Electric vehicle charging station energy management system and scheduling method

Abstract

The invention discloses an electric vehicle charging station energy management system and a scheduling method. The electric vehicle charging station energy management system comprises an energy management controller, and a wind power generation scheduling module, a photovoltaic power generation scheduling module, an energy storage system scheduling module and an electric vehicle charging control module which are in communication connection with the energy management controller; the energy management controller is used for dividing the operation process of the charging, storing and discharging integrated electric vehicle charging station into a plurality of discrete control periods, obtaining the operation state of each module in each control period, and generating a joint scheduling instruction based on a preset multi-agent constraint Markov decision process model. Cooperative control over wind power generation, photovoltaic power generation, battery energy storage and electric vehicle charging is achieved. According to the electric vehicle charging station energy management system and the scheduling method, the problem that the revenue maximization and the safety of the charging station cannot be considered in a random environment in the prior art is solved.

Description

Technical Field

[0001] This invention belongs to the technical fields of electric vehicle (EV) charging station energy management system (EMS), multi-agent deep reinforcement learning (MARL), constrained Markov decision process (CMDP), and safe reinforcement learning, specifically relating to an electric vehicle charging station energy management system. This invention also relates to a scheduling method for the electric vehicle charging station energy management system. Background Technology

[0002] With the large-scale integration of EVs into the power grid, "charge-storage-discharge integrated" EV charging stations, which combine charging, energy storage, and discharging functions, have become an important infrastructure for alleviating grid pressure and improving the absorption capacity of new energy sources. However, under complex operating conditions, including fluctuations in renewable energy output, uncertainties in EV arrival and departure behavior, and strong coupling of multiple energy units, how to achieve economical, efficient, and coordinated operation of integrated charging stations while ensuring system safety constraints has become a major technical problem that urgently needs to be solved in the current technology.

[0003] Existing integrated charging, storage, and discharging EV charging stations typically consist of EV charging facilities, a Battery Energy Storage System (BESS), distributed renewable energy sources, a two-way grid interface, and an Energy Management System (EMS). The EMS coordinates and controls each unit to achieve the coordinated operation of charging, energy storage, and discharging. Existing technologies for the operation and scheduling of such systems mainly include model-based and optimization-based methods, methods based on single-agent deep reinforcement learning (DRL), and methods based on multi-agent reinforcement learning (MARL). Among these, optimization-based solutions heavily rely on load and renewable energy forecasts, making them ill-suited to highly stochastic operating environments. Single-agent DRL methods merge multiple heterogeneous energy units into a single model, resulting in excessively high dimensionality in the state and action spaces and limited scalability. While existing MARL methods are structurally closer to real-world systems, they are mostly based on unconstrained Markov Decision Processes (MDPs), where system operational safety constraints are typically indirectly reflected through rewards and penalties, potentially leading to constraint violations under complex or extreme conditions. From a causal perspective, the main reasons for these problems in existing integrated charging, storage, and discharging station scheduling technologies are: 1) EV charging demand and renewable energy output exhibit strong stochasticity, making long-term effectiveness difficult for prediction-based or static model-based scheduling methods; 2) Existing reinforcement learning methods often prioritize profit maximization, lacking systematic modeling of safety constraints; 3) In multi-agent scenarios, coupling relationships exist between subsystems, making simple reward and penalty mechanisms insufficient to guarantee global safety and feasibility.

[0004] In summary, while existing energy management technologies for integrated charging, storage, and discharging EV charging stations have achieved coordinated operation of charging, energy storage, and discharging to a certain extent, they still generally suffer from technical defects such as strong reliance on prediction, simplified modeling, and difficulty in strictly guaranteeing safety constraints when facing multi-source uncertainties and complex safety constraints. Therefore, they cannot effectively meet the safety and economic synergy requirements in complex operating environments. Summary of the Invention

[0005] The purpose of this invention is to provide an energy management system for electric vehicle charging stations, which solves the problem that existing technologies cannot simultaneously maximize the revenue and safety of charging stations under random conditions.

[0006] Another object of the present invention is to provide a scheduling method for an energy management system for electric vehicle charging stations.

[0007] The first technical solution adopted in this invention is: an energy management system for electric vehicle charging stations, including an energy management controller and its communication-connected wind power generation scheduling module, photovoltaic power generation scheduling module, energy storage system scheduling module, and electric vehicle charging control module. The energy management controller is used to divide the operation process of the charging station into multiple discrete control cycles, acquire the operating status of each module in each control cycle, and generate corresponding scheduling instructions to realize the coordinated control of photovoltaic power generation, wind power generation, battery energy storage, grid interaction, and electric vehicle charging. Each module is controlled by a corresponding intelligent agent. Each intelligent agent is constructed based on a constrained Markov decision process, has an independent benefit function and cost function, and is jointly optimized through a multi-agent deep reinforcement learning algorithm.

[0008] The first technical solution of the present invention is further characterized in that,

[0009] The wind power dispatch module's inputs include the actual wind power generation and wind power change trend information at the current time and multiple historical times. The output is the wind power output dispatch value, which is used to determine the amount of electricity transmitted to the station's common bus within the current dispatch cycle. The revenue function of the wind power dispatch module reflects the electricity purchase cost generated by wind power consumption. The cost function includes adjustment penalty terms for exceeding the actual available wind power output and wind curtailment penalty terms for underutilized available wind energy.

[0010] The inputs to the photovoltaic power generation dispatch module include the actual photovoltaic power generation and photovoltaic output trend information at the current time and multiple historical times, and the output is the photovoltaic output dispatch value. The revenue function of the photovoltaic power generation dispatch module reflects the electricity purchase cost generated by photovoltaic consumption. The cost function includes the adjustment penalty term for exceeding the actual available photovoltaic output and the curtailment penalty term for the unutilized available photovoltaic energy.

[0011] The inputs to the energy storage system scheduling module include the grid electricity price sequence and the current state of charge of the energy storage system. The outputs are the interaction power between the energy storage system and the grid, and the charging and discharging power. The revenue function of the energy storage system scheduling module reflects the economic benefits generated by the interaction with the grid, and the cost function includes a penalty term for the state of charge exceeding the preset safety range.

[0012] The inputs to the electric vehicle charging control module include time period information, new energy intensity index, current number of charging vehicles, average state of charge of vehicles, and charging demand. The output is the charging price threshold parameter. The revenue function of the electric vehicle charging control module reflects the charging service revenue, and the cost function includes a penalty for setting the threshold too low during the period of insufficient new energy and a penalty for setting the charging price too high during the period of sufficient new energy.

[0013] The optimization model of the multi-agent constrained Markov decision process for the energy management controller is as follows:

[0014]

[0015] In the formula, This refers to the set of modules participating in the system's coordinated scheduling, including wind power generation scheduling module, photovoltaic power generation scheduling module, electric vehicle charging control module, and energy storage system scheduling module; This indicates the overall system control strategy; This represents the overall system operating benefit function; Indicates the first Each functional module in the control strategy The corresponding cumulative operating revenue is as follows: the revenue of the wind and photovoltaic power generation dispatch module reflects the consumption of new energy, the revenue of the electric vehicle charging control module reflects the charging service revenue and green incentive effect, and the revenue of the energy storage system dispatch module reflects the economic benefits or costs generated by interaction with the public power grid. These are the weighting coefficients for the revenue of each module; Indicates the first Each module in the control strategy The corresponding cumulative operating costs; Indicates the first The maximum cost threshold allowed for each functional module.

[0016] The energy management controller constructs its optimization objective based on the Lagrangian function, transforming the constrained optimization problem into an unconstrained problem with penalties. The Lagrangian function is:

[0017] In the formula, This represents the weighted sum of the revenues from all modules. This indicates penalties for exceeding the security budget; Indicates the maximum allowed limit; Indicates the severity of the punishment. For the first The safety adjustment factor of each module is dynamically adjusted to achieve the constraint strength through adaptive updating; The control policy is updated using a soft Q function:

[0018] In the formula, This is an estimate of the severity of the punishment. For evaluating the network's benefits, For cost evaluation networks.

[0019] The update of the safety adjustment factor satisfies:

[0020] In the formula, Restricted to the range Inside, This represents the upper limit of the Lagrange multiplier; The learning rate for the safety factor automatically increases the penalty intensity when the actual operating cost exceeds the safety threshold, and decreases it when the actual operating cost does not exceed the safety threshold. This indicates the current actual risk.

[0021] The second technical solution adopted in this invention is: a scheduling method for an electric vehicle charging station energy management system. First, the control strategies of each module, the revenue and cost dual evaluation network, and the Lagrange safety factor are initialized. Then, the status of wind power, photovoltaic, energy storage, and EVs is collected. Based on the strategy, a joint scheduling command including wind power, photovoltaic output, charging price, and energy storage power is generated and executed to update the system status. Next, the revenue and cost feedback of each module is obtained and the data is stored in an experience pool. Finally, based on the target return, the mean square error loss function is calculated, and the dual evaluation network is updated through gradient descent to continuously correct the evaluation accuracy of the economic effect and safety impact of scheduling actions.

[0022] The second technical solution of the present invention is further characterized in that, The scheduling method for the energy management system of electric vehicle charging stations includes the following steps: Step 1: Initialize the control strategies, benefit evaluation networks, cost evaluation networks, Lagrange safety factors, and experience pools for each module; Step 2: Collect the current operating status of the wind power generation scheduling module, photovoltaic power generation scheduling module, energy storage system scheduling module, and electric vehicle charging control module as input; Step 3: Based on the current control strategies of each module, sample and generate corresponding scheduling actions, including wind power output, photovoltaic power output, energy storage charging and discharging power, and charging price signals; Step 4: Execute scheduling actions, update system status, and obtain feedback on the benefits and costs of each module; Step 5: Store the input, scheduling actions, revenue, and cost information into the experience pool; Step 6: Update the revenue evaluation network and cost evaluation network based on the data in the experience pool; Step 7: Update the control strategy based on the Lagrange function and adaptively adjust the safety adjustment factor to maximize the overall operating benefits of the system while meeting safety constraints.

[0023] The beneficial effects of this invention are as follows: The electric vehicle charging station energy management system and scheduling method of this invention construct a multi-agent constrained Markov decision process and introduce learnable Lagrange multipliers. By replacing the traditional single-benefit optimization with a "benefit-cost" dual-channel evaluation mechanism, each functional module maximizes economic benefits while dynamically controlling operational risks through adaptive safety adjustment factors. This achieves collaborative optimization scheduling of wind, solar, energy storage, and charging modules under stochastic environments. Simultaneously, by utilizing a price-new energy coupling mechanism, the charging price is dynamically adjusted according to the new energy intensity index, guiding the electric vehicle load to shift to the surplus period of new energy. Under the premise of ensuring strict satisfaction of system safety constraints, this significantly improves the renewable energy absorption capacity and the comprehensive operational efficiency of the charging station, forming a closed-loop learning scheduling framework of "perception-decision-execution-evaluation-update," enabling the system to have long-term adaptive evolution capabilities. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the structure of the electric vehicle charging station energy management system of the present invention; Figure 2 This is a flowchart illustrating the scheduling method of the electric vehicle charging station energy management system of the present invention. Detailed Implementation

[0025] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0026] Example 1 This invention provides an energy management system for electric vehicle charging stations, explicitly constructing a multi-agent (CMDP) system. Each agent possesses a dual-channel reward and cost mechanism, and a learnable Lagrange multiplier is introduced, thereby maximizing the revenue and achieving a dynamic balance of safety for the intelligent charging station. The agents are autonomous decision-making units corresponding to wind power scheduling, photovoltaic power scheduling, energy storage system scheduling, and EV charging control modules. Based on the Constrained Markov Decision Process (CMDP), each agent has an independent policy network (Actor), a revenue evaluation network (double Q), a cost evaluation network (double Q), and a learnable Lagrange safety factor. This allows for maximizing the economic benefits of its own module while adaptively adjusting the strength of safety constraints to achieve collaborative optimization scheduling with other agents.

[0027] like Figure 1As shown, the operation of the integrated charging, storage, and discharging EV charging station is divided into multiple discrete control cycles, each corresponding to a fixed time interval (e.g., 30 minutes), with one day as a scheduling cycle. The energy management controller acquires the operating status of each subsystem within each control cycle and generates corresponding scheduling commands to achieve coordinated control of photovoltaic power generation, wind power generation, battery energy storage, grid interaction, and EV charging. The charging station includes photovoltaic power generation scheduling, wind power generation scheduling, energy storage system scheduling modules, grid interaction module, and EV charging control module. Each subsystem is managed by its corresponding control module. After completing the local modeling of the wind power generation scheduling module, photovoltaic power generation scheduling module, EV scheduling module, and energy storage system scheduling module, the overall operation of the integrated charging, storage, and discharging EV charging station is uniformly constructed as a multi-module collaborative constraint optimization problem.

[0028]

[0029] The meanings of each symbol are as follows: This refers to the set of functional modules participating in the system's coordinated scheduling, including wind power generation modules, photovoltaic power generation modules, EV scheduling modules, and energy storage system modules.

[0030] This represents the overall system control strategy, used to describe the joint decision-making method for generating corresponding control quantities by each functional module in each scheduling cycle, including new energy output scheduling, EV charging prices and threshold parameters, and the power interaction between energy storage system charging / discharging and the grid.

[0031] This represents the overall system operating benefit function, used to characterize the system under a given control strategy. The overall economic benefits obtained by the charging station throughout the entire scheduling cycle.

[0032] Indicates the first Each functional module in the control strategy The corresponding cumulative operating revenue is as follows: the revenue of wind power and photovoltaic modules reflects the consumption of new energy, the revenue of EV modules reflects charging service revenue and green incentive effects, and the revenue of energy storage system modules reflects the economic benefits or costs generated by interaction with the public power grid.

[0033] The weighting coefficients for the revenue of each functional module are used to adjust the relative importance of different modules in the overall system objectives. By setting different weights, a coordinated balance can be achieved between the utilization of new energy, the quality of EV services, and the economical operation of energy storage.

[0034] Indicates the first Each functional module in the control strategy The corresponding cumulative operating cost is used to characterize the degree of breach of physical constraints, security boundaries, and operational stability of the module.

[0035] Indicates the first The maximum allowable cost threshold for each functional module is used as the security budget for that module, which is used to constrain the feasibility of new energy dispatch, the safe range of the state of charge of the battery energy storage system, and the stability of the EV charging process.

[0036] The aforementioned optimization model aims to maximize the overall operational benefits of the system while requiring that the operating costs of each functional module do not exceed their respective preset safety thresholds, thereby establishing a unified coordination mechanism between economy and safety. Through this optimization framework that combines multi-module weighted revenue maximization with independent safety constraints, synergistic optimization among wind and solar power integration, EV charging services, and energy storage system operation is achieved.

[0037] Example 2 This invention provides an energy management system for electric vehicle charging stations. Based on Embodiment 1, the preferred modeling method for the wind power generation dispatch module is as follows: The inputs to the wind power generation dispatch module include: This represents the actual wind power generation at the current moment and multiple historical moments, as well as wind power trend information extracted from historical power sequences. These parameters characterize the current availability of wind energy resources and their short-term variation characteristics. The output is: ,in This represents the wind power output dispatch value output by the wind power dispatch module, used to determine the amount of electrical energy transmitted from the wind power dispatch system to the station's common bus within the current dispatch cycle. This output is subject to the rated capacity of the wind power system. And the constraint of currently available wind power. Its payoff function is... ,in This indicates the feed-in tariff for renewable energy. This represents the amount of wind power that can be dispatched at that moment. This term reflects the operating cost incurred by the charging station when purchasing wind power from the wind turbine generator unit, and is therefore represented by a negative value. The cost function is: Among them, symbols } represents the positive part operator, which takes the non-negative part of the variable; These are used to represent the constraint strength corresponding to the adjustment behavior exceeding the actual available wind power output and the situation where available wind energy is not fully utilized, respectively. By applying penalties with different weights to these two types of situations, wind power dispatch is ensured to always meet the physical feasibility conditions and reduce the wind curtailment rate, thereby reducing the waste of renewable energy. This represents the energy actually dispatched by the wind power generation dispatch module at that moment. The wind power generation module primarily aims to improve wind energy utilization, and its revenue is related to the actual wind power energy consumed. Specifically, charging stations purchase electricity from the wind power generation system at a fixed renewable energy settlement price; therefore, the revenue of the wind power generation module reflects the electricity purchase cost corresponding to wind power consumption.

[0038] Example 3 This invention provides an energy management system for electric vehicle charging stations. Based on Embodiment 1, the preferred modeling method for the photovoltaic power generation scheduling module is as follows: The inputs to the photovoltaic power generation dispatch module include: This represents the actual photovoltaic (PV) power generation at the current moment and at multiple historical moments; as well as the PV power output trend information extracted from historical power sequences. These parameters characterize the current availability of solar energy resources and their short-term variation characteristics. The output is: This represents the photovoltaic (PV) output dispatch value output by the PV power generation dispatch module, used to determine the amount of electrical energy transmitted from the PV power generation system to the station's common bus within the current dispatch cycle. This output is subject to the rated capacity of the PV system. And the constraint of currently available photovoltaic power. Its payoff function is... ,in This indicates the feed-in tariff for renewable energy. This represents the amount of photovoltaic power generation that can be dispatched at that moment. This term reflects the operating cost incurred by the charging station when purchasing photovoltaic power from the photovoltaic power generation unit, and is therefore represented as a negative value. Its cost function is: .in, , These are used to represent the constraint strength corresponding to the adjustment of dispatch behavior exceeding the actual available photovoltaic output and the underutilization of available photovoltaic energy, respectively. By applying penalties with different weights to these two types of situations, the photovoltaic power generation dispatch is ensured to always meet the physical feasibility conditions, and the curtailment rate is reduced, thereby reducing the waste of renewable energy. This represents the energy actually dispatched by the photovoltaic (PV) power generation dispatch module at that moment. The PV power generation module primarily aims to improve solar energy utilization, and its revenue is related to the actual PV energy consumed. Specifically, charging stations purchase electricity from the PV power generation system at a fixed renewable energy settlement price; therefore, the revenue of the PV power generation module reflects the electricity purchase cost corresponding to PV energy consumption.

[0039] Example 4 This invention provides an energy management system for electric vehicle charging stations. Based on Embodiment 1, the preferred modeling method for the energy storage system scheduling module is as follows: The inputs to the energy storage system scheduling module include: ,in This indicates the current state of charge of the battery energy storage system; This represents the electricity price information for the current period and multiple historical scheduling cycles. The above input parameters collectively characterize the adjustability of the charging station's energy storage system and the trend of electricity price changes within the current cycle. The output is: ,in The charging and discharging power of the battery energy storage system during the current dispatch cycle, and implicitly includes the power interaction with the public power grid. These represent the upper and lower limits of the grid interface power, respectively. The output is a continuous value and is subject to the upper and lower limits of the grid interface power, where a positive value indicates electricity purchased from the public grid, and a negative value indicates electricity sold to the public grid. When the renewable energy output exceeds the EV charging load, the energy storage system prioritizes charging operations to absorb the surplus renewable energy; when renewable energy is insufficient to meet charging demand, the energy storage system performs discharging operations and compensates for the power shortfall through the public grid interface, thereby maintaining the overall energy balance of the charging station. Its revenue function is... This revenue item is jointly determined by the current electricity price and the power purchased and sold by the energy storage system via the grid interface. It reflects the economic benefits or costs incurred by the charging station participating in grid energy trading during the current dispatch cycle. In this embodiment, the electricity purchase price and the electricity sales price are set to be the same. Its cost function is:

[0040] in, and These represent the penalty weights for when the state of charge of the energy storage system is below the lower safety limit and above the upper safety limit, respectively. This represents the energy storage capacity and is used to standardize the penalty range. By imposing penalties on the above two types of states, a two-way safety constraint mechanism is constructed: on the one hand, it constrains the energy storage system to avoid overcharging when there is a surplus of new energy to avoid the risk of overcharging; on the other hand, it constrains the energy storage system to maintain necessary energy reserves when there is a shortage of new energy to avoid over-discharge, thereby ensuring the continuous satisfaction of EV charging needs and improving the overall safety and reliability of the charging station operation.

[0041] Example 5 This invention provides an energy management system for electric vehicle charging stations. Based on Embodiment 1, the preferred modeling method for the EV charging control module is as follows: The inputs to the EV charging control module include: ,in, Indicates the current scheduling time; This represents the normalized renewable energy intensity index; This indicates the number of EVs currently connected to the charging station; This represents the average state of charge of all vehicles at the station; and This represents the aggregated charging demand within the current scheduling cycle. The above parameters are used to comprehensively characterize the vehicle's operating status and the level of renewable energy supply. The output is: ,in This indicates the charging price signal released to EV users, and its value is constrained by preset minimum and maximum charging prices. This represents the threshold parameter used to switch between green-priority mode and revenue-priority mode, with a value range of [0,1]. The charging price signal is input to each EV terminal as an external control variable. Each vehicle autonomously determines its charging power based on its remaining dwell time and battery status, guided by the price signal, thus forming a station-level aggregated charging load. Its revenue function is... ,in This indicates whether the process has entered the green phase. This represents the electricity price during the green phase. The revenue structure employs a dual-branch mechanism, including a revenue-priority branch and a green-priority branch, switched by the relationship between a threshold parameter and a renewable energy intensity index: when the renewable energy supply level is below the threshold parameter, the system enters the revenue-priority mode, increasing charging prices to suppress load growth and boost direct charging revenue; when the renewable energy supply level is above the threshold parameter, the system enters the green-priority mode, lowering charging prices to incentivize vehicles to increase charging power, thereby promoting renewable energy consumption. In the green-priority mode, a green value signal is constructed based on a reverse mapping of charging prices, causing green revenue to increase as user charging prices decrease. Simultaneously, the threshold parameter is used to adjust the magnitude of green revenue, thus enabling the threshold parameter to simultaneously perform branch switching and gain adjustment functions. Through this design, the system proactively lowers prices to absorb surplus renewable energy when renewable energy is abundant, and automatically reverts to the revenue-priority mode when renewable energy is insufficient, thereby achieving a dynamic trade-off between economic benefits and renewable energy utilization. Its cost function is:

[0042] in, and These are used to adjust the penalty intensity for setting the threshold too low during the period of insufficient new energy, and the penalty intensity for setting the charging price too high during the period of sufficient new energy. This represents the minimum charging price for electric vehicles. The first penalty term constrains prematurely entering a low-price mode when new energy sources are scarce, thus preventing excessive growth in charging demand from impacting system stability. The second penalty term constrains maintaining excessively high prices when new energy sources are abundant, thus avoiding inhibiting new energy absorption. Through this dual constraint mechanism, the EV scheduling module maintains conservative pricing to ensure system operational safety when new energy sources are insufficient, and proactively lowers prices to improve new energy absorption efficiency when new energy sources are abundant, thereby achieving a balance between safety and economy in the EV charging process.

[0043] Example 6 This invention also provides a method for generating joint scheduling instructions in an electric vehicle charging station energy management system, such as... Figure 2 As shown, it includes the following steps: Step 1: Initialization Phase (Model and Safety Factor Initialization) Control Strategy (Actor): ,in Indicates the first i Network parameters for each controller, revenue evaluation network (double Q): Cost Evaluation Network (Double Q): Lagrange safety factor: Experience Pool: Initialize the strategies, revenue evaluators, cost evaluators, safety adjustment factors, and historical caches for each module.

[0044] Step 2: Collect input from each module: Wind power dispatch module input Photovoltaic dispatch module input Energy storage system scheduling module input and EV scheduling module input .

[0045] Step 3: Generate joint scheduling actions. Outputs from each module: 1. Wind power: 2. Photovoltaics: 3. EV: 4. Energy storage: Sampling method: ,in, This indicates that i The output of a controller at time t.

[0046] Step 4: Perform scheduling and system input updates Energy Storage SOC Update:

[0047] Step 5: Obtain Revenue and Cost Feedback Revenue of each module: wind: ,Light: Energy storage: and EV: Cost of each module: Photovoltaic: Wind power Energy storage: EV: .

[0048] Step Six: Store the inputs, outputs, and corresponding revenue and cost information of each module in the experience pool. middle.

[0049] Step 7: Update the benefit / cost evaluation network (Critic), target value: Loss function:

[0050]

[0051] in, Indicates the first Each module at time Target return on investment; Indicates the first Each module at time The target cost-return. It means that when the first When the module receives input and outputs a, the first... How much revenue can the module generate in the future? This indicates the security cost incurred for the same output. This loss function continuously refines the model, making it increasingly accurate in assessing the economic benefits and security impacts of scheduling actions.

[0052] Step 8: Update control policies and security factors Lagrange function:

[0053] in, This represents the weighted sum of the revenues from all modules. This indicates penalties for exceeding the security budget, specifically... Current real risks : Maximum allowable limit Severity of punishment.

[0054] Soft Q:

[0055] in, This function is used to estimate the penalty intensity and is used to optimize the scheduling strategy based on the "risk-discounted payoff".

[0056] Strategy Update:

[0057] in, Indicates the first New control strategies for functional modules (such as EV scheduling module and energy storage module); Indicates the first The set of all possible outputs of each module; This represents the KL divergence, used to measure the "distance" between two probability distributions; This represents entropy temperature. This represents the partition function.

[0058] Safety factor update:

[0059] in, This represents the upper limit of the Lagrange multiplier. Indicates the safety factor learning rate. This is the projection operator.

[0060] Steps seven through nine use a dual benefit / cost evaluation model to assess the economic efficiency and safety of scheduling actions, and employ safety factors to adaptively penalize over-limit risks, thereby maximizing the overall benefits of the system while ensuring that the operation of each functional module always meets safety constraints.

[0061] Through the above methods, the electric vehicle charging station energy management system and scheduling method of the present invention exhibit significant advantages in practical applications, mainly reflected in the following aspects: 1) Achieving coordinated and optimized scheduling of multiple modules including wind, solar, energy storage, and charging. This invention constructs a multi-functional coordinated energy management architecture comprising wind power generation modules, photovoltaic power generation modules, EV scheduling modules, and energy storage system modules. A unified scheduling mechanism is used to jointly optimize the operating status and control variables of each module. Unlike traditional methods that separately manage renewable energy generation, energy storage control, and charging load, this invention introduces a unified control framework at the system level, incorporating renewable energy output, EV charging prices, and energy storage charging and discharging power into the same optimization model. A weighted comprehensive benefit function is used to coordinate the objectives of multiple modules. This avoids the local optima problem caused by the independent operation of each subsystem, achieving a coordinated balance between wind and solar energy consumption, EV charging services, and the economical operation of energy storage, significantly improving the overall operating efficiency and economic benefits of integrated charging, storage, and discharging stations.

[0062] 2) Introducing a dual benefit-cost mechanism to achieve economic optimization under safety constraints. This invention constructs a benefit function and a cost function for each functional module and sets independent safety thresholds. An adaptive penalty for exceeding these thresholds is applied through a safety adjustment factor. Traditional scheduling methods often focus solely on economic benefits, making it difficult to guarantee the state of charge of energy storage, the stability of EV charging, and the feasibility of renewable energy output. This invention, however, achieves a unified expression of multi-dimensional safety constraints by setting curtailment and over-scheduling costs for wind and solar modules, state of charge safety costs for energy storage modules, and price stability costs for EV modules. This effectively prevents overcharging and over-discharging of energy storage, suppresses renewable energy curtailment, and avoids sudden increases in EV load without sacrificing economic performance, thus improving system operational safety and reliability.

[0063] 3) An adaptive safety adjustment mechanism is adopted to dynamically adjust the constraint strength. This invention introduces a safety adjustment factor to provide real-time feedback adjustment on the operational risk of each module. When a constraint is triggered, the penalty intensity is automatically increased; when the system returns to a safe range, the constraint is automatically released. Unlike fixed threshold control methods, this invention dynamically updates the safety adjustment factor based on operational feedback, enabling the system to automatically adjust the conservatism of the scheduling strategy according to the actual risk level. This avoids the scheduling rigidity problem caused by traditional hard constraints, allowing the system to more actively absorb renewable energy when there is a surplus and automatically become more conservative when renewable energy is insufficient, achieving a dynamic balance between economy and safety.

[0064] 4) Enhancing Renewable Energy Absorption Capacity through a Price-Renewable Energy Coupling Mechanism. This invention introduces a linkage mechanism between renewable energy intensity indicators and charging prices into the EV scheduling module, and sets up dual operating modes: green priority and revenue priority. When renewable energy is abundant, the system automatically lowers charging prices to encourage vehicles to increase charging load; when renewable energy is insufficient, charging prices are increased to suppress demand growth. This mechanism achieves mode switching through threshold parameters and amplifies the incentive effect of renewable energy through inverse price mapping. This effectively guides EV load to shift to periods of surplus renewable energy, significantly reduces the proportion of wind and solar curtailment, improves renewable energy utilization, and simultaneously ensures stable charging station revenue.

[0065] 5) Constructing a closed-loop learning-based scheduling framework to achieve continuous improvement in long-term operational performance. This invention adopts a closed-loop optimization structure based on historical operational data, continuously updating the control strategies and evaluation models of each module to achieve online learning and adaptive optimization. By collecting system operating status, executing scheduling decisions, obtaining benefit feedback, and updating control parameters, a closed-loop process of "perception—decision—execution—evaluation—update" is formed, enabling the scheduling strategy to continuously evolve with environmental changes. Compared with traditional scheduling methods based on static models or offline parameters, this invention can adapt to uncertainties such as new energy fluctuations and random EV access, gradually approaching the optimal scheduling state during long-term operation, thus improving system robustness and adaptability.

Claims

1. An energy management system for electric vehicle charging stations, characterized in that, The system includes an energy management controller and its communication connections to a wind power generation scheduling module, a photovoltaic power generation scheduling module, an energy storage system scheduling module, and an electric vehicle charging control module. The energy management controller is used to divide the operation of the charging station into multiple discrete control cycles. In each control cycle, it acquires the operating status of each module and generates corresponding scheduling instructions to achieve coordinated control of photovoltaic power generation, wind power generation, battery energy storage, grid interaction, and electric vehicle charging. Each module is controlled by a corresponding intelligent agent. Each intelligent agent is constructed based on a constrained Markov decision process and has an independent benefit function and cost function. The agents are jointly optimized through a multi-agent deep reinforcement learning algorithm.

2. The energy management system for electric vehicle charging stations as described in claim 1, characterized in that, The wind power generation dispatch module's inputs include the actual wind power generation at the current moment and multiple historical moments, as well as wind power change trend information. Its output is the wind power output dispatch value, which is used to determine the amount of electrical energy transmitted to the station's common bus within the current dispatch cycle. The revenue function of the wind power dispatch module reflects the electricity purchase cost generated by wind power consumption. The cost function includes a regulation penalty term for exceeding the actual available wind power output and a wind curtailment penalty term for underutilization of available wind energy.

3. The energy management system for electric vehicle charging stations as described in claim 1, characterized in that, The inputs of the photovoltaic power generation scheduling module include the actual photovoltaic power generation and photovoltaic output change trend information at the current time and multiple historical times, and the output is the photovoltaic output scheduling value. The revenue function of the photovoltaic power generation scheduling module reflects the electricity purchase cost generated by photovoltaic consumption. The cost function includes the adjustment penalty term for exceeding the actual available photovoltaic output and the curtailment penalty term for the unutilized available photovoltaic energy.

4. The energy management system for electric vehicle charging stations as described in claim 1, characterized in that, The inputs of the energy storage system scheduling module include the grid electricity price sequence and the current state of charge of the energy storage system, and the outputs are the interaction power between the energy storage system and the grid and the charging and discharging power. The revenue function of the energy storage system scheduling module reflects the economic benefits generated by the interaction with the grid, and the cost function includes a penalty term for the state of charge exceeding the preset safety range.

5. The energy management system for electric vehicle charging stations as described in claim 1, characterized in that, The inputs to the electric vehicle charging control module include time period information, new energy intensity index, current number of charging vehicles, average state of charge of vehicles, and charging demand. The output is a charging price threshold parameter. The revenue function of the electric vehicle charging control module reflects the charging service revenue, and the cost function includes a penalty for setting the threshold too low during the period of insufficient new energy and a penalty for setting the charging price too high during the period of sufficient new energy.

6. The energy management system for electric vehicle charging stations as described in any one of claims 1 to 5, characterized in that, The optimization model of the multi-agent constrained Markov decision process constructed by the energy management controller is as follows: In the formula, This refers to the set of modules participating in the system's coordinated scheduling, including wind power generation scheduling module, photovoltaic power generation scheduling module, electric vehicle charging control module, and energy storage system scheduling module; This indicates the overall system control strategy; This represents the overall system operating benefit function; Indicates the first Each functional module in the control strategy The corresponding cumulative operating revenue is as follows: the revenue of the wind and photovoltaic power generation dispatch module reflects the consumption of new energy, the revenue of the electric vehicle charging control module reflects the charging service revenue and green incentive effect, and the revenue of the energy storage system dispatch module reflects the economic benefits or costs generated by interaction with the public power grid. These are the weighting coefficients for the revenue of each module; Indicates the first Each module in the control strategy The corresponding cumulative operating costs; Indicates the first The maximum cost threshold allowed for each functional module.

7. The energy management system for electric vehicle charging stations as described in claim 6, characterized in that, The energy management controller constructs an optimization objective based on a Lagrangian function, transforming the constrained optimization problem into an unconstrained problem with penalties. The Lagrangian function is: In the formula, This represents the weighted sum of the revenues from all modules. This indicates penalties for exceeding the security budget; Indicates the maximum allowed limit; Indicates the severity of the punishment. For the first The safety adjustment factor of each module is dynamically adjusted to achieve the constraint strength through adaptive updating; The control policy is updated using a soft Q function: In the formula, This is an estimate of the severity of the punishment. For evaluating the network's benefits, For cost evaluation networks.

8. The energy management system for electric vehicle charging stations as described in claim 7, characterized in that, The update of the security adjustment factor satisfies: In the formula, Restricted to the range Inside, This represents the upper limit of the Lagrange multiplier; The learning rate for the safety factor automatically increases the penalty intensity when the actual operating cost exceeds the safety threshold, and decreases it when the actual operating cost does not exceed the safety threshold. This indicates the current actual risk.

9. The scheduling method of the energy management system for electric vehicle charging stations as described in claim 1, characterized in that, First, initialize the control strategies for each module, the dual evaluation network for revenue and cost, and the Lagrange safety factor. Then, collect the status of wind power, photovoltaic, energy storage, and EVs, generate joint dispatch instructions based on the strategies that include wind power, photovoltaic output, charging price, and energy storage power, and execute them to update the system status. Next, obtain the revenue and cost feedback of each module and store the data in the experience pool. Finally, calculate the mean squared error loss function based on the target return, and update the dual evaluation network through gradient descent to continuously correct the evaluation accuracy of the economic effect and safety impact of dispatch actions.

10. The scheduling method of the energy management system for electric vehicle charging stations as described in claim 9, characterized in that, Includes the following steps: Step 1: Initialize the control strategies, benefit evaluation networks, cost evaluation networks, Lagrange safety factors, and experience pools for each module; Step 2: Collect the current operating status of the wind power generation scheduling module, photovoltaic power generation scheduling module, energy storage system scheduling module, and electric vehicle charging control module as input; Step 3: Based on the current control strategies of each module, sample and generate corresponding scheduling actions, including wind power output, photovoltaic power output, energy storage charging and discharging power, and charging price signals; Step 4: Execute scheduling actions, update system status, and obtain feedback on the benefits and costs of each module; Step 5: Store the input, scheduling actions, revenue, and cost information into the experience pool; Step 6: Update the revenue evaluation network and cost evaluation network based on the data in the experience pool; Step 7: Update the control strategy based on the Lagrange function and adaptively adjust the safety adjustment factor to maximize the overall operating benefits of the system while meeting safety constraints.

Images