifically, the system can access the global data plane to retrieve the optimal response power benchmark dataset, calculate the absolute deviation between the real-time scheduling power variable and the recorded value in the dataset, multiply it by a preset deviation penalty coefficient to construct a power deviation penalty term, and then add the penalty term to the original objective function of the initial distribution network scheduling optimization model. At the same time, the node voltage amplitude constraints, branch current thermal stability constraints, and node power balance constraints obtained by analysis are encapsulated into a constraint matrix and used as the feasible domain boundary to embed the model.

[0030] For example, the system sets the power deviation penalty coefficient to 2.5 yuan per kilowatt, constructs the penalty term "2.5 × |real-time power - reference power|", and adds it to the initial model with the goal of minimizing the total operating cost. At the same time, the system analyzes the topology parameters of the IEEE 33-node standard test system, sets the node voltage amplitude constraint to 0.95 kV to 1.05 kV, sets the branch current thermal stability constraint to no more than 0.4 kA, and converts these constraints into a constraint matrix containing 132 variables. This matrix is ​​then embedded into the optimization model to construct a distribution network dispatch optimization model that takes into account the reference dataset.

[0031] Step S150: Using a risk measurement model built based on probability distribution distance, perform sub-Bruker optimization on the distribution network dispatch optimization model and output a pricing guidance strategy.

[0032] Among them, probability distribution distance refers to a mathematical metric used to measure the degree of difference between two probability distributions. In this scheme, the Wasserstein Distance (WD) is preferred, which quantifies uncertainty by calculating the minimum "work" required to move one distribution to another. Risk measurement model refers to a mathematical model used to evaluate and optimize the potential losses of decisions in uncertain environments. In this scheme, Conditional Value at Risk (CVaR) is introduced to measure costs in extreme scenarios. Split-Brow optimization is an optimization technique that comprehensively considers the ambiguity of probability distributions. It finds decision schemes that are immune to distribution estimation errors by performing worst-case analysis within a fuzzy set centered on the empirical distribution and with the probability distribution distance as the radius.

[0033] Specifically, the system can fit an empirical distribution based on historical source load data samples, define the fluctuation boundary of prediction error according to the preset probability distribution distance to construct a fuzzy set, establish an affine mapping relationship between scheduling decision variables and uncertainties, and reconstruct the semi-infinite programming problem containing fuzzy sets into a deterministic mathematical programming problem through duality theory. After solving the problem, the pricing guidance strategy corresponding to each scheduling period is output analytically.

[0034] For example, the system collects output data and concurrent load data of photovoltaic power plants in a certain region over the past 365 days with a time resolution of 1 hour, calculates the prediction error, and constructs an empirical distribution containing 8760 samples; sets the distance radius of the probability distribution to 0.05 and constructs a fuzzy set; introduces linear decision rules to establish a linear relationship between scheduling decision variables for each time period and the prediction errors of photovoltaic and load; finally, it solves for 24 pricing guidance strategy values ​​for the next day, with a value range of 0.45 yuan per kilowatt-hour to 1.25 yuan per kilowatt-hour. This strategy achieves a balance between aggregator revenue and user costs while ensuring that the grid voltage qualification rate is maintained above 99.9%.

[0035] In some embodiments, the system can calculate the charging pricing revenue of the aggregator at time t on the nth resource control node using the following formula (1): (1); Formula (1) is the objective function for optimizing pricing at the aggregator's upper level, which is used to maximize the net profit per unit time when the aggregator acts as an agent for electric vehicles to participate in the day-ahead spot market transaction. This represents the profit maximization objective of the aggregator at the nth resource control node; T is the total set of scheduling periods (e.g., ...). (where I represents 24 hours in a day); I is the set of the total number of controllable electric vehicles under the nth resource management node; The charging price set for the nth resource management node at time t, in yuan per kilowatt-hour; The day-ahead spot price of electricity in the electricity market at time t, expressed in yuan per kilowatt-hour; Let represent the charging power of the i-th electric vehicle at time t in the n-th resource management node, expressed in kilowatts.

[0036] Next, in this embodiment, the system can also calculate the upper and lower limits and average price constraints that the aggregator must meet when setting charging prices using formula (2): (2); Formula (2) represents the set of constraints for aggregated commercial electricity pricing. To constrain the lower limit of pricing, ensure that the price set by aggregators will not be lower than the spot market price minus the adjustment coefficient c; c is the pricing adjustment coefficient, which is determined by the type and scale of internal electricity resources contained in each node, and the unit is yuan per kilowatt-hour. For example, set c = 0.05 yuan per kilowatt-hour. To impose a pricing cap and ensure that aggregators do not set excessively high prices, among which... The electricity purchase price for the enterprise at time t is expressed in yuan per kilowatt-hour. The pricing average constraint is used to prevent aggregators from engaging in excessive profit-seeking behavior and to provide users with economic protection. The total number of scheduling periods (e.g., 24). The average price for charging and discharging is usually set by the government, in yuan per kilowatt-hour. For example, it is set as follows: =0.6 yuan per kilowatt-hour.

[0037] Furthermore, in this embodiment, the system can also calculate the electricity cost of the i-th electric vehicle at time t using formula (3): (3); Formula (3) is the objective function of the lower-level user response model, which is used to minimize the overall charging expenditure cost of electric vehicle users. This represents the goal of minimizing electricity costs for all electric vehicle users under the nth resource control node; The charging pricing set for aggregators, in yuan per kilowatt-hour; Let represent the charging power of the i-th electric vehicle at time t in the n-th resource management node, in kilowatts; T is the total scheduling time period set; and I is the set of electric vehicle numbers.

[0038] Furthermore, in this embodiment, the system can also calculate the various physical operating constraints of the electric vehicle during the charging process using formula (4): (4); Formula (4) is a set of constraints on the charging behavior of electric vehicles. To impose boundary constraints on charging power, the charging power of a single vehicle is limited to between 0 and the maximum allowable power. For example, setting... =7 kilowatts; To impose boundary constraints on battery capacity, limit the battery state of charge between a minimum and a maximum threshold, for example, by setting... =10 kWh, =60 kWh; The constraints for dynamically updating battery power describe the cumulative relationship of power levels in adjacent time periods, where... Electric vehicle charging efficiency (dimensionless, e.g., 0.9). The unit of time (e.g., 1 hour); To ensure a minimum off-grid power limit, this ensures that the vehicle's battery level does not fall below the user's expected level when disconnected from the grid. Let be the off-grid time of the i-th electric vehicle. Let be the desired energy level of the i-th electric vehicle, in kilowatt-hours, for example, set as follows: =50 kWh.

[0039] Therefore, by combining the above formulas (1) to (4), the system can construct a two-level optimization mathematical model that takes into account the master-slave game relationship. The model clarifies the non-zero-sum game objective between the aggregator's pursuit of profit maximization and the user's pursuit of cost minimization through formulas (1) and (3), and introduces rigid constraints that conform to the requirements of electricity market regulation and physical operation laws through formulas (2) and (4), thus providing a rigorous mathematical theoretical basis for solving the optimal charging and discharging pricing strategy and guiding the user-side response.

[0040] In some embodiments, the system can realize the transformation and solution of the master-slave game two-layer model through the following formula (5): (5); Formula (5) is the objective function of the transformed upper-level model, used to maximize the overall net profit of the aggregator after taking into account the optimal response of the lower-level users. `max` represents the maximum value operation; `T` is the total scheduling time period set; `I` is the set of electric vehicle numbers. to These are the dual variables (real variables) of the inequality constraints introduced via the KKT conditions. , , and These are the slack variables for the upper limit constraint of charging power, lower limit constraint of battery capacity, upper limit constraint of battery capacity, and off-grid battery capacity guarantee constraint of the corresponding lower-level user model; Let t be the day-ahead spot price of electricity.

[0041] Next, in this embodiment, the system can also calculate the complementary relaxation conditions of the lower-level user model using formula (6): (6); Formula (6) is a set of KKT complementary relaxation conditions, which is used to characterize the relationship that the product between the Lagrange multipliers of the inequality constraints in the original lower-level model and their corresponding relaxation variables is zero. For the dual variable of the non-negativity constraint on charging power; Maximum charging power for EVs (e.g., 7 kW); and These are the lower and upper limits of electricity consumption (e.g., 10 kWh and 60 kWh). The desired power consumption is 50 kWh; the symbol ⊥ indicates a complementary relationship, that is, A⊥B means A×B=0.

[0042] Furthermore, in this embodiment, the system can also construct the distribution network dispatch optimization model using formula (7) to take into account the optimal response power benchmark dataset: (7); Formula (7) is the objective function of the distribution network dispatch optimization model that takes into account the optimal response power benchmark dataset, aiming to minimize the overall operating cost of the system. min represents the minimum value operation; G is the set of micro fuel oil generators; P is the photovoltaic set; N is the set of control resource nodes; The coal cost coefficient for the g-th micro-oil generator unit (e.g., 0.3 yuan per kilowatt-hour); The active power output (in kilowatts) of the g-th micro-oil generator unit during time period t. and These are the operation and maintenance cost coefficients for micro gas turbines and photovoltaics, respectively; and These are the cost coefficients for turning on and off (e.g., 50 yuan and 20 yuan respectively). and These are binary variables representing the start-up and shutdown actions, respectively. The amount of electricity purchased in the spot market before time t; The incentive price for the charging power deviation at time t (e.g., 0.1 yuan per kilowatt-hour); The EV charging power scheduling deviation at time t; Let t be the absolute value of the active power compensation equipment output of the first node; Let t be the active power compensation cost.

[0043] Furthermore, in this embodiment, the system can also calculate the absolute value constraints in the operation of the distribution network using formula (8): (8); Formula (8) represents the absolute value and deviation constraint group. The first and second lines define the absolute value form of the EV charging power scheduling deviation, ensuring that the deviation is accurately captured regardless of whether the actual charging power is higher or lower than the reference value. The third line defines the baseline value of the optimal response charging power under market equilibrium; it defines the absolute value of the output of the active power compensation equipment at the first node. The actual output of the active power compensation equipment at the first node (in kilowatts).

[0044] Therefore, by combining the above formulas (5) to (8), the system can transform the complex nonlinear master-slave game model into a mixed-integer linear programming model that can be solved efficiently, and construct a multi-energy complementary dispatching framework for the distribution network that takes into account market deviation penalties. This combination effectively eliminates the inherent nonlinear terms in the traditional model. By introducing deviation penalty terms and compensation mechanisms, it achieves a high degree of synergy between the market game results and the operation of the physical power grid, thereby improving the economy and security of the distribution network operation under a high proportion of new energy access.

[0045] In some embodiments, the system can calculate the capacity constraints of micro-oil generators and photovoltaic output in the operation of the distribution network using the following formula (9): (9); Among them, formula (9) is the generator operation constraint group including unit start-up and shutdown. Let g be the binary variable representing the state of the g-th micro fuel oil generator unit at time t (with a value of 0 or 1, where 1 indicates the unit is in operation). and These are binary variables representing the start-up and shutdown actions, respectively. and These are the active power output and the upper limit (for example, setting the upper limit of a certain unit to 500 kilowatts). The ramp rate (e.g., 50 kW per time slot) is used to limit the range of output variation between adjacent time slots; , and Here are the coefficients used to describe the polygon approximation, and here are the linear approximation parameters for the capacity constraint; and These are reactive power and apparent power, respectively.

[0046] Furthermore, in this embodiment, the system can also calculate the second-order cone optimal power flow constraint of the distribution network using formula (10): (10); Formula (10) is the second-order cone optimal power flow constraint set of the distribution network, which is used to ensure the consistency of the physical laws of the power grid operation. and The active and reactive power are injected into the node, mainly including the micro-oil generator and photovoltaic output at the node; and These are active and reactive loads, respectively. and These are the node voltage and the square of the line current, respectively. and These are the line resistance and reactance, respectively. and To take into account the active load after EV rescheduling; and These represent the overall bus and line set, respectively; π(j) represents the line sub-node; It is a 2-norm.

[0047] Furthermore, in this embodiment, the system can also calculate the upper and lower limits of node voltage and line current using formula (11): (11); Formula (11) represents the safety boundary constraints for node voltage and line current. and These are the upper and lower limits of the node voltage (e.g., setting them). =0.95 per unit value, =1.05 per unit value); I is the upper limit of line current (for example, setting I=1.2 per unit value) to prevent line overload.

[0048] Furthermore, in this embodiment, the system can also construct a bibliophilic optimal scheduling model based on Wasserstein distance using formula (12): (12); Formula (12) is the objective function for the bibliometric optimization based on Wasserstein distance, used to find the optimal day-ahead scheduling decision under the worst-case distribution, considering the uncertainty of source load. min represents finding the minimum value; It is the supremum function, used to find the worst probability distribution that maximizes the expected cost in the fuzzy set B(r); is the two-stage intraday scheduling cost function; r is a fuzzy set range parameter that limits the distribution of prediction errors.

[0049] Next, in this embodiment, the system can also define a risk measurement function under a certain confidence interval using formula (13), taking into account the CVaR split-bar optimization scheduling model: (13); Formula (13) is the risk measure and bi-level optimization objective function that takes into account CVaR. α represents the risk measure function (CVaR) at the α confidence interval level; ς is an auxiliary variable for scheduling cost; h is the intraday stage scheduling cost coefficient. ω represents the vector of decision variables for the two-stage intraday phase; ω represents the weighting coefficients of scheduling cost and CVaR. It is a positive part function.

[0050] Furthermore, in this embodiment, the system can also establish a linear affine dependency between the decision variable and the uncertainty using formula (14): (14); Formula (14) is the set of constraints after introducing LDR (linear decision rule). These are introduced auxiliary variables used to approximate the relationship between decision variables and uncertainties. This is the integral form of the distance metric between the true distribution and the empirical distribution, expressed as the Wasserstein distance metric.

[0051] Therefore, by combining the above formulas (9) to (14), the system can deeply integrate the physical operation constraints of the distribution network (such as unit start-up and shutdown, power flow distribution, voltage and current boundaries) with the sub-Bruker robust optimization model that considers extreme risk scenarios. By introducing Wasserstein distance and CVaR risk measure, the model can effectively resist the prediction deviation risk of new energy and load; at the same time, with the help of linear affine dependency (Linear Decision Rule, abbreviated as LDR) and duality theory, the system successfully transforms the complex nonlinear robust optimization problem into a mixed integer linear programming (MILP) model that can be solved efficiently, thereby achieving the optimal balance between day-ahead market electricity purchase cost and operation risk under the premise of ensuring the safe and stable operation of the power grid.

[0052] In some embodiments, the system can obtain the final mixed integer linear programming (MILP) model after introducing auxiliary variables using the following formula, so as to solve the distribution network distributed bar optimization scheduling problem with uncertainty: ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; The above formula is the final mixed integer linear programming (MILP) model after introducing auxiliary variables. By transforming the Wasserstein distance constraint and CVaR risk measure constraint into a finite number of linear constraints, it achieves an efficient solution to the distribution network distributed bar optimization scheduling problem with uncertainty. Let be the objective function, representing minimizing the sum of the total operating cost and risk cost of the system; This is the unit cost coefficient; For decision variable vectors; and These are the upper and lower bound vectors for the decision variables, respectively; to The vector of intercept variables; to The dual variable vector; to Here is the weight coefficient vector; r is the Wasserstein distance radius; For the uncertainty of photovoltaic power output prediction error; to and to The auxiliary variables introduced are used to transform the norm constraint into a linear form; A and It is a coefficient matrix; For day-ahead scheduling decision variables; and This represents the initial values ​​and coefficient vector of the intraday scheduling decision variables; and is a constant vector; h is the intraday stage scheduling cost coefficient; ω is the weighting coefficient of scheduling cost and CVaR; α is the confidence level; ς is an auxiliary variable; and κ represents the variable relevant to risk measurement; κ is the dual variable. For example, setting weight coefficients. The Wasserstein distance radius is r=100, the confidence level is α=0.95, and the weighting coefficient is ω=0.5, in order to achieve the optimal balance between cost and risk.

[0053] Therefore, by establishing the above formula, the system can transform the complex sub-Bruker optimization problem into a standard mixed-integer linear programming model, and quickly obtain the global optimal solution through commercial solvers (such as Gurobi or CPLEX), thus ensuring the safe and economical operation of the distribution network even in harsh scenarios such as extreme weather or sudden load changes.

[0054] This embodiment provides a market and operational coordination method for guiding electric vehicle charging that considers source-load uncertainty. This method is achieved collaboratively through six core steps: acquiring bilateral market interaction data, constructing a master-slave game decision model, single-layer model reconstruction and linearization, iterative optimization of mixed-integer linear programming problems, constructing a distribution network scheduling optimization model, and performing distributed bar optimization. Specifically, based on bilateral market interaction data obtained from aggregators and electric vehicle users, a master-slave game decision model is constructed with maximizing aggregator revenue as the objective function and minimizing user charging costs as the response constraint, providing a data foundation and game framework for market-based pricing decisions. The master-slave game decision model is reconstructed into a single layer based on the Caro-Kuhn-Tucker conditions, and the complementary relaxation conditions are linearized to generate a mixed-integer linear programming problem, transforming the two-layer game model into a single-layer solvable optimization problem and reducing the solution complexity of the decision model. Iterative optimization of the mixed-integer linear programming problem is performed, extracting the solution vector to obtain the optimal solution. The response power benchmark dataset provides a benchmark reference that conforms to market game equilibrium for subsequent scheduling optimization. By incorporating the power deviation penalty term into the objective function and embedding the grid operation state constraints as the feasible domain boundary into the initial distribution network scheduling optimization model, a distribution network scheduling optimization model that takes into account the optimal response power benchmark dataset is constructed, achieving deep coupling between market decision objectives and grid physical operation constraints. Through a risk measurement model based on probability distribution distance, the distribution network scheduling optimization model is subjected to bibliometric optimization, outputting a pricing guidance strategy. This effectively quantifies the impact of source and load uncertainties on the decision results and improves the adaptability of the pricing strategy in complex operating environments.

[0055] In this technical solution, this embodiment addresses the technical problem described in the background section: existing master-slave game pricing models often focus on the economic gains of a single node, failing to incorporate pricing decisions with the actual physical constraints of the distribution network into a unified decision-making framework, and neglecting the impact of uncertainties on both the source and load sides. This leads to the technical problem that purely economically oriented pricing strategies are prone to exceeding the grid safety boundary and thus cannot be implemented. By constructing an integrated decision-making mechanism of "market game-physical scheduling-uncertainty optimization," the constraints of the grid operating state are embedded as the feasible domain boundary into the optimization model, fundamentally solving the core defect of the disconnect between market decisions and physical operation. To address the consequences of this defect, such as pricing strategy execution failure, increased grid safety operation risks, and insufficient renewable energy absorption capacity, a power deviation penalty mechanism and a risk measurement model based on probability distribution distance are established. On the one hand, the deviation between market response results and scheduling execution results is quantitatively penalized, forcing the pricing strategy to proactively adapt to grid operation requirements. On the other hand, through bibliometric optimization, the fluctuation range of source and load prediction errors is comprehensively considered, achieving a balance between aggregator revenue and user utility while ensuring the safe and stable operation of the grid. Therefore, the technical solution of this embodiment solves the technical problem that the existing electric vehicle charging pricing decision method is difficult to adapt to the uncertainty of both the source and load sides, resulting in insufficient grid operation security. It improves the feasibility of the pricing guidance strategy, the safety and stability of grid operation, and the efficiency of power market resource allocation, and provides reliable information processing and decision support for the market-oriented operation of distribution and consumption side resources.

[0056] In another embodiment, Tables 1-3 below show the multi-dimensional performance comparison data of this solution under different scenarios and models, which intuitively verify the comprehensive advantages of this solution in power grid security, computing efficiency and commercial economy.

[0057] Table 1 focuses on the operational performance of the one-stage comparative model and the deterministic optimization scheduling model. Regarding the average voltage drop index, the model corresponding to this scheme has a value of 0.04, slightly higher than the deterministic model's 0.02. This is mainly because this scheme introduces more realistic distributed bar constraints to cope with extreme uncertainties and adopts a more conservative strategy at the extreme safety boundary. However, in terms of load distribution balance, this scheme optimizes the value from 0.20 to 0.24, significantly improving the power flow distribution of the grid, effectively avoiding the overload risk caused by excessive load on local lines, and greatly enhancing the physical safety of the distribution network operation.

[0058] Table 2 further compares the computational efficiency and optimization results of different risk preference models. It can be seen that as the model complexity increases towards robust optimization, the solution time exhibits a stepwise upward trend. For example, the solution time of the CVaR robust model reaches 584.61 seconds, while the CVaR partially robust model increases to 6324.92 seconds. Nevertheless, this scheme, while ensuring solution accuracy, successfully transforms the originally extremely complex nonlinear problem into an efficiently solvable mixed-integer linear programming model through ingenious linearization transformation and the application of duality theory, providing a feasible technical path for practical engineering applications. In terms of the optimal value, the partially robust model achieves 52536.25 yuan, surpassing the traditional CVaR model's 45172.68 yuan, demonstrating the commercial value of fully considering the worst-case distribution in reducing long-term systemic risk.

[0059] Table 3 provides an in-depth evaluation of the risk resistance and solution stability of each model under different levels of prediction error fluctuation (low, medium, and high). In the low-fluctuation scenario, all models successfully obtained the optimal solution, but this solution still maintained the lowest average cost (39,148.93 yuan). When faced with adverse market environments of medium (σ=0.3) and high (σ=0.5) fluctuations, the solution rates of the traditional deterministic model dropped sharply to 14.53% and 32.85%, respectively, completely losing their practical engineering value; even the traditional CVaR model had a 12.63% no-solution rate under high fluctuations. In contrast, the CVaR split bar model in this solution maintained an extremely low no-solution rate (0% and 3.25%) under medium and high fluctuations, and the average cost was reduced by approximately 3.3% and 10.1% respectively compared to the deterministic model.

[0060] The data presented in the tables above clearly demonstrate that this solution avoids the pitfall of solely pursuing theoretical perfection while neglecting practical engineering applications. Through sophisticated mathematical modeling and algorithm design, it achieves a perfect balance between cost control and risk mitigation at the business management level, while providing grid dispatchers with a robust and reliable decision-making tool even under extreme data disturbances at the administrative supervision level. This fully showcases the powerful enabling role of information and communication technologies in the commercial operation and macro-level supervision of complex power systems.

[0061] In another embodiment, as follows: Figure 2 The graph shows a comparison of electric vehicle charging load before and after the implementation of this solution. The horizontal axis represents a 24-hour time series, and the vertical axis represents the total amperage (A) of the charging stations, intuitively reflecting the dynamic changes in the grid load.

[0062] The red line in the graph represents the "disordered charging curve" without this solution. It can be seen that in the disordered state, influenced by users' concentrated return home from get off work (e.g., after 6 PM) and daily routines, the charging load experiences extremely concentrated peaks during the evening hours (especially from 9 PM to 11 PM), with the highest amperage approaching 300 amperes. This "evening peak" phenomenon not only leads to local distribution network overload and transformer capacity shortages but also significantly increases the grid's peak-shaving pressure and operation and maintenance costs, and even poses safety hazards. The blue line represents the "ordered charging curve" after applying this solution. Based on the aforementioned master-slave game-theoretic two-layer optimization model and distributed bar scheduling framework, the system successfully guides users' electricity consumption behavior by bidirectionally distributing reasonable dynamic charging and discharging pricing strategies to aggregators and EV users. Specifically, during the evening peak hours when grid load is high, the system appropriately increases charging prices or provides discharging subsidies through price levers, prompting users to actively reduce unnecessary charging power or participate in vehicle-to-grid (V2G) discharging; while during the early morning and midday hours when grid load is low, the system encourages users to engage in off-peak charging by lowering electricity prices. The resulting blue curve was significantly "flattened" over time, with sharp peaks effectively dispersed, load fluctuations smoothed out, and overall operation more stable. This comparison of actual operational data fully demonstrates that this solution is not merely a theoretical model deduction, but can play a crucial role in real-world business and administrative management scenarios. Through advanced information and communication technologies, the system successfully breaks down the barriers between individual user interests and overall grid security, achieving refined guidance and collaborative management of large-scale electric vehicle charging behavior without significantly sacrificing user travel experience. This not only greatly improves the operational resilience and asset utilization of the distribution network but also creates considerable economic benefits for aggregators, perfectly aligning with the core technological goal of achieving a win-win situation for all parties at the administrative, commercial, and management levels.

[0063] In another embodiment, as follows: Figure 3 This chart shows a comparison curve between the aggregator's dynamic charging pricing strategy and the electricity market spot price after the implementation of this scheme. The horizontal axis represents the division of a 24-hour day, and the vertical axis represents the electricity price (in yuan per kilowatt-hour). The solid red line represents the agent's electricity purchase price, the solid blue line represents the day-ahead spot market price, and the dashed orange line represents the dynamic charging pricing generated by this scheme.

[0064] Looking at the overall trend throughout the day, this solution demonstrated a high degree of adaptability and refined management capabilities in a complex market environment. During the off-peak load period from early morning to late afternoon (e.g., 1:00 AM to 8:00 AM), the spot market price was low, and the system accordingly maintained charging pricing at an extremely low level of around 0.3 yuan / kWh. This incentivized users with flexible travel plans to recharge in advance, effectively absorbing surplus clean energy from the grid and improving the utilization rate of green energy. As daytime industrial and commercial loads increased and photovoltaic power generation decreased marginally, the spot market price began to fluctuate dramatically. For example, the first small price peak occurred around 10:00 AM, followed by a brief decline at noon, and then the highest peak of the day occurred between 5:00 PM and 7:00 PM. Faced with this highly volatile market environment, the charging pricing of this solution did not blindly follow the trend, but rather exhibited clear characteristics of "peak shaving and valley filling" and "price smoothing." During peak electricity consumption periods in the morning and evening, charging prices are strictly limited to a reasonable range of 0.8 to 1.25 yuan, preventing large-scale user "charging abandonment" due to soaring electricity prices. During transitional periods with relatively stable price fluctuations, pricing can be flexibly adjusted to follow the market benchmark price. This dynamic pricing mechanism has significant advantages in terms of commercial management and administrative supervision. For aggregators, by precisely targeting off-peak market price periods for low-cost electricity purchases and offering services to users at a reasonable premium during peak periods, their profit margins and market competitiveness are greatly expanded. For grid managers, this strategy is equivalent to installing an invisible "virtual energy storage" regulating valve on the user side, guiding a massive number of electric vehicles to form a combined force for off-peak charging through price signals. Without increasing additional hardware investment, it significantly alleviates the power supply pressure on the grid during peak hours, ensuring the safe and stable operation of the power system. This fully demonstrates the superior efficiency of this solution in handling complex commercial transactions and macro-level grid dispatch and collaborative management.

[0065] In another embodiment, as follows: Figure 4 The figure displays a comparison curve of the day-ahead forecast and measured values ​​of distribution network load and photovoltaic power after the implementation of this scheme. The figure is divided into two sub-figures: the left side shows the normalized load power (including forecast and measured values), and the right side shows the normalized photovoltaic power (including forecast and measured values), which intuitively reflects the system's forecast accuracy and dynamic adjustment capability in dealing with uncertainties on both the source and load sides.

[0066] Looking at the load power comparison on the left, the measured load value represented by the red solid line exhibits a typical "double-peak" characteristic throughout the day, peaking during the morning peak (approximately 8:00 to 10:00) and the evening peak (approximately 18:00 to 21:00), which closely matches the daily routines of residential and industrial users. The load forecast value represented by the green dashed line closely follows the fluctuation trend of the measured value. Although there are slight deviations in local periods, the overall trend is highly consistent. This indicates that the forecasting module on the system can accurately capture the dynamic changes in load, providing a reliable data foundation for subsequent scheduling decisions. More importantly, based on this high-precision forecast data, this solution can formulate off-peak charging and discharging strategies in advance, guiding flexible loads such as electric vehicles to charge more during off-peak periods (such as early morning and midday) and charge less or send power back to the grid during peak periods, thereby effectively "smoothing" the peak load of the evening peak and alleviating the pressure on grid expansion. Looking at the photovoltaic power comparison on the right, the measured photovoltaic value, represented by the orange solid line, shows a clear inverted "U" shaped trend during the daytime (approximately 6:00 AM to 6:00 PM), peaking between 12:00 PM and 2:00 PM, and then gradually declining as sunlight weakens. The photovoltaic forecast value, represented by the blue dashed line, also accurately replicates this pattern. High-precision photovoltaic forecasting is crucial for the safe and economical operation of the power grid. In traditional models, the randomness and volatility of photovoltaic power generation can easily lead to curtailment or grid frequency fluctuations. However, with the help of this solution, aggregators and grid managers can predict the output of new energy sources in advance for each period. When a large photovoltaic output is predicted at noon while the regular load is relatively low, the system will automatically trigger a low-price incentive mechanism to encourage more electric vehicles to connect to the grid and consume surplus green electricity; when a sharp drop in photovoltaic output is predicted in the evening while residential load begins to rise, the system will reserve adjustment resources in advance to ensure a smooth power supply transition. In summary, this solution, through precise forecasting and coordinated management of load and renewable energy output, helps aggregators achieve refined allocation of charging and discharging resources at the commercial operation level, maximizing the use of low-priced green electricity and peak-valley price differences, significantly reducing operating costs and improving service quality. At the grid management level, the solution effectively smooths load fluctuations, promotes renewable energy consumption, and significantly enhances the distribution network's capacity to accommodate high proportions of distributed energy and its operational resilience, fully demonstrating the deep integration and core value of information and communication technologies in the commercial dispatch and administrative supervision of modern power systems.

[0067] In some embodiments, the master-slave game decision model is reconstructed into a single layer based on the Caro-Kun-Tucker conditions, and the complementary relaxation conditions are linearized to generate a mixed-integer linear programming problem, including: The constraint boundary of the lower-level response model is analyzed, the Lagrangian function corresponding to the lower-level response model is written, and the Carlow-Kun-Tucker condition and complementary relaxation condition are derived.

[0068] Among them, the constraint boundary of the lower-level response model refers to the set of physical and logical constraints that electric vehicle users must follow when optimizing their charging behavior. It includes at least the upper and lower limits of charging power (in kilowatts), the state of charge (SOC) change boundary, and the power demand boundary at the time of off-grid. The Lagrange function is an auxiliary function formed by linearly weighting the objective function of the lower-level response model and its constraint boundary by introducing Lagrange multipliers (LM). It is used to transform the constrained optimization problem into a solution form of an unconstrained optimization problem. The Carlow-Kuhn-Tucker condition is a set of equations in which the gradient of the Lagrange function is a zero vector and all multipliers are non-negative. It is a necessary first-order condition for determining the optimal solution of a nonlinear programming problem. The complementary relaxation condition refers to the mathematical relationship that the product of the Lagrange multiplier and the relaxation variable of the corresponding constraint boundary is always equal to zero. It reflects the duality that "the multiplier is non-zero when the constraint is in effect and zero when the constraint is not in effect".

[0069] Specifically, the system can use a symbolic computation engine to parse the equality constraints on the dynamic update of the battery state of charge in the lower-level response model, as well as the inequality constraints on the upper limit of charging power and off-grid power. It can automatically write the Lagrangian function containing the above constraints and derive the Carrousel-Kuhn-Tucker condition equation system containing 48 equations and 52 variables through partial derivative operations, while identifying 12 sets of complementary relaxation conditions. For example, for the charging power constraint "0≤P{n,i,t}^{EV,c}≤P{max}^{EV,c}" of the i-th electric vehicle at the n-th control node, the system lists the corresponding complementary relaxation conditions: "μ{1,n,i,t}×P{n,i,t}^{EV,c}=0" and "μ{2,n,i,t}×(P{max}^{EV,c}-P{n,i,t}^{EV,c})=0", where μ{1,n,i,t} and μ_{2,n,i,t} are the corresponding Lagrange multipliers.

[0070] Add the Caro-Kun-Tucker conditions to the constraint set of the upper-level optimization problem to construct a single-level optimization data structure.

[0071] Among them, the constraint set of the upper-level optimization problem refers to the set of restrictions that aggregators must meet when formulating pricing strategies, including price fluctuation range constraints and full-cycle average price constraints; the single-level optimization data structure refers to transforming the original two-level game model with nested upper and lower levels of logic into a flat data structure containing only single-level variables and single-level constraints through mathematical transformation. This structure is convenient for mathematical programming solvers to uniformly identify and solve.

[0072] Specifically, the system can directly insert the 48 derived Caro-Kuhn-Tucker conditional equations as new algebraic constraints into the constraint list of the upper-level aggregator pricing optimization model. It also includes decision variables (such as charging power variables) from the lower-level response model and decision variables (such as pricing variables) from the upper-level model into the same variable pool, constructing a single-level optimization data structure containing approximately 2000 variables and 1500 constraints. For example, the system can directly add constraints derived from the lower level, such as “λ{n,t}^P-μ{1,n,i,t}+μ{2,n,i,t}+...=0”, to the constraint matrix of the upper-level model, ensuring that the upper-level pricing variable λ{n,t}^P and the lower-level power variable P_{n,i,t}^{EV,c} are subject to global constraint verification within the same optimization framework.

[0073] By introducing binary decision variables, the Big M method is used to transform complementary relaxation conditions into linear inequality constraints.

[0074] Among them, binary decision variables refer to integer variables that take only 0 or 1 values, and are often used in optimization models to represent logical relationships or switch states; the Big-M method refers to a mathematical modeling technique that introduces a sufficiently large positive number (usually denoted as M, with a value much larger than the order of magnitude of other parameters in the model) and binary decision variables to transform the nonlinear complementary relationship of "product equal to zero" into a system of linear inequalities; linear inequality constraints refer to constraints composed of linear expressions, expressed in the form of "less than or equal to" or "greater than or equal to", and are the standard form of mixed integer linear programming problems.

[0075] Specifically, for each set of complementary relaxation conditions identified, the system can introduce a corresponding binary decision variable and set a sufficiently large positive value. Two sets of linear inequality constraints are constructed to equivalently transform the nonlinear complementary relaxation condition into a linear constraint form acceptable to the mixed-integer linear programming problem. For example, for the complementary relaxation condition "μ{1,n,i,t}×P{n,i,t}^{EV,c}=0", the system introduces a binary decision variable b{n,i,t} and constructs a set of linear constraints: "0≤μ{1,n,i,t}≤ ×(1-b{n,i,t})” and “0≤P{n,i,t}^{EV,c}≤ "×b{n,i,t}". When b{n,i,t}=1, the charging power variable P{n,i,t}^{EV,c} can take positive values, but the multiplier μ{1,n,i,t} must be zero; when b{n,i,t}=0, the multiplier μ{1,n,i,t} can take positive values, but the charging power variable P_{n,i,t}^{EV,c} must be zero, thus perfectly replicating the complementary relaxation logic.

[0076] Based on the optimal duality theory, a linear equivalent replacement is performed on the nonlinear terms in a single-layer optimized data structure to generate a mixed-integer linear programming problem.

[0077] Among them, optimal duality theory refers to the mathematical theory that studies the intrinsic relationship between optimization problems (primary problems) and their dual problems. Through the weak duality theorem and the strong duality theorem, certain nonlinear terms can be replaced by linear expressions of their dual spaces. Nonlinear terms refer to expressions that include variable products (such as the product of price and power variables) or nonlinear functions. Mixed integer linear programming (MILP) is a class of standardized mathematical programming problems that include continuous variables, integer variables and linear constraints, and can be solved efficiently by commercial solvers.

[0078] Specifically, the system can identify bilinear terms in the form of "pricing variable × power variable" in single-level optimization data structures, and replace them with linear expressions based on Caro-Kuhn-Tucker multipliers using optimal duality theory. This eliminates the nonlinear characteristics in the model and ultimately generates a standard mixed-integer linear programming problem containing approximately 3000 continuous variables, 500 binary decision variables, and 2500 linear constraints. For example, the system replaces the nonlinear term "∑(λ{n,t}^P×P{n,i,t}^{EV,c})" in the original upper-level objective function with the linear expression "-∑(μ{2,n,i,t}×P{max}^{EV,c}+...)+∑(-λ_t^{LMP}×P{n,i,t}^{EV,c})", where μ{2,n,i,t} is the dual variable in the Caro-Kuhn-Tucker conditions. This transforms the entire model into a mixed-integer linear programming problem that can be directly processed by solvers such as Gurobi or CPLEX.

[0079] Therefore, according to the above implementation method, the system can transform the complex two-level master-slave game model into a mixed-integer linear programming problem that can be efficiently processed by a computer. Without losing the lower-level user response logic, the solution complexity is greatly reduced, providing a feasible computational path for subsequent large-scale power grid dispatch optimization.

[0080] In some embodiments, the mixed-integer linear programming problem is iteratively optimized to extract the solution vector of the mixed-integer linear programming problem, resulting in an optimal response power benchmark dataset, including: Perform iterative optimization calculations on mixed-integer linear programming problems and output a high-dimensional solution vector containing continuous and discrete variable components.

[0081] Among them, iterative optimization calculation refers to the mathematical programming solver using a joint search algorithm that combines the branch and bound method (B&B) with the cutting plane method (CPM). By continuously dividing the feasible region and pruning invalid branches in a multidimensional space defined by constraints, it seeks the combination of variables that makes the objective function reach its optimal value. The high-dimensional solution vector refers to the one-dimensional array generated after the optimization calculation converges. This array stores the optimal values ​​of all decision variables in the model in sequence, including continuous variable components representing continuous physical quantities (such as charging power and voltage amplitude) and discrete variable components representing logical states (such as binary decision variables and integer variables).

[0082] Specifically, the system can call a high-performance mathematical programming solver (such as Gurobi 10.0 or CPLEX 22.1) deployed on a cloud server cluster, load the mixed integer linear programming problem file generated by the preprocessing module, set the solution parameters as follows: relative gap tolerance of 10^{-4}, time limit of 300 seconds, and number of concurrent threads of 16, start the solver kernel to perform iterative calculations until the solution status returns "Optimal" or the time limit is reached, and then extract the high-dimensional solution vector containing the values ​​of all variables from the memory stack. For example, the system calls the Gurobi solver to solve a mixed-integer linear programming problem with 3240 variables (2240 ​​of which are continuous variables and 1000 are binary decision variables) and 2800 constraints. After 127 branch-and-bound iterations and 43 seconds of computation time, it outputs a one-dimensional floating-point array of length 3240 as a high-dimensional solution vector. The first to second 2240 bits of this array store the continuous variable components, and the second to third 3240 bits store the discrete variable components.

[0083] The high-dimensional solution vector is analyzed, and the charging power values ​​mapped to each control node and each scheduling period are extracted to generate standardized response records.

[0084] The standardized response record refers to a structured data entry used to uniformly describe the optimal charging behavior of electric vehicles under specific spatiotemporal coordinates. Each record contains at least a unique identifier for the control node (Node ID, no unit), a scheduling time index (Time Index, no unit), and the optimal charging power value (in kilowatts). It serves as a bridge connecting mathematical optimization results with actual scheduling operations.

[0085] Specifically, the system can traverse the index positions corresponding to the charging power decision variables in the high-dimensional solution vector according to the preset Variable Index Mapping Table (VIMT), read the corresponding floating-point values, and bind these values ​​with the node identifier and time period index corresponding to the variable index. It then assembles these values ​​into a standardized response record according to the predefined JavaScript Object Notation (JSON) format. For example, the system knows from the mapping table that the variable with index 1024 corresponds to the charging power of "Node ID: N23" at "Time Period Index: T08". It reads the value of 18.5 from the solution vector at this position and generates a standardized response record: "{"node_id":"N23","time_index":8,"power_kw":18.5}". This record indicates that the optimal response power of node N23 is 18.5 kilowatts in the 8th time period of the following day.

[0086] A multidimensional index structure is constructed based on standardized response records. The multidimensional index structure is associated with node index keys, time-series index keys, and power value fields.

[0087] Among them, the multidimensional index structure refers to a data organization method that supports fast retrieval under multiple conditions. It is usually constructed in the form of B+ tree (B-plus tree) or hash dictionary (HD). By simultaneously establishing the node-dimensional index key (NK) and the time-series-dimensional index key (TK), it can achieve millisecond-level positioning of massive response records. The power value field refers to the data slot in the multidimensional index structure used to store the specific physical quantity value, and it usually maintains a strong consistency association with the index key.

[0088] Specifically, the system can batch-write the generated standardized response records into a hash set of an in-memory database (such as Redis), using the concatenated string "node ID:time period index" as the primary key (PK) and the power value as the value, thus constructing a key-value mapping relationship. This forms a multi-dimensional index structure that supports efficient queries by node, by time period, or by a combination of node and time period. For example, the system writes 1000 standardized response records into a Redis cluster, constructs a hash mapping with the key name "N23:8" and the key value "18.5", and simultaneously creates a B+ tree index file with the node ID as the first-level index and the time period index as the second-level index. This allows the scheduling module to return the result within 1 millisecond without traversing the entire table when querying the baseline power of node "N23" in the "T08" time period.

[0089] The multidimensional index structure is persistently stored in the global data plane to form the optimal response power benchmark dataset.

[0090] Among them, persistent storage refers to the process of dumping volatile data structures in memory to non-volatile storage media (such as solid-state drives or distributed file systems) to ensure that data is not lost after system restart or failure; the global data plane refers to the logical area in the system used to uniformly store and manage cross-module shared data, which is usually implemented using a distributed file system (such as Hadoop Distributed File System, abbreviated as HDFS) or object storage service (such as Amazon S3) to provide a unified data access interface for upper-layer applications; the optimal response power benchmark dataset refers to a set of read-only data files formed by solidifying a multi-dimensional index structure, which can be called by subsequent scheduling optimization models.

[0091] Specifically, the system can call the application programming interface (API) of the distributed file system to serialize the multidimensional index structure in memory into a binary data stream or a comma-separated values ​​(CSV) format file, add a timestamp, and write it to a specified directory in the global data plane. Simultaneously, it updates the metadata index table in the global data plane, recording the storage path, creation time, and data size of the dataset. For example, the system uses the Snappy compression algorithm to compress the multidimensional index structure in memory into a binary package, stores it as a data file with the path " / gdp / ev_data / baseline / 20260115_0800.dat" through the HDFS API, with a file size of approximately 2.5 megabytes, and registers 2880 records (corresponding to 24 time periods × 120 nodes) in the metadata table for subsequent loading and use by the distribution network scheduling optimization model.

[0092] Therefore, according to the above implementation method, the system can transform the abstract mathematical optimization results into a structured, persistently stored benchmark dataset. This not only enables the seamless transfer of market game decision results to the physical scheduling level, but also ensures efficient reading and writing of massive amounts of data through multi-dimensional indexing technology, laying a solid data foundation for building a market-operation collaborative charging guidance strategy.

[0093] In some embodiments, the steps of constructing a distribution network dispatch optimization model that takes into account the optimal response power benchmark dataset include: Access the global data plane, calculate the deviation between the power values ​​of the scheduling power variable and the power benchmark dataset of the optimal response power, and construct a power deviation penalty term based on the deviation.

[0094] Among them, the deviation refers to the absolute difference between the charging power command value issued in the real-time scheduling decision and the corresponding power value in the optimal response power benchmark dataset stored in the global data plane, with the unit being kilowatts; the power deviation penalty term refers to a scalar function that quantifies the economic cost or adjustment cost brought about by the above deviation, which is usually modeled as the product of the deviation and the preset penalty coefficient, with the unit being yuan, and is used to suppress the disconnect between the market game result and the physical scheduling execution.

[0095] Specifically, the system can call the application programming interface of the distributed file system to read the data file with the extension ".dat" under the storage path " / gdp / ev_data / baseline / ", deserialize the file content into a multi-dimensional index structure, then traverse all spatiotemporal coordinates, calculate the absolute value of the difference between the real-time scheduling power variable (in kilowatts) and the corresponding power value (in kilowatts) in the benchmark dataset, and multiply the absolute value by the preset deviation penalty coefficient α (in yuan per kilowatt), and accumulate them to obtain the power deviation penalty term. For example, after the system reads that the baseline power value of node N23 in time period T08 is 18.5 kW, if the scheduling power variable generated by the real-time scheduling module is 20.0 kW, then the absolute deviation is calculated as |20.0−18.5|=1.5 kW; the system presets the deviation penalty coefficient α to be 2.0 yuan per kW, then the instantaneous penalty value under this spatiotemporal coordinate is 2.0×1.5=3.0 yuan; the system traverses all 2880 spatiotemporal coordinate points, adds up all the instantaneous penalty values, and constructs a total power deviation penalty term of 1560.5 yuan.

[0096] The power deviation penalty term is superimposed on the original objective function of the initial distribution network dispatch optimization model to generate a composite objective function.

[0097] Among them, the original objective function of the initial distribution network dispatch optimization model refers to the minimization function that takes ensuring the safe operation of the power grid as the core and only considers the physical operating cost. It usually includes the unit start-up and shutdown cost, coal cost and network loss cost. The composite objective function refers to a new type of optimization objective that integrates economic incentive components (such as power deviation penalty terms) on the basis of the original objective function, aiming to achieve the comprehensive optimization of physical operating cost and market regulation cost.

[0098] Specifically, the system can insert the power deviation penalty term obtained through analysis as a new sub-term into the summation formula of the original objective function, constructing a dual minimization objective that includes both physical cost and deviation penalty, and compiling this composite objective function into a mathematical expression tree that the solver can recognize. For example, the system obtains the original objective function expression of the initial distribution network scheduling optimization model as " ",in For generator set operating costs, For network loss costs; the system adds the previously calculated power deviation penalty term Ppenalty (valued at 1560.5 yuan) to the objective function, generating a composite objective function expression " The expression is compiled into an expression tree containing addition and multiplication nodes, which is then used by the optimization engine.

[0099] The distribution network topology parameters are analyzed, and node voltage amplitude constraints, branch current thermal stability constraints, and node power balance constraints are extracted. These constraints are then converted into constraint matrices.

[0100] Among them, node voltage amplitude constraint refers to the inequality restriction that forces the effective value of the voltage at each node of the distribution network to be maintained within the rated range (usually 0.95 to 1.05 times the per-unit value) in order to ensure power quality; branch current thermal stability constraint refers to the inequality restriction that forces the effective value of the current flowing through each branch to not exceed the thermal stability limit of the conductor in order to prevent the conductor from overheating and melting; node power balance constraint refers to the equality restriction that forces the sum of the active power flowing into any node and the power flowing out to be zero according to Kirchhoff's laws; constraint matrix refers to the data structure that compactly stores all the above constraints in matrix form, usually including coefficient matrix A (real number matrix), right-hand constant vector b (real number vector) and variable type vector c (integer vector), which conforms to the standard mathematical programming expression form of "Ax≤b".

[0101] Specifically, the system can load standard format distribution network topology files (such as MatPower format or IEEE standard example files), parse the node admittance matrix and branch impedance parameters, derive the mathematical expressions for the above three types of constraints based on Ohm's law and Kirchhoff's law, extract the variable coefficients in the expressions, fill them into the corresponding columns of the coefficient matrix A, and fill the constant threshold into the constant vector b on the right. For example, the system loads the topology data of the IEEE 33-node test system and parses the impedance parameters of 32 branches; sets the node voltage amplitude constraint range to 0.95 kV to 1.05 kV and the branch current thermal stability constraint upper limit to 0.4 kA; the system automatically generates a coefficient matrix A with a size of 132 rows (number of constraints) multiplied by 1056 columns (number of variables), where rows 1 to 33 correspond to the node power balance equation, rows 34 to 65 correspond to the lower limit constraint of node voltage amplitude, rows 66 to 97 correspond to the upper limit constraint of node voltage amplitude, and rows 98 to 129 correspond to the branch current thermal stability constraint.

[0102] By setting the constraint matrix as the feasible region boundary and mapping the composite objective function to the feasible region boundary, a distribution network dispatch optimization model considering the optimal response power benchmark dataset is obtained.

[0103] The feasible region boundary refers to the closed geometric region enclosed by the constraint matrix A and the vector b in the mathematical space. Any point in this region represents a set of potential scheduling schemes that satisfy all physical operation constraints. Mapping refers to the computational process of establishing a connection between an abstract mathematical function (such as a composite objective function) and a specific geometric space (such as the feasible region boundary), so that the optimization solution process is restricted to the safe region.

[0104] Specifically, the system can associate and bind the constructed composite objective function expression with constraint matrix A and vector b, encapsulating it into a standard mathematical programming model object (MO). The solution domain of this model object is explicitly set to a multidimensional convex set defined by the constraint matrix, thereby completing the construction of the distribution network scheduling optimization model. For example, the system can encapsulate the aforementioned composite objective function " "The constraint matrix of 132 rows multiplied by 1056 columns is encapsulated to generate a model object named "ED_Schedule_Opt_Model". The system sets the solution domain of this model object to "Ax ≤ b", which means that in the subsequent iterative optimization process, any scheduling scheme that attempts to exceed the lower voltage limit of 0.95 kV or the upper current limit of 0.4 kA will be automatically eliminated, thereby ensuring that the output pricing guidance strategy is physically safe."

[0105] Therefore, according to the above implementation method, the system can construct a collaborative mechanism between market game results and physical power grid operation. It achieves a combination of economic incentives and physical constraints through the power deviation penalty term, and ensures that all scheduling decisions are executed within the power grid safety margin through the rigid boundary of the constraint matrix. This effectively solves the technical problem of strategy failure caused by neglecting physical feasibility in traditional pricing methods.

[0106] In some embodiments, a risk metric model based on probability distribution distance is used to perform sub-Bruker optimization on the distribution network dispatch optimization model, outputting a pricing guidance strategy, including: Based on the empirical distribution of historical source load data samples, and according to the preset probability distribution distance, the fluctuation boundary of prediction error is defined, and a fuzzy set covering multiple scenarios is constructed.

[0107] The historical source-load data sample refers to the collection of actual power output data and actual load consumption data collected over a period of time, including at least distributed photovoltaic power output samples, wind power output samples, and residential and industrial load consumption samples, with a sampling time interval of no more than 1 hour. The empirical distribution refers to the probability distribution obtained by statistically fitting the historical source-load data sample through kernel density estimation (KDE), used to approximately characterize the true distribution law of source-load prediction error. The probability distribution distance is a dimensionless index used to quantify the degree of difference between two probability distributions. This scheme preferably uses the Wasserstein distance (WD), which measures the distribution difference by calculating the minimum "work" required to move a sample from one distribution to another. The fluctuation boundary refers to the boundary that allows the probability distribution to fluctuate within a preset probability distribution distance, centered on the empirical distribution, used to define the maximum impact range of uncertainty. The fuzzy set refers to a feasible set containing multiple probability distributions that meet the fluctuation boundary constraints, used to characterize the uncertainty range of source-load prediction and cover all possible error distributions.

[0108] Specifically, the system can call the data storage module to read historical source-load data samples from the past 1 to 3 years, calculate the difference between the actual and predicted source-load values ​​at each sampling point to obtain a prediction error sequence, and use a Gaussian kernel function to perform kernel density estimation on the prediction error sequence to fit an empirical distribution. Then, it reads a preset probability distribution distance radius parameter, uses the empirical distribution as the center and the preset radius as the sphere, and constructs a fuzzy set containing all probability distributions whose distance from the empirical distribution does not exceed the radius. For example, the system reads 8760 hours of photovoltaic output and load data for a 10kV distribution network in Zhejiang throughout 2024, calculates the photovoltaic output prediction error range to be -28.3 kW to +24.7 kW, and the load prediction error range to be -15.2 kW to +18.6 kW; it uses a Gaussian kernel function with a bandwidth parameter of 2.1 to perform kernel density estimation on the above error sequence and fits an empirical distribution; with a preset Wasserstein distance radius of 0.048, the constructed fuzzy set contains 1024 possible error distribution scenarios that meet the constraints, covering 99.2% of historical fluctuations.

[0109] Establish an affine mapping relationship between scheduling decision variables and uncertainties, embed the affine mapping relationship into the objective function and constraints of the distribution network scheduling optimization model, and generate the decision model to be optimized.

[0110] Uncertainty refers to the random errors generated during the source-load forecasting process, including photovoltaic power output forecasting errors, wind power output forecasting errors, and base load forecasting errors, with units in kilowatts. Affine mapping refers to a linear function association rule that represents the scheduling decision variable as a linear combination of uncertainties, in the form of "scheduling decision variable = intercept term + slope term × uncertainty". The intercept term and slope term are the decision parameters to be optimized, used to achieve adaptive adjustment of scheduling decisions as uncertainties change. The decision model to be optimized refers to a new optimization model that retains all constraints of the original model after substituting the affine mapping relationship into the original distribution network scheduling optimization model, using the intercept term and slope term as optimization variables. The variable dimension increases linearly with the number of scheduling periods.

[0111] Specifically, the system can set a set of affine mapping coefficients for each scheduling period, establish a linear relationship between the scheduling decision variables such as charging scheduling power, distributed power output and conventional unit output in each period and the photovoltaic and load uncertainties in the corresponding period, substitute the relationship into the power balance constraints, voltage constraints and current constraints of the original distribution network scheduling optimization model, replace the original fixed scheduling variables, and generate a decision model to be optimized that only contains affine mapping coefficients as variables. For example, the system sets affine mapping coefficients for each of the 24 scheduling periods on the next day. The affine mapping relationship for the charging scheduling power in the 8th period (08:00~09:00) is P8sch=18.5+0.32×ξ8pv, where 18.5 kW is the intercept term, ξ8pv is the photovoltaic output prediction error in the 8th period (in kW), and 0.32 is the slope term. Substituting this relationship into the node power balance constraints of the original distribution network scheduling optimization model, replacing the original fixed charging power variables, generates a decision model to be optimized that includes 48 intercept term variables, 48 ​​slope term variables, and the original constraints. The total number of variables in the model increases from the original 1056 to 1152.

[0112] By performing a dual transformation on the decision-making model to optimize it, the semi-infinite programming problem containing fuzzy sets is reconstructed into a deterministic mathematical programming problem.

[0113] Among them, semi-infinite programming (SIP) refers to optimization problems with an infinite number of constraints. Since fuzzy sets contain infinitely many possible error distributions, each distribution corresponds to a set of constraints, the decision model to be optimized belongs to the category of semi-infinite programming problems. Dual transformation (DT) refers to a mathematical method based on optimal duality theory that transforms the infinite-dimensional constraints in semi-infinite programming problems into finite-dimensional dual constraints. By introducing dual variables, the extreme value search at the distribution level is transformed into linear computation in the dual space. Deterministic mathematical programming problems refer to conventional optimization problems with a finite number of constraints and decision variables, which can be directly solved by commercial mathematical programming solvers. In this scheme, the problem obtained after transformation is a mixed integer linear programming problem.

[0114] Specifically, the system can call the built-in dual transformation engine to introduce two sets of dual variables for the Wasserstein distance constraint of fuzzy sets. This transforms the infinite-dimensional constraint "which holds true for all distributions in all fuzzy sets" in the decision model to be optimized into a finite-dimensional linear inequality constraint containing only dual variables. At the same time, the expected value calculation in the objective function is transformed into a weighted summation form based on historical samples, eliminating the uncertainty dependency in the model and obtaining a deterministic mathematical programming problem containing only finite variables and finite constraints. For example, for a fuzzy set containing 1024 potential scenarios, the system calls the dual transformation engine to introduce 2048 dual variables, transforming the infinite-dimensional probabilistic constraints in the original semi-infinite programming problem into a deterministic mixed-integer linear programming problem containing 21888 variables and 18272 linear constraints. The variables include 48 intercept terms, 48 ​​slope terms, 2048 dual variables, and other continuous variables of the original model. The constraints include 1056 physical constraints and 17216 dual constraints of the original model. The transformed problem can be directly identified and processed by the Gurobi 10.0 solver.

[0115] Solve a deterministic mathematical programming problem and analytically output the pricing guidance strategy corresponding to each scheduling period.

[0116] Among them, the pricing guidance strategy refers to the charging service price for each dispatch period issued by the aggregator to the electric vehicle users under its jurisdiction, in yuan per kilowatt-hour, which is used to guide users to adjust their charging behavior according to the price signal, while ensuring that the power grid operates within the safety margin.

[0117] Specifically, the system can call a high-performance mathematical programming solver deployed on a cloud server cluster to load the transformed deterministic mathematical programming problem and set the solution parameters to relative gap tolerance. The maximum solution time is 600 seconds, and iterative optimization calculations are performed. After the solution is completed, pricing parameters for each time period are extracted from the optimal solution vector, and after format conversion, a standardized pricing guidance strategy file is generated, stored in the global data plane, and pushed to the aggregator operation platform. For example, the system calls Gurobi. The 10.0 solver solved the aforementioned deterministic mixed-integer linear programming problem. After 213 branch-and-bound iterations and 187 seconds, it obtained the optimal solution. Pricing parameters for 24 time periods were extracted and rounded to two decimal places to generate a pricing guidance strategy: 00:00-01:00 RMB 0.42 per kilowatt-hour, 04:00-05:00 RMB 0.38 per kilowatt-hour, 08:00-09:00 RMB 1.18 per kilowatt-hour, 12:00-13:00 RMB 0.67 per kilowatt-hour, 18:00-19:00 RMB 1.25 per kilowatt-hour, and 22:00-23:00 RMB 0.45 per kilowatt-hour. In simulation tests, this strategy ensured a distribution network voltage qualification rate of 99.93%, a daily average revenue for aggregators of RMB 1278.6, and a 17.4% reduction in average charging costs for users compared to disordered charging.

[0118] Therefore, according to the above implementation method, the system can incorporate the uncertainties on both the source and load sides into the entire pricing decision process, quantify the influence range of prediction errors through fuzzy sets, achieve adaptive adjustment of scheduling decisions through affine mapping, and transform the complex semi-infinite programming problem into a solvable deterministic model through dual transformation. The final output pricing guidance strategy takes into account both the economic efficiency of market operation and the physical security of power grid operation, effectively solving the problem of insufficient adaptability of traditional pricing methods in high-penetration distributed power access scenarios.

[0119] In some embodiments, before fitting an empirical model based on historical source load data samples, preprocessing of the historical source load dataset is further included, including: The source load historical dataset is sorted according to the preset risk measurement index, and the source load historical dataset is divided into tail dataset and normal dataset based on the sorting results.

[0120] Among them, the risk measurement index refers to the assessment parameter used to quantify the degree of threat to the power grid operation safety of the source load data sample. This scheme preferably adopts the Conditional Value at Risk (CVaR) as the core index, which is used to measure the expected value of potential loss when the source load prediction error exceeds the threshold at a given confidence level. The source load historical dataset refers to a multi-dimensional data set composed of historical source load data samples arranged in chronological order. Each row represents a sampling time, and each column represents the prediction error value of a source load type (such as photovoltaic power output, wind power output, and base load). The tail dataset refers to the sample set selected from the source load historical dataset where the risk measurement index exceeds the preset threshold, representing extreme operating conditions that may lead to power grid safety risks. The normal dataset refers to the data sample set in the source load historical dataset other than the tail dataset where the risk measurement index is within the normal range, representing typical operating conditions of the power grid in daily operation.

[0121] Specifically, the system can invoke the risk assessment engine to iterate through each data sample in the historical source-load dataset, calculate the corresponding conditional risk value for that sample, and sort all samples in descending order of conditional risk value. Then, based on a preset quantile threshold (e.g., the 95th percentile), the sorted samples are divided into two parts: the top 5% are assigned to the tail dataset, and the remaining 95% are assigned to the normal dataset. For example, the system loads a historical source-load dataset containing 8760 hours of sampling points from the past year. Each sampling point includes two dimensions: photovoltaic prediction error and load prediction error. With a confidence level of 95%, the system calculates the conditional risk value for each hourly sampling point to be in the range of -12.8 yuan to +15.4 yuan. After sorting the conditional risk values ​​from high to low, 438 samples (5%) with conditional risk values ​​greater than 12.0 yuan are assigned to the tail dataset, and the remaining 8322 samples are assigned to the normal dataset. The tail dataset mainly corresponds to scenarios of sudden drops in photovoltaic output or sudden increases in load under extreme weather conditions.

[0122] Perform cluster analysis on the tail dataset to extract a subset of extreme scenarios, and perform cluster analysis on the normal dataset to extract a subset of typical scenarios.

[0123] Clustering analysis is an unsupervised learning algorithm that automatically groups similar data samples into several clusters by calculating the feature distances between samples (such as Euclidean distance or Manhattan distance). This scheme preferably adopts the density-based spatial clustering of applications with noise (DBSCAN) or the K-means clustering algorithm (KMA). The extreme scenario subset refers to a representative set of extreme operating condition samples extracted from the tail dataset. Redundant similar samples are removed through clustering analysis, and typical scenarios that best reflect the characteristics of extreme risks are retained. The typical scenario subset refers to a set of samples extracted from the normal dataset that reflects the daily operation patterns of the power grid and is used to construct the main part of the experience distribution.

[0124] Specifically, for the tail dataset, the system can set a smaller neighborhood radius and a lower minimum number of samples, and use a density-based spatial clustering algorithm to identify multiple extreme scene clusters, extract the centroid of each cluster or the most representative sample point to form an extreme scene subset; for the normal dataset, the system can determine the appropriate number of clusters based on experience or the elbow rule, and use the K-means clustering algorithm to cluster the massive normal samples into several typical clusters, and extract the centroid of each cluster as a typical scene subset. For example, the system performed density-based spatial clustering analysis on a tail dataset containing 438 samples, setting the neighborhood radius to 0.8 and the minimum number of samples to 5. It successfully identified three extreme scenario clusters, corresponding to three extreme operating conditions: "PV midday output gap," "evening peak load superposition," and "nighttime load surge." The centroids of the three clusters were extracted to form a subset of extreme scenarios containing three samples. On a normal dataset containing 8322 samples, K-means clustering analysis was performed. Based on the elbow rule, the optimal number of clusters was determined to be 50. The centroids of 50 clusters were extracted to form a subset of typical scenarios containing 50 samples, compressing the total number of samples from 8760 to 53, achieving a data compression rate of 99.4%.

[0125] Configure a first weighting coefficient for the extreme scenario subset and a second weighting coefficient for the typical scenario subset. Generate a training dataset based on the extreme scenario subset, the typical scenario subset and the corresponding weighting coefficients, and input the training dataset into the risk measurement model.

[0126] The first weighting coefficient refers to the weight value assigned to each sample in the extreme scenario subset. It is usually set to a high value (e.g., greater than 1 / N, where N is the number of samples in the extreme scenario subset) to amplify the impact of extreme scenarios in the risk measurement model and reflect the importance of tail risk. The second weighting coefficient refers to the weight value assigned to each sample in the typical scenario subset. It is usually normalized according to the frequency of the sample in the original dataset or the cluster size and is used to construct an empirical distribution that reflects the daily operating rules. The training dataset is a structured data set composed of the extreme scenario subset, the typical scenario subset and their corresponding weighting coefficients. It is used to train the risk measurement model so that it can accurately perceive the distribution characteristics of source load uncertainty.

[0127] Specifically, the system can configure weights using either inverse frequency weighting or uniform weighting based on the number of samples in the extreme scenario subset and the typical scenario subset. To ensure the risk measurement model's sensitivity to extreme scenarios, the first weighting coefficient can be set to a value much larger than the second weighting coefficient, resulting in a higher "contribution" of extreme scenarios to the probability distribution distance when constructing the fuzzy set subsequently. For example, the system configures the first weighting coefficient for an extreme scenario subset containing 3 samples, with each sample having a weight of 0.1 and a total weight of 0.3; and configures the second weighting coefficient for a typical scenario subset containing 50 samples, with each sample having a weight of 0.014 after normalization based on its cluster size, and a total weight of 0.7. This generates a training dataset containing 53 samples and their corresponding weights, where, although the extreme scenario sample size is small, its total weight accounts for 30%, far exceeding its 5% proportion in the original dataset. The system stores this training dataset in comma-separated value format and inputs it into the risk measurement model through the application programming interface for fitting empirical distributions and constructing fuzzy sets.

[0128] Therefore, according to the above implementation method, the system can accurately identify and enhance extreme risk scenarios in historical source-load data through data preprocessing mechanism, effectively solving the shortcomings of traditional uniform distribution sampling in dealing with long-tail risks; the data dimensionality is greatly reduced through cluster analysis, improving the solution efficiency of subsequent sub-Bruker optimization problems; and the trained risk measurement model can more sensitively capture tail risk characteristics through differentiated configuration of weighting coefficients, thereby outputting a more robust pricing guidance strategy under the premise of ensuring power grid safety.

[0129] In some embodiments, prior to basing the data on bilateral market interactions obtained from aggregators and electric vehicle users, the method further includes: Perform desensitization processing on the original model files and device configuration files stored in the distributed persistent storage area to mask sensitive fields in the original model files and device configuration files.

[0130] Among them, the distributed persistent storage area refers to a non-volatile data storage space implemented with a distributed architecture, possessing data redundancy backup and fault recovery capabilities, used for long-term preservation of core data assets for power system planning and operation; the original model file refers to a structured data file that conforms to power system data exchange standards (such as the Common Information Model, or CIM) and contains power grid topology connections and equipment parameters, with the file extension usually ".xml" or ".rdf"; the equipment configuration file refers to a key-value pair collection that records the static attributes and dynamic control parameters of specific power equipment (such as smart meters, charging piles, and photovoltaic inverters), usually stored in JSON or Extensible Markup Language (XML) format; de-identification processing refers to the process of removing or hiding information involving personal privacy, trade secrets, or critical infrastructure in the data through data transformation, replacement, or masking, aiming to ensure the security of data sharing and processing; sensitive fields refer to data items contained in the original model file and equipment configuration file that, if leaked, may cause damage to personal rights or public safety, including at least the user's name, ID number, contact number, precise geographical coordinates of the equipment, and the name of the property owner.

[0131] Specifically, the system can call the de-identification engine deployed in the secure isolation zone, load the predefined de-identification rule library, and perform a full field scan of the original model files and device configuration files in the distributed persistent storage area. For the identified sensitive fields, an irreversible hash algorithm (such as Secure Hash Algorithm 256, or SHA-256 for short) is used to convert identifier fields, or format-preserving encryption (FPE for short) is used to encrypt numeric fields with fixed formats. The de-identified data is then overwritten to the original storage path or exported to a new file. For example, if the system scans a device configuration file containing the fields "user_name": "Zhang San", "id_card": "330102199001011234", and "geo_coord": "120.1551,30.2741", according to the de-identification rules, the system will replace "user_name" with "user_8f9a2c", "id_card" with "330102********1234", and "geo_coord" with "120.15xx,30.27xx", generating a de-identified file to ensure that the original privacy information cannot be reverse-engineered.

[0132] In response to the project instantiation request, the original model file and device configuration file after de-identification are loaded, and a business operation template containing job activity instances is compiled and generated.

[0133] Among them, an engineering instantiation request refers to a trigger signal issued by a system administrator or external scheduling system to initiate a specific charging guidance task. This request usually includes a task identifier, execution time window, and target control area. Compilation and generation refers to the process of converting human-readable or standard-format structured data files into internal data structures or bytecode that can be efficiently loaded and executed by computer programs. A job activity instance refers to the logical encapsulation of a specific task (such as adjusting charging power during a specific period) performed by a specific device (such as a charging pile) in a specific engineering instance. It includes task input parameters, execution status, and output result placeholders. A business operation template refers to a blueprint file composed of multiple interrelated job activity instances combined according to specific temporal logic and dependencies, used to guide the automated execution of the system. It usually exists in memory in the form of a Directed Acyclic Graph (DAG).

[0134] Specifically, the system can parse the task identifier in the engineering instantiation request, locate the corresponding anonymized original model file and device configuration file, call the model parser to load the file content into memory, and instantiate each job activity according to the topological connection relationship and preset business logic in the file, establishing dependency edges between job activities (e.g., "Charging power optimization" can only be executed after "photovoltaic power output prediction" is completed), and finally generating a business operation template object containing all job activity instances and their topological relationships. For example, when the system receives a charging guidance task request for "Hangzhou Binjiang District, January 15, 2026", it loads the anonymized IEEE 33-node model file and configuration files of 120 charging piles for that area; the system instantiates a total of 352 job activity instances, such as "data acquisition", "power prediction", "optimization calculation", and "policy issuance", and builds dependency relationships: only when the status of the "power prediction" instance changes to "completed" will the "optimization calculation" instance that depends on it be scheduled for execution, and finally generates a business operation template containing 352 nodes and about 500 edges, stored as a directed acyclic graph structure in memory.

[0135] Establish a mapping relationship between communication sessions and unique identifiers of job activity instances, parse the semantic features of communication sessions, and generate structured pending instructions associated with job activity instances.

[0136] Among them, a communication session refers to a logical connection with the system established by a user (such as an aggregator dispatcher) through a human-machine interface (HMI) and has an independent lifecycle, usually uniquely identified by a session ID (SID); a unique identifier for a job activity instance refers to a unique string code assigned to each job activity instance across the entire network, used to accurately locate the instance in the distributed system; semantic features refer to the business meaning contained in the text, voice, or operation instructions carried in the communication session, such as "start the 08:00 optimization task" or "query the voltage of node N23"; structured pending instructions refer to the standardized instructions that the unstructured or semi-structured semantic features parsed from the communication session are converted into standardized instructions that the system can automatically execute, containing clear actions, objects, and parameters, usually conforming to JSON format.

[0137] Specifically, the system can bind the session identifier to the business operation template identifier associated with the current project instance at the beginning of the communication session; when it receives the operation instruction sent by the user through the communication session, it calls the Natural Language Processing (NLP) module to parse the semantic features of the instruction, identify the unique identifier of the target job activity instance and the operation to be performed, and then generate a structured to-do instruction containing the operation type, target instance identifier and input parameters, and push the instruction into the message queue of the corresponding job activity instance. For example, a user initiates a session through the front-end interface, with the session identifier being "SID_abc123". The system binds this session to the current business operation template "TEMPLATE_20260115". When the user inputs the text command "Please execute optimization calculation immediately", the system parses the semantic features, identifies the intent as "execute", and determines the target as the "optimization calculation" job activity instance (unique identifier "ACT_OPT_001"). The system then generates a structured to-do instruction: {"action":"execute","target_id":"ACT_OPT_001","params":{}}, and sends this instruction to the message queue of the "ACT_OPT_001" instance, triggering it to transition from a waiting state to a ready state.

[0138] Therefore, according to the above implementation method, the system can build a tight security defense at the source of data processing, effectively avoid the risk of sensitive information leakage through desensitization processing, and meet the data security compliance requirements of the power industry; by compiling and generating business operation templates, the system realizes the transformation from static model to dynamic executable task, and improves the system's automated processing capability; by establishing a mapping between sessions and instances and generating structured instructions, the system opens up the channel between human-computer interaction and background automated execution.

[0139] According to a second aspect, this invention provides a market and operational collaborative electric vehicle charging guidance system that considers source-load uncertainty. This system is deployed on a cloud server cluster. The system includes a processing unit and a storage unit. The storage unit stores executable instructions, which, when invoked by the processing unit, implement any of the electric vehicle charging guidance methods described in this invention.

[0140] According to a third aspect, the present invention provides an electronic device comprising: at least one processing unit; and a storage unit communicatively connected to the at least one processing unit. The storage unit stores executable instructions that can be invoked by the at least one processing unit. When the executable instructions are executed by the at least one processing unit, the at least one processing unit causes it to execute any electric vehicle charging guidance method according to an embodiment of the present invention.

[0141] According to another aspect of the present invention, a medium for storing executable instructions is provided, wherein the computer instructions are used to cause a computer to execute any of the electric vehicle charging guidance methods in the embodiments of the present invention.

[0142] Figure 5 A schematic block diagram of an electronic device 600 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0143] like Figure 5 As shown, the electronic device 600 includes a computing unit 601, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 602 or a computer program loaded from a storage unit 608 into a random access memory (RAM) 603. The RAM 603 may also store various programs and data required for the operation of the electronic device 600. The computing unit 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0144] Multiple components in electronic device 600 are connected to I / O interface 605, including: input unit 606, such as keyboard, mouse, etc.; output unit 607, such as various types of displays, speakers, etc.; storage unit 608, such as disk, optical disk, etc.; and communication unit 609, such as network card, modem, wireless transceiver, etc. Communication unit 609 allows electronic device 600 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0145] The computing unit 601 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 601 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 601 performs the various methods and processes described above, such as an electric vehicle charging bootstrapping method. For example, in some embodiments, an electric vehicle charging bootstrapping method can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 608. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 600 via ROM 602 and / or communication unit 609. When the computer program is loaded into RAM 603 and executed by the computing unit 601, one or more steps of an electric vehicle charging bootstrapping method described above can be performed. Alternatively, in other embodiments, the computing unit 601 can be configured to perform an electric vehicle charging bootstrapping method by any other suitable means (e.g., by means of firmware).

[0146] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0147] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0148] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0149] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0150] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0151] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0152] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0153] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for guiding electric vehicle charging, characterized in that, include: Based on bilateral market interaction data obtained from aggregators and electric vehicle users, a master-slave game decision-making model is constructed with the aggregator's revenue maximization as the objective function and the user's charging cost minimization as the response constraint. The master-slave game decision model is reconstructed into a single layer based on the Caro-Kun-Tucker conditions, and the complementary relaxation conditions are linearized to generate a mixed integer linear programming problem. The mixed-integer linear programming problem is iteratively optimized to extract the solution vector of the mixed-integer linear programming problem, thereby obtaining the optimal response power benchmark dataset; The power deviation penalty term is incorporated into the objective function, and the grid operation state constraint is embedded as the feasible domain boundary into the initial distribution network scheduling optimization model to construct a distribution network scheduling optimization model that takes into account the optimal response power benchmark dataset. The distribution network scheduling optimization model is subjected to bibliometric optimization by using a risk measurement model based on probability distribution distance, and a pricing guidance strategy is output.

2. The method according to claim 1, characterized in that, The master-slave game decision model is reconstructed into a single layer based on the Caro-Kun-Tucker conditions, and the complementary relaxation conditions are linearized to generate a mixed-integer linear programming problem, including: The constraint boundary of the lower-level response model is analyzed, the Lagrangian function corresponding to the lower-level response model is listed, and the Carlow-Kun-Tucker condition and complementary relaxation condition are derived. The Caro-Kun-Tucker conditions are added to the constraint set of the upper-level optimization problem to construct a single-layer optimization data structure. By introducing binary decision variables, the complementary relaxation conditions are transformed into linear inequality constraints using the Big M method. Based on the optimal duality theory, the nonlinear terms in the single-layer optimized data structure are replaced by linear equivalents to generate the mixed-integer linear programming problem.

3. The method according to claim 2, characterized in that, The iterative optimization of the mixed-integer linear programming problem, extracting the solution vector of the mixed-integer linear programming problem, and obtaining the optimal response power benchmark dataset includes: Perform iterative optimization calculations on the mixed-integer linear programming problem and output a high-dimensional solution vector containing continuous and discrete variable components; The high-dimensional solution vector is analyzed to extract the charging power values ​​mapped to each control node and each scheduling period, and a standardized response record is generated. A multidimensional index structure is constructed based on the standardized response record, and the multidimensional index structure is associated with node index keys, time series index keys, and power value fields. The multidimensional index structure is persistently stored in the global data plane to form the optimal response power benchmark dataset.

4. The method according to claim 3, characterized in that, The steps for constructing the distribution network dispatch optimization model that takes into account the optimal response power benchmark dataset include: Access the global data plane, calculate the deviation between the scheduling power variable and the power value of the optimal response power benchmark dataset, and construct a power deviation penalty term based on the deviation. The power deviation penalty term is superimposed on the original objective function of the initial distribution network dispatch optimization model to generate a composite objective function; The distribution network topology parameters are analyzed, and node voltage amplitude constraints, branch current thermal stability constraints, and node power balance constraints are extracted. The node voltage amplitude constraints, branch current thermal stability constraints, and node power balance constraints are then converted into constraint matrices. The constraint matrix is ​​set as the feasible region boundary, and the composite objective function is mapped to the feasible region boundary to obtain the distribution network dispatch optimization model that takes into account the optimal response power benchmark dataset.

5. The method according to claim 4, characterized in that, The risk measurement model constructed based on probability distribution distance is used to perform bibliometric optimization on the distribution network dispatch optimization model, and output a pricing guidance strategy, including: Based on the empirical distribution of historical source load data samples, the fluctuation boundary of prediction error is defined according to the preset probability distribution distance, and a fuzzy set covering multiple scenarios is constructed. Establish an affine mapping relationship between scheduling decision variables and uncertainties, embed the affine mapping relationship into the objective function and constraints of the power grid scheduling optimization model, and generate a decision model to be optimized. Perform a dual transformation on the decision model to be optimized to reconstruct the semi-infinite programming problem containing the fuzzy set into a deterministic mathematical programming problem; Solve the deterministic mathematical programming problem and analytically output the pricing guidance strategy corresponding to each scheduling period.

6. The method according to claim 5, characterized in that, Before fitting the empirical model based on historical source load data samples, the process also includes preprocessing the historical source load dataset, including: The source load historical dataset is sorted according to a preset risk measurement index, and the source load historical dataset is divided into tail dataset and normal dataset based on the sorting result. Cluster analysis is performed on the tail dataset to extract a subset of extreme scenarios, and cluster analysis is performed on the normal dataset to extract a subset of typical scenarios; Configure a first weighting coefficient for the extreme scenario subset, configure a second weighting coefficient for the typical scenario subset, generate a training dataset based on the extreme scenario subset, the typical scenario subset and the corresponding weighting coefficients, and input the training dataset into the risk measurement model.

7. The method according to claim 1, characterized in that, Prior to relying on bilateral market interaction data obtained from aggregators and electric vehicle users, the following are also included: The original model files and device configuration files stored in the distributed persistent storage area are de-identified to mask sensitive fields in the original model files and device configuration files. In response to the project instantiation request, the original model file and the device configuration file after de-identification are loaded, and a business operation template containing job activity instances is compiled and generated. Establish a mapping relationship between communication sessions and unique identifiers of job activity instances, parse the semantic features of the communication sessions, and generate structured pending instructions associated with the job activity instances.

8. A market- and operationally coordinated electric vehicle charging guidance system considering source-load uncertainty, characterized in that, The system is deployed on a cloud server cluster; The system includes a processing unit and a storage unit. The storage unit stores executable instructions. When the executable instructions are invoked by the processing unit, the electric vehicle charging guidance method as described in any one of claims 1 to 7 is executed.

9. An electronic device, characterized in that, include: At least one processing unit; and a storage unit communicatively connected to the at least one processing unit; The storage unit stores executable instructions that can be invoked by the at least one processing unit. When the executable instructions are executed by the at least one processing unit, the at least one processing unit causes the at least one processing unit to perform the method as described in any one of claims 1 to 7.

10. A medium for storing executable instructions, characterized in that, When the executable instruction is invoked by the processing unit, it causes the processing unit to perform the method as described in any one of claims 1 to 7.

Images