Cloud-side collaborative energy storage power station dynamic control method

Abstract

The invention relates to the technical field of electric energy storage systems, and discloses a cloud-side collaborative energy storage power station dynamic control method, which comprises the following steps: a cloud server obtains an instruction of an energy storage power station to generate required input information, and an edge controller obtains real-time state information of the energy storage power station; based on the input information and the real-time state information, a charge-discharge instruction sequence of a future period is generated through a dynamic planning model, and the dynamic planning model constructs a joint state space containing the residual electric quantity, the residual charge-discharge cycle number and the battery health degree and performs inverse recursive calculation by taking the operation income as a target function, so that the charge-discharge instruction sequence of the future period is obtained. Finally, an optimal charging and discharging instruction sequence is extracted in the forward direction; and issuing the optimal charging and discharging instruction sequence to an edge controller, and controlling an energy storage power station to operate. The problems of low instruction maintenance efficiency, response lag, insufficient income optimization and dependence on artificial experience in the prior art are solved, and the purposes of automatic instruction generation, real-time response, income maximization and labor cost reduction are achieved.

Description

Technical Field

[0001] This invention relates to the field of energy storage system technology, and in particular to a dynamic control method for a cloud-edge collaborative energy storage power station. Background Technology

[0002] The core challenge currently facing the commercial and industrial energy storage sector lies in the frequent adjustments to local policies. Various regions are continuously optimizing their peak-valley electricity pricing mechanisms, with frequent changes not only in time-of-use settings but also in the dynamic adjustment of electricity prices themselves. These changes bring enormous workload and technical difficulty to the maintenance of energy storage power station operation strategies. Under traditional operation and maintenance models, staff must manually configure charging and discharging strategies for different time periods based on the specific load characteristics of each power station. This method is inefficient, significantly increases labor costs, and heavily relies on manual experience to calculate charging and discharging power to approximate the optimal profit solution, resulting in response lag and insufficient optimization. Industry practice indicates that insufficient optimization, especially the failure to effectively utilize the latest peak / valley periods, of static strategies may lead to potential revenue losses. Furthermore, existing technologies mainly rely on the power station's local EMS (Energy Management System) to execute fixed strategies, lacking effective linkage with broader information sources. Strategy adjustments require manual intervention, making it difficult to respond to policy changes in real time and unable to achieve flexible scheduling of massive cloud data and computing power to support global optimization and automated decision-making, thus restricting the economic benefits and operational efficiency of energy storage power stations.

[0003] For example, Chinese patent CN119520601A discloses a method for updating control strategies of energy storage power stations based on cloud-edge collaborative technology. It provides the following technical solutions: when the edge computing gateway collects data or receives alarm information or task processing requests actively sent by the energy storage power station, it starts calling a data processing program; the edge computing gateway data processing program determines the type of task currently being executed. If it is an alarm message, it proceeds to an alarm message analysis and processing program; if it is a computation task actively requested by the energy storage power station, it performs a computation task analysis and processing program; if it is a high-complexity computation task, it requests assistance from the cloud server for computation. This patent designs a bidirectional update method for control strategies, allowing both the edge control gateway and the cloud server to proactively initiate strategy updates. Compared to previous unidirectional strategy update methods that primarily rely on the edge or cloud side, this patent method has greater applicability. However, the aforementioned method for updating control strategies for energy storage power stations based on cloud-edge collaboration technology cannot achieve dynamic strategy optimization for peak-valley arbitrage, lacks a real-time response mechanism to changes in electricity price policies, has a simple battery health model, and does not integrate a profit maximization algorithm, resulting in poor strategy flexibility, limited economic benefit optimization, and difficulty in adapting to the complex needs of industrial and commercial energy storage. Summary of the Invention

[0004] This invention addresses the problems of low strategy maintenance efficiency, delayed response, insufficient benefit optimization, and reliance on human experience in existing technologies. It proposes a cloud-edge collaborative dynamic control method for energy storage power stations, achieving the goals of automated command generation, real-time response, maximized benefits, and reduced labor costs.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A dynamic control method for a cloud-edge collaborative energy storage power station includes: The cloud server obtains the input information required for the strategy generation of the energy storage power station, and the edge controller obtains the real-time status information of the energy storage power station. Based on the input information and real-time status information, a charging and discharging instruction sequence for future cycles is generated through a dynamic programming model. The dynamic programming model constructs a joint state space that includes the remaining power, the remaining number of charging and discharging cycles, and the battery health, and performs reverse recursive calculation with operating revenue as the objective function, and finally extracts the optimal charging and discharging instruction sequence in the forward direction. The optimal charge and discharge command sequence is sent to the edge controller to control the operation of the energy storage power station.

[0006] It enables automated command generation for cloud-edge collaboration, significantly reducing manual maintenance workload and enabling real-time response to changes in electricity pricing policies, thereby improving the operational efficiency and economic benefits of energy storage power stations.

[0007] Preferably, the dynamic programming model specifically includes: dividing the future period into n equally spaced time periods; defining state variables as triplets, wherein the triplets include: the remaining energy storage capacity at the beginning of the time period, the value of which ranges from 0 to the rated capacity of the energy storage system, and is discretized during calculation, with the discretization step size being half of the full-power charge and discharge energy of the energy storage within the minimum time interval; the remaining number of charge and discharge cycles before the start of the time period, the value of which ranges from 0 to the upper limit of the number of charge and discharge cycles allowed per day; and the limitation of the battery health status on the charge and discharge capacity before the start of the time period, the value of which is selected from a preset discrete set.

[0008] The critical states of the energy storage system are accurately modeled to ensure the feasibility and optimality of strategy calculations under physical constraints. Discretization reduces computational complexity while improving accuracy, enabling instructions to effectively adapt to battery life and operational needs, and avoiding loss of benefits or safety hazards due to model simplification.

[0009] Preferably, the generation of the charging and discharging instruction sequence for future cycles through the dynamic programming model specifically includes: in each time period, the allowed actions include charging, discharging, and standby; when the action is charging, the revenue is calculated as the product of the negative charging power, the time period length, and the electricity market transaction rate for that time period, and must meet the following constraints: the power after charging does not exceed the capacity, and the power constraint: the sum of the charging power and the predicted load for that time period does not exceed the smaller value between the transformer capacity and the maximum demand.

[0010] The charging action takes into account real-time electricity prices and load to avoid additional electricity cost losses caused by overcapacity operation, supports maximum revenue, and complies with power station hardware limitations to improve the practicality and reliability of commands.

[0011] Preferably, the generation of the charging and discharging instruction sequence for future cycles through the dynamic programming model further includes: when the action is discharging, the revenue is calculated as the product of the discharge power, the duration, and the electricity market transaction rate for that duration, and must meet the following constraints: the discharge power is not lower than the safety threshold considering the discharge depth and health, and the power constraint is that the discharge power does not exceed the smaller of the system rated power and the current load; when the action is standby, the revenue is zero and there is no power constraint.

[0012] The calculation of discharge benefits and the design of constraints ensure battery safety and operational efficiency. By dynamically adjusting the depth of discharge in conjunction with battery health, battery life is extended. Standby options avoid unnecessary charge-discharge switching, reduce equipment wear and tear, and make commands more aligned with actual operating scenarios.

[0013] Preferably, the generation of the charging and discharging command sequence for future cycles through the dynamic programming model further includes establishing state transition equations, specifically including: the power transition equation is defined as follows: if charging, the power at the end of the period is the power at the beginning of the period plus the charging energy considering charging efficiency; if discharging, the power at the end of the period is the power at the beginning of the period minus the discharging energy considering discharging efficiency; if standby, the power remains unchanged; the cycle number transition equation is defined as follows: if the action in the current period is different from the action in the previous period and involves a switching between charging and discharging states, the remaining cycle number is reduced by one; otherwise, it remains unchanged; the state transition of health is calculated based on the cumulative equivalent full cycle number through a decay model function.

[0014] The state transition equation accurately simulates the dynamic behavior of the battery, while the energy equation considers efficiency loss to improve the realism of the instructions. The cycle number equation strictly constrains the daily switching frequency to protect battery life. The health decay model incorporates long-term life factors into the optimization, so that the instructions do not only pursue short-term gains, but also comprehensively consider economic benefits and equipment durability.

[0015] Preferably, the reverse recursive calculation specifically includes: starting from the last period of the cycle and working backwards to the first period, initializing the final state value function to zero, and setting a final state power safety constraint; for each possible state in each period, calculating the sum of the immediate benefit after executing each allowed action and the optimal value function of the next period, taking the maximum value as the optimal value function of that state, and constructing a complete state-value mapping table.

[0016] The reverse recursive method efficiently solves for the global optimum by covering all possible states through reverse computation, avoiding local optima traps. A value mapping table is constructed to provide a foundation for forward decision-making, ensuring mathematical rigor and computational efficiency of instructions, reducing computational overhead, and making it suitable for cloud-edge collaborative environments.

[0017] Preferably, the forward extraction of the optimal charge and discharge command sequence specifically includes: starting from a known initial state, determining the optimal charge and discharge actions for each time period in chronological order based on the state-value mapping table generated in the reverse recursive stage; ensuring that the action sequence simultaneously satisfies the power transfer equation and the number of charge and discharge cycles by tracking state transitions in real time, and finally outputting the complete charge and discharge command sequence for future cycles.

[0018] The forward extraction process transforms theoretical calculations into executable instructions, while real-time state tracking ensures the continuity of action sequences and compliance with constraints. It enables smooth instruction execution, reduces runtime errors, improves system reliability and automation, and allows energy storage power stations to adapt to changing conditions.

[0019] Preferably, the edge controller periodically uploads the actual power and actual load of the energy storage system to the cloud server during the execution of the instruction sequence; the cloud server recalculates and optimizes the instructions for the remaining unexecuted periods based on the latest actual data, correcting deviations caused by prediction errors; when a significant change in the load prediction value or a fluctuation in the electricity market transaction rate is detected, the system immediately triggers the instruction recalculation process to dynamically adjust subsequent charging and discharging instructions.

[0020] The rolling optimization mechanism enhances the stability of instructions by correcting prediction errors through periodic feedback and recalculation, ensuring that instructions are always close to the optimal state. Event-driven recalculation enables real-time response to policy or load changes, improving system adaptability and avoiding revenue loss.

[0021] Preferably, the required input information includes the attribute parameters, user constraint parameters, and dynamic input parameters of the energy storage power station; the attribute parameters include at least the geographical location information, rated installed power, and rated capacity of the power station.

[0022] Comprehensive input parameter design enables personalized instructions, linking geographical location to electricity pricing policies, and defining physical boundaries based on installed capacity and power output, ensuring that instructions are generated based on the specific characteristics of the power plant. This improves the accuracy and applicability of instructions and reduces the limitations of a one-size-fits-all approach.

[0023] Preferably, the user constraint parameters include at least the electricity consumption type classification and billing method, wherein the billing method is based on transformer capacity or maximum demand; the dynamic input parameters include at least the remaining power of the energy storage system at the time of model triggering, the upper limit of the daily allowable charge-discharge cycle count, the battery charge-discharge depth safety constraint, and the historical cumulative charge-discharge cycle count.

[0024] User constraints and dynamic parameters refine the instruction conditions; electricity usage characteristics and billing methods influence electricity price selection; and cycle count and health constraints protect battery life. These parameters make the dynamic programming model closer to actual operational needs, supporting maximum revenue while also considering safety and compliance.

[0025] Compared with the prior art, the beneficial effects of the present invention are as follows.

[0026] 1. This invention adopts a cloud-edge collaborative architecture, with the cloud responsible for policy monitoring and instruction calculation, and the edge automatically executing the instructions. This achieves full automation of instruction maintenance, minimizes human intervention, significantly improves operational efficiency and response speed, and reduces reliance on manual intervention.

[0027] 2. This invention uses a dynamic programming algorithm to accurately calculate charge and discharge commands, taking into account factors such as power station attributes, user constraints, and battery health, to ensure maximum profitability within lifespan constraints. The system has real-time response capabilities, automatically parsing new policies in the cloud and immediately issuing updated commands, enabling power stations to quickly adapt to changes, maximize arbitrage opportunities, and improve economic efficiency.

[0028] 3. This invention quantifies energy storage status into a multi-dimensional state space, breaking through the traditional single-capacity model and accurately matching the cycle life limitations of industrial and commercial scenarios. By introducing a battery health degradation model, lifespan degradation is correlated with operational commands, achieving long-term comprehensive benefit optimization. Furthermore, a rolling optimization mechanism is designed to periodically correct commands based on actual data, addressing prediction errors, ensuring the reliability and adaptability of commands, and improving overall operational robustness. Attached Figure Description

[0029] Figure 1 This is an overall flowchart of a cloud-edge collaborative dynamic control method for energy storage power stations according to the present invention. Detailed Implementation

[0030] See Figure 1 As shown, a dynamic control method for a cloud-edge collaborative energy storage power station includes: The cloud server obtains instructions from the energy storage power station to generate the required input information, and the edge controller obtains the real-time status information of the energy storage power station. Based on the input information and real-time status information, a sequence of charge and discharge instructions for future cycles is generated through a dynamic programming model. The dynamic programming model constructs a joint state space that includes the remaining power, the remaining number of charge and discharge cycles, and the battery health, and performs reverse recursive calculation with operating revenue as the objective function, and finally extracts the optimal instruction in the forward direction. The charging and discharging command sequence is sent to the edge controller to control the operation of the energy storage power station.

[0031] This patent proposes a cloud-edge collaborative dynamic instruction generation system. By centrally managing electricity pricing policies in the cloud and executing optimized instructions in real time at the edge, it achieves fully automated and high-precision maximization of peak-valley arbitrage profits for energy storage power stations.

[0032] like Figure 1 In one embodiment shown, Figure 1 This is an overall flowchart of a cloud-edge collaborative dynamic control method for energy storage power stations according to the present invention. The present invention designs a cloud-edge collaborative dynamic control method for energy storage power stations. First, the cloud server is responsible for acquiring various input information required for generating commands from the energy storage power station, while the edge controller collects real-time operating status data of the energy storage power station. Then, based on this input information and real-time status, the system generates a sequence of charge and discharge commands for a complete future cycle through a dynamic programming model. The core of this model lies in constructing a joint state space, which includes three key state variables: remaining battery capacity, remaining charge and discharge cycle count, and battery health. The optimal command is solved using a reverse recursive calculation method with the objective function of maximizing operational revenue. Finally, the generated charge and discharge command sequence is sent to the edge controller to control the actual operation of the energy storage power station.

[0033] Specifically, the process of this invention begins with the initialization of a dynamic programming model. First, the future period (e.g., a day) is evenly divided into n equally spaced time periods, the length of which can be set according to actual needs, for example, 5 to 15 minutes. Then, the state variables at the beginning of each time period are defined as a triplet structure. The remaining energy storage capacity ranges from 0 to the system's rated capacity and is discretized during the calculation. The discretization step size is typically half of the full-power charge / discharge energy within the minimum time interval to balance calculation accuracy and efficiency. The remaining charge / discharge cycle count ranges from 0 to the maximum allowed charge / discharge cycle count per day, which is usually set to a smaller value (e.g., 1 or 2 times) based on battery life constraints. The battery health status is used to limit charge / discharge capacity, and its value is selected from a preset discrete set, for example, 0.8, 0.9, or 1.0, corresponding to different capacity decay levels.

[0034] After defining the state variables, the next step is the action decision and revenue calculation phase of the model. In each time period, the system allows three actions: charging, discharging, or standby. When charging is selected, the revenue calculation is negative, specifically the inverse of the product of charging power, time period length, and the electricity market transaction rate for that time period. Two constraints must be met: first, the amount of electricity generated after charging must not exceed the rated capacity of the energy storage system; second, the sum of the charging power and the predicted load for that time period must not exceed the smaller of the transformer capacity and the maximum demand, to avoid a surge in demand-based electricity charges. When discharging is selected, the revenue is calculated as the product of discharging power, time period length, and the rate, and must meet the following constraints: the amount of electricity generated after discharging must not be lower than a safe threshold considering the depth of discharge and battery health; and the power constraint, the discharging power must not exceed the smaller of the system's rated power and the current load. If standby is selected, the revenue is zero, and no power limit needs to be considered.

[0035] Then, the system establishes state transition equations to describe the dynamic changes of state variables. The energy transition equation is defined as follows: if charging is performed, the energy at the end of the period equals the energy at the beginning of the period plus the charging energy considering charging efficiency; if discharging is performed, the energy at the end of the period equals the energy at the beginning of the period minus the discharging energy considering discharging efficiency; if in standby mode, the energy remains unchanged. The cycle count transition equation is defined as follows: if the action in the current period is different from the action in the previous period and involves a switch between charging and discharging states (such as switching from charging to discharging), the remaining cycle count is decreased by one; otherwise, the remaining cycle count remains unchanged. The state transition for battery health is calculated based on the cumulative equivalent full cycle count through a decay model function, which maps the historical cycle count to discrete values ​​of health.

[0036] Subsequently, the reverse recursive calculation phase begins. This phase starts from the last time period of the future cycle and calculates backwards to the first time period. First, the final state value function is initialized to zero, and a safety constraint on the final state energy is set to ensure that the stored energy does not fall below a safety threshold at the end of the cycle. Then, for each possible state in each time period t, the sum of the immediate reward after executing each allowed action and the optimal value function for the next time period is calculated, and the maximum value is taken as the optimal value function for that state. Through recursion across time periods, a complete state-value mapping table is constructed, which records the maximum cumulative reward that can be obtained in each state.

[0037] After the reverse recursion is completed, the optimal instruction is extracted in the forward direction. This stage starts from the known initial state and, based on the state-value mapping table generated in the reverse stage, determines the optimal charging and discharging actions for each time period in chronological order. By tracking the state transition process in real time, it is ensured that the action sequence simultaneously satisfies the energy transfer equation and the charging and discharging cycle number constraint, and finally outputs the complete charging and discharging instruction sequence for future cycles. This sequence specifies the charging, discharging, or standby operation in detail on a time-period basis.

[0038] To address uncertainties in actual operation, the method also introduces a rolling optimization and feedback correction mechanism. During the execution of the command sequence, the edge controller periodically uploads the actual power generation and load of the energy storage system to the cloud server. Based on the latest actual data, the cloud server recalculates and optimizes the commands for the remaining unexecuted periods, thereby correcting deviations caused by prediction errors. Furthermore, when the system detects a significant change in the load forecast or fluctuations in the electricity market transaction rate, it immediately triggers a command recalculation process, dynamically adjusting subsequent charging and discharging commands to ensure that the commands are always close to the optimal level.

[0039] Throughout the process, the required input information includes the energy storage power station's attribute parameters, user constraint parameters, and dynamic input parameters. Attribute parameters include at least the power station's geographical location, rated installed capacity, and rated capacity; these parameters determine the applicable electricity pricing policy and physical boundaries. User constraint parameters include at least the electricity consumption classification and billing method, which can be based on transformer capacity or maximum demand; these constraints directly affect the setting of power limits. Dynamic input parameters include at least the remaining energy capacity of the energy storage system at the time of model triggering, the maximum number of allowed charge / discharge cycles per day, battery charge / discharge depth safety constraints, and the historical cumulative number of charge / discharge cycles; these parameters ensure the real-time nature and adaptability of the commands.

[0040] In another embodiment, the core parameters are defined as follows: Parameter types include power plant attributes, user constraints, and dynamic inputs.

[0041] Power plant attributes include geographical location / electricity pricing area, installed capacity (P) max ), capacity (E) max The geographic location / electricity pricing area is used to determine the applicable electricity pricing policy and is a core parameter that directly determines the input electricity price curve. Incorrect settings will cause the strategy to fail completely. Installed power (Pmax) and capacity (Emax) serve as boundaries between charging / discharging power and electricity volume. Installed power (Pmax) determines the charging / discharging rate, affecting profitability during short-term high electricity price windows. Capacity (Emax) determines the total energy throughput. Both are physical hard constraints on the algorithm.

[0042] User constraints include electricity usage type (commercial / industrial) and billing method (capacity-based / demand-based). The electricity usage type (commercial / industrial) influences the choice of electricity price tier, determining which tier of price is used. The billing method (capacity-based / demand-based) is specifically based on the required transformer capacity S. tra Or, based on demand, the maximum demand D max This is a strongly constrained parameter. The discharge power generated by the strategy must ensure that (Pd + Loadt) does not exceed the required transformer capacity S.tra Or, based on demand, the maximum demand D max Otherwise, it will lead to a surge in electricity demand and costs for users, resulting in economic losses, and strict control measures are required.

[0043] Dynamic inputs include initial charge level E0, maximum charge / discharge cycle limit K, depth of charge / discharge DoD, and cumulative charge / discharge cycle count C. num The initial charge level E0 is the remaining energy storage capacity when the algorithm is triggered, which is the starting point affecting the strategy. If E0 is too low, it may not be able to discharge enough energy in the following discharge period; if E0 is too high, it may not be able to fully charge in the following charging period. The algorithm needs to be able to adapt to different E0 values. The charge / discharge cycle limit K is the number of charge / discharge mode switching allowed per day (K∈[1,2]), which is a key lifetime constraint. The smaller the K value (e.g., K=1), the worse the strategy flexibility, and some arbitrage opportunities may be missed; the larger the K value, the greater the battery life loss. This algorithm treats it as a state variable for the first time, accurately guaranteeing the optimality of the strategy under the lifetime constraint. The charge / discharge depth DoD is a battery safety constraint (e.g., DoD≤80%), which is a safety constraint. The shallower the DoD setting, the longer the battery life, but the smaller the available energy throughput, affecting the upper limit of revenue. A balance needs to be struck between revenue and life. The cumulative number of charge / discharge cycles C num This number is accumulated for each charge-discharge cycle and is the input for calculating Ht (health). Its absolute value reflects the battery's historical degree of wear and tear.

[0044] The specific dynamic programming algorithm is designed as follows: Time period discretization: Divide a calculation period into n equally spaced time periods (e.g., Δt = 5-15 minutes), with time period numbers t = 1, 2, ..., n.

[0045] Define state variable: S t = (E t C t H t ).

[0046] E t The remaining energy stored at the start of time period t (0≤E) t ≤E max E t Discretize the data during calculation. It is recommended to store half of the full-power charge / discharge energy at the minimum time interval. This reduces computational load and improves accuracy.

[0047] C t The remaining number of charge-discharge cycles before the start of time period t (0≤C) t ≤K).

[0048] H tThe limitation of State of Health (SOH) on charge / discharge capacity before the start of time period t, H t Discretization is performed during calculation. For example: H t ∈ {0.8, 0.9, 1.0} (corresponding to 80% / 90% / 100% of rated capacity).

[0049] Decision variables and payoff function: Action a_t includes charging, discharging, and standby; The benefit of charging is calculated as: Pc multiplied by Δt, then multiplied by Price. t The discharge benefit is calculated as Pd multiplied by Δt and then multiplied by Price. t The standby revenue is calculated as 0.

[0050] The charging constraint is the product of ηc, Pc, and Δt, plus E. t The sum of the values ​​is less than or equal to the capacity Emax, and at the same time, Pc is increased by the Load. t The sum ≤ min(S) tra D max ).

[0051] The constraint condition for discharge is E t Subtract (Pd multiplied by Δt divided by ηd) ≥ (1-DoD•H) t Then multiply by the capacity Emax, and at the same time, Pd≤(P max Load t ).

[0052] There is no power constraint in standby mode.

[0053] Where: ηc / ηd is the charge / discharge efficiency (default 85%), Load t User load during time period t (can be based on prediction).

[0054] State transition equation: The equation for charge transfer is: E t In the charging state, it equals ηc×Pc×∆t plus E(t-1); E t In the discharge state, it is equal to E(t-1) minus (Pd multiplied by Δt divided by ηd); E t In standby mode, it equals E(t-1).

[0055] The recurrence relation is: C t Equals C t -1, when at ≠ a(t-1) and at and a(t-1) are switching between charging and discharging; in other cases, Ct Equals C t . at represents the charging and discharging action decision during time period t.

[0056] Where H t The state transition rule is: H t equals f(H) t-1 C num The function f represents the mapping relationship between the cumulative number of cycles and the discrete value of battery health. This function can be provided by experimental data or battery manufacturers.

[0057] In another embodiment, the present invention also designs a battery health degradation model based on the cumulative number of cycles to quantify H. t Changes: H t equals f(C) total ) equals H initial Subtract α(C) total +ΔC) β Among them, H initial For the initial battery health (usually 1.0), C total The cumulative equivalent full cycle count (accumulated from the start of battery use) is denoted by ΔC, which represents the additional equivalent cycle count since the last strategy calculation (calculated based on the depth of charge / discharge for each cycle; for example, an 80% DoD cycle is equivalent to 0.8 full cycles). α and β are battery degradation model coefficients, obtained by fitting battery aging experimental data. For example, for a certain type of lithium iron phosphate battery, experimental measurements show α = 0.001 and β = 1.2.

[0058] During the state transition in this period, H t The value is calculated from the theoretical value by the above model and mapped to the closest discrete value (such as 0.8, 0.9, 1.0). This model directly links battery life degradation with operating strategies, enabling dynamic programming to pursue not only maximizing short-term gains, but also maximizing overall gains after considering long-term life degradation.

[0059] Objective function and recursive solution: The reverse recursive process involves calculating backwards from the last time period, solving for the optimal value function for each time period. In each time period t, based on the optimal states of subsequent time periods, the maximum cumulative reward obtainable under the current state St is calculated. This stage constructs a complete state-value mapping table, laying the foundation for positive decision-making.

[0060] The detailed steps are as follows: Initialize the boundary state and set the final state value: V n+1 (S n+1 ) equals 0 ∀S n+1 Define the safety final state constraint En+1 ≥(1-DoD) multiplied by Emax.

[0061] The optimal value function is calculated recursively in reverse order, decreasing from time period t=n to t=1: V t (S t ) equals max{R t (S t ,a t )+V t+1 S t+1}; where at∈{charging, discharging, standby}, S t = (E t C t H t S represents the state at the beginning of time period t. t+1 After executing action at, the state transitions to the end of time period t / start of t+1; the state transition sequence is S. t via at to S t+1 .

[0062] Forward extraction phase: Starting from the initial state S0, the optimal charging and discharging actions for each time period are extracted sequentially based on the value mapping table generated in the reverse phase. Through a real-time state tracking mechanism, it is ensured that the action sequence simultaneously satisfies the power transfer equation and the cycle count constraint, forming the optimal strategy for the entire day.

[0063] This invention establishes a maximum profit function and uses basic power plant information and electricity price information to calculate a charging and discharging strategy that maximizes profits. Compared with intelligent algorithms, it has lower computational overhead and requires less data.

[0064] To verify the effectiveness of this invention, a comparative test was conducted for one month on a 500kW / 1MWh commercial and industrial energy storage power station in actual operation in the Yangtze River Delta region. During the test, the local EMS adopted a fixed two-charge, two-discharge strategy, while this system automatically generated and issued a dynamic strategy daily.

[0065] Based on the test, the following data and conclusions were obtained: The total monthly revenue of the traditional fixed strategy is 42,580 yuan, the strategy maintenance time (person-hours) is 3.5, and the strategy adjustment response delay is 4-5 working days; The total monthly revenue of the dynamic strategy of this invention is RMB 48,750, the strategy maintenance time (person-hours) is 0.5, and the strategy adjustment response delay is <15 minutes; Compared to traditional fixed strategies, the advantages of this invention are: Monthly total revenue (RMB) increased by 14.5%, strategy maintenance time (person-hours) decreased by 85.7%, and strategy adjustment response latency was basically real-time.

[0066] The core advantage of this system lies in its real-time and automatic response to policy changes, and its logic is as follows: Cloud-based monitoring and analysis: The cloud platform has a policy monitoring module that continuously crawls and analyzes electricity price policy documents (such as notices and detailed rules) published on the official websites of power grid companies in various provinces and cities.

[0067] Policy Feature Extraction and Standardization: Automatically extract key features from policies, including but not limited to: effective date, applicable billing area, new peak / flat / valley / peak / valley time period division, electricity price for each time period, special electricity price for major holidays, etc., and convert them into standard data formats within the system.

[0068] Policy triggering and recalculation: Planned triggering: The system initiates a new day's strategy calculation for all power stations at a fixed time each day (such as midnight).

[0069] Event-driven triggering: Once a new electricity price policy is identified as being issued and effective in a certain region, the system immediately sends a policy recalculation instruction to all energy storage power stations within the policy's coverage area.

[0070] Edge Policy Reception and Execution: The edge controller (EMS) receives the latest and optimal policy sequence from the cloud in real time through the API interface and seamlessly replaces the old policy without manual confirmation (alarms can be set to notify maintenance personnel), thus realizing a fully automatic closed loop from policy change to policy execution, with response latency controllable in the minute level.

[0071] Rolling optimization and feedback correction: The system does not execute the strategy only once a day. When the load forecast changes or the electricity price fluctuates, the calculation process will be triggered again, that is, the charging and discharging strategy will be readjusted from time to time.

[0072] All data collection and extraction in this invention are carried out under compliant and legal conditions.

Claims

1. A cloud-edge collaborative energy storage power station dynamic control method, characterized in that, include: The cloud server obtains instructions from the energy storage power station to generate the required input information, and the edge controller obtains the real-time status information of the energy storage power station. Based on the input information and real-time status information, a charging and discharging instruction sequence for future cycles is generated through a dynamic programming model. The dynamic programming model constructs a joint state space that includes the remaining power, the remaining number of charging and discharging cycles, and the battery health, and performs reverse recursive calculation with operating revenue as the objective function, and finally extracts the optimal charging and discharging instruction sequence in the forward direction. The optimal charge and discharge command sequence is sent to the edge controller to control the operation of the energy storage power station. 2.The cloud-edge collaborative energy storage power station dynamic control method of claim 1, wherein, The dynamic programming model specifically includes: dividing the future period into n equally spaced time periods; defining state variables as triples, wherein the triples include: The remaining energy storage capacity at the start of the time period ranges from 0 to the rated capacity of the energy storage system. It is discretized during the calculation, and the discretization step size is half of the full-power charging and discharging energy of the energy storage within the minimum time interval. The number of charge / discharge cycles remaining before the start of the time period, ranging from 0 to the maximum number of charge / discharge cycles allowed per day; The battery health status before the start of the time period limits the charging and discharging capacity, and its value is selected from a preset discrete set.

3. The cloud-edge collaborative energy storage plant dynamic control method of claim 2, wherein, The generation of the charging and discharging instruction sequence for future cycles through the dynamic programming model specifically includes: in each time period, the allowed actions include charging, discharging, and standby; when the action is charging, the revenue is calculated as the product of the negative charging power, the time period length, and the electricity market transaction rate for that time period, and must meet the following constraints: the power after charging does not exceed the capacity, and the power constraint, that is, the sum of the charging power and the predicted load for that time period does not exceed the smaller value between the transformer capacity and the maximum demand.

4. The cloud-edge collaborative energy storage plant dynamic control method of claim 3, wherein, The generation of charging and discharging instruction sequences for future cycles through a dynamic programming model also includes: when the action is discharging, the revenue is calculated as the product of the discharge power, the duration, and the electricity market transaction rate for that duration, and must meet the following constraints: the discharge power must not be lower than the safety threshold considering the discharge depth and health, and the power constraint must not exceed the smaller of the system rated power and the current load; when the action is standby, the revenue is zero and there is no power constraint.

5. The cloud-edge collaborative energy storage plant dynamic control method of claim 4, wherein, The generation of charging and discharging command sequences for future cycles through the dynamic programming model also includes establishing state transition equations, specifically including: the power transition equation is defined as follows: if charging, the power at the end of the period is the power at the beginning of the period plus the charging energy after considering charging efficiency; if discharging, the power at the end of the period is the power at the beginning of the period minus the discharging energy after considering discharging efficiency; if in standby mode, the power remains unchanged. The cycle number transition equation is defined as follows: if the action in the current period is different from the action in the previous period and involves a switching of charging and discharging states, the remaining cycle number is reduced by one; otherwise, it remains unchanged. The state transition of health is calculated based on the cumulative equivalent full cycle number through the decay model function.

6. The cloud-edge collaborative energy storage plant dynamic control method of claim 5, wherein, The reverse recursive calculation specifically includes: starting from the last period of the cycle and working backwards to the first period, initializing the final state value function to zero, and setting a final state power safety constraint; for each possible state in each period, calculating the sum of the immediate benefit after executing each allowed action and the optimal value function of the next period, taking the maximum value as the optimal value function of that state, and constructing a complete state-value mapping table.

7. The cloud-edge collaborative energy storage plant dynamic control method according to claim 1 or 6, characterized in that, The forward extraction of the optimal charge and discharge command sequence specifically includes: starting from the known initial state, determining the optimal charge and discharge actions for each time period in chronological order based on the state and value mapping table generated in the reverse recursive stage; ensuring that the action sequence simultaneously satisfies the power transfer equation and the number of charge and discharge cycles by tracking the state transition in real time, and finally outputting the complete charge and discharge command sequence for future cycles. 8.The cloud-edge collaborative energy storage power station dynamic control method of claim 7, wherein, During the execution of the instruction sequence, the edge controller periodically uploads the actual power and actual load of the energy storage system to the cloud server. The cloud server recalculates and optimizes the instructions for the remaining unexecuted periods based on the latest actual data, correcting deviations caused by prediction errors. When a significant change in the load forecast value or fluctuations in the electricity market transaction rate are detected, the system immediately triggers the instruction recalculation process to dynamically adjust subsequent charging and discharging instructions. 9.The cloud-edge collaborative energy storage power station dynamic control method of claim 1 or 8, wherein, The required input information includes the attribute parameters, user constraint parameters, and dynamic input parameters of the energy storage power station; the attribute parameters include at least the geographical location information, rated installed power, and rated capacity of the power station.

10. The cloud-edge collaborative energy storage plant dynamic control method of claim 9, wherein, The user constraint parameters include at least the electricity consumption type classification and billing method, wherein the billing method is based on transformer capacity or maximum demand; the dynamic input parameters include at least the remaining power of the energy storage system at the time of model triggering, the upper limit of the daily allowable charge and discharge cycle count, the battery charge and discharge depth safety constraint, and the historical cumulative charge and discharge cycle count.

Images