An energy storage control method, device, equipment and medium of an integrated energy system

Abstract

This invention provides an energy storage control method, device, equipment, and medium for an integrated energy system. The method includes: acquiring time-of-use electricity price information, bus voltage deviation, current wind and solar power output data, and current state of charge (SOC); performing time series decomposition on the current wind and solar power output data to obtain long-cycle trend components and short-cycle fluctuation components; invoking a first intelligent agent to determine a target SOC and a charge / discharge reference power sequence based on the time-of-use electricity price information, current SOC, and long-cycle trend components; invoking a second intelligent agent to determine a power correction coefficient based on the current wind and solar power output data, current SOC, target SOC, bus voltage deviation, and short-cycle fluctuation components; and generating a target energy storage charge / discharge power sequence based on the power correction coefficient and the charge / discharge reference power sequence. By having a long-cycle intelligent agent responsible for establishing stable charge / discharge cycles and a short-cycle intelligent agent performing minor corrections, the equivalent number of charge / discharge cycles of the battery is significantly reduced.

Description

Technical Field

[0001] This invention belongs to the field of integrated energy optimization and scheduling technology, and particularly relates to an energy storage control method for an integrated energy system, an energy storage control device for an integrated energy system, an electronic device, and a computer-readable storage medium. Background Technology

[0002] In a combined wind, solar, and energy storage system, there is a significant temporal and spatial mismatch in the output characteristics of wind and solar power. The peak output of solar power is concentrated from noon to afternoon, while the peak output of wind power often occurs from night to early morning. The daily power generation curves of the two have a low overlap rate. This inherent characteristic means that a single energy storage system needs to absorb excess solar power during the day and release power at night to make up for insufficient wind power output.

[0003] Existing energy storage scheduling strategies typically employ a uniform control cycle, which requires the same group of energy storage batteries to frequently switch between charge and discharge states. This not only leads to a decrease in system efficiency but also accelerates battery life degradation due to a surge in equivalent charge and discharge cycles. Summary of the Invention

[0004] In view of the above problems, embodiments of the present invention are proposed to provide an energy storage control method for an integrated energy system, an energy storage control device for an integrated energy system, an electronic device, and a computer-readable storage medium that overcome or at least partially solve the above problems.

[0005] To address the aforementioned problems, a first aspect of the present invention provides an energy storage control method for an integrated energy system, the method comprising: The system acquires time-of-use electricity price information, bus voltage deviation, and current wind and solar power output data and current state of charge of the wind-solar-storage combined power generation system; the current wind and solar power output data is a data sequence of the active power output of the wind turbine generator and photovoltaic array of the wind-solar-storage combined power generation system over time. The current wind and solar power output data is decomposed into a time series to obtain the long-term trend component and the short-term fluctuation component of the current wind and solar power output data; the long-term trend component is the power component in the current wind and solar power output data whose rate of change is lower than a first rate threshold and whose duration is longer than a first duration; the short-term fluctuation component is the power component in the current wind and solar power output data whose rate of change is higher than a second threshold and whose duration is shorter than a second duration. The first intelligent agent is invoked with a first preset time scale as the control period, and based on the time-of-use electricity price information, the current state of charge, and the long-period trend component, the target state of charge and the charging / discharging reference power sequence are determined; the first preset time scale and the long-period trend component are periodically matched. The second intelligent agent is invoked with a second preset time scale as the control period, and a power correction coefficient is determined based on the current wind and solar power output data, the current state of charge, the target state of charge, the bus voltage deviation, and the short-period fluctuation component; the second preset time scale is matched with the period of the short-period fluctuation component. Based on the power correction coefficient and the charge / discharge reference power sequence, a target energy storage charge / discharge power sequence is generated.

[0006] Optionally, the step of performing time series decomposition on the current wind and solar power output data to obtain the long-term trend component and short-term fluctuation component of the current wind and solar power output data includes: The long-period trend component is extracted by performing wavelet decomposition or moving average filtering on the current wind and solar power output data. Based on the long-cycle trend component and the current wind and solar power output data, the short-cycle fluctuation component is determined.

[0007] Optionally, the first agent and the second agent are trained in the following manner: Obtain the coupled reward function and historical landscape output data; The historical power output data is preprocessed, and a dataset is constructed based on the preprocessed historical power output data. The initial first agent and the initial second agent are jointly trained based on the dataset and the coupled reward function. The joint training takes the coupled reward function as the optimization objective, the target state of charge and the charge-discharge reference power sequence output by the first agent as the input constraints of the second agent, and the power correction coefficient output by the second agent adjusts the charge-discharge reference power sequence. After training is completed, the first agent and the second agent are obtained.

[0008] Optionally, the coupled reward function is obtained by weighted summation of time-shifting efficiency reward, power smoothing reward, and lifetime decay penalty; the time-shifting efficiency reward is determined based on the energy transfer efficiency and peak-valley arbitrage revenue within each scheduling cycle; the power smoothing reward is determined based on the root mean square value of the rate of change of grid-connected power within the evaluation time interval of the power smoothing effect; and the lifetime decay penalty is determined based on the equivalent number of charge-discharge cycles.

[0009] Optionally, the invocation of the first intelligent agent, using a first preset time scale as the control period, determines the target state of charge and the charging / discharging reference power sequence based on the time-of-use electricity price information, the current state of charge, and the long-period trend component, including: Obtain the predicted wind and solar power output data within the first preset time period in the future; The first intelligent agent is invoked to perform inference based on the predicted wind and solar power output data within the first preset time period, the time-of-use electricity price information, the current state of charge and the long-term trend component, with the first preset time scale as the control period, and outputs the target state of charge and the charging and discharging reference power sequence.

[0010] Optionally, the invocation of the second intelligent agent, using a second preset time scale as the control period, determines a power correction coefficient based on the current wind and solar power output data, the current state of charge, the target state of charge, the bus voltage deviation, and the short-period fluctuation component, including: Determine the deviation between the current wind and solar power output data and the predicted wind and solar power output data; Determine the deviation between the current state of charge and the target state of charge; The second intelligent agent is invoked with a second preset time scale as the control period. Based on the deviation between the current wind and solar power output data and the predicted wind and solar power output data, the deviation between the current state of charge and the target state of charge, the bus voltage deviation, and the short-period fluctuation component, inference is performed, and a power correction coefficient is output.

[0011] Optionally, generating the target energy storage charge / discharge power sequence based on the power correction coefficient and the charge / discharge reference power sequence includes: When the deviation between the current wind and solar power output data and the predicted wind and solar power output data is less than or equal to a preset deviation threshold, the charging and discharging reference power sequence is used as the target energy storage charging and discharging power sequence. When the deviation between the current wind and solar power output data and the predicted wind and solar power output data is greater than the preset deviation threshold, the power correction coefficient is superimposed on the charging and discharging reference power sequence to generate the target energy storage charging and discharging power sequence.

[0012] Optionally, before generating the target energy storage charge / discharge power sequence, the method further includes: If the deviation between the current state of charge and the target state of charge exceeds a preset deviation threshold, the second agent is invoked to output a state of charge correction instruction in order to make the deviation between the current state of charge and the target state of charge less than or equal to the preset deviation threshold.

[0013] Optionally, the method further includes: Obtain the actual operating results of executing the target energy storage charging and discharging power sequence to update the historical wind and solar power output dataset; Based on the updated wind and solar power output dataset, the first and second agents are optimized and trained, and the performance of the optimized first and second agents is evaluated with that of the current first and second agents. If the performance of the first and second agents after optimization training is better than that of the currently deployed first and second agents, then the current first and second agents should be updated.

[0014] According to a second aspect of the present invention, an energy storage control device for an integrated energy system is provided, the device comprising: The parameter acquisition module is used to acquire time-of-use electricity price information, bus voltage deviation, and the current wind and solar power output data and current state of charge of the wind-solar-storage combined power generation system; the current wind and solar power output data is a data sequence of the active power output of the wind turbine generator and photovoltaic array of the wind-solar-storage combined power generation system changing over time. The time-series decomposition module is used to perform time-series decomposition on the current wind and solar power output data to obtain the long-cycle trend component and the short-cycle fluctuation component of the current wind and solar power output data; the long-cycle trend component is the power component in the current wind and solar power output data whose rate of change is lower than a first rate threshold and whose duration is longer than a first duration; the short-cycle fluctuation component is the power component in the current wind and solar power output data whose rate of change is higher than a second threshold and whose duration is shorter than a second duration. A power sequence determination module is used to call a first intelligent agent to determine a target state of charge and a charging / discharging reference power sequence based on the time-of-charge price information, the current state of charge, and the long-period trend component, using a first preset time scale as the control period; the first preset time scale is matched with the period of the long-period trend component; The correction coefficient determination module is used to call the second intelligent agent to determine the power correction coefficient based on the current wind and solar power output data, the current state of charge, the target state of charge, the bus voltage deviation, and the short-period fluctuation component, with the second preset time scale as the control period; the second preset time scale is matched with the period of the short-period fluctuation component. The target power sequence generation module is used to generate a target energy storage charge and discharge power sequence based on the power correction coefficient and the charge and discharge reference power sequence.

[0015] According to a third aspect of the present invention, an electronic device is provided, comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the energy storage control method of the integrated energy system as described in any of the preceding claims.

[0016] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, on which a program is stored, which, when executed by a processor, implements the steps of the energy storage control method of the integrated energy system as described in any of the preceding claims.

[0017] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses an energy storage control method, device, equipment, and medium for an integrated energy system. The method includes: acquiring time-of-use electricity price information, bus voltage deviation, and current wind and solar power output data and current state of charge (SOC) of a wind-solar-storage combined power generation system; performing time series decomposition on the current wind and solar power output data to obtain long-term trend components and short-term fluctuation components; invoking a first intelligent agent with a first preset time scale as the control period to determine a target SOC and a charging / discharging reference power sequence based on time-of-use electricity price information, current SOC, and long-term trend components; invoking a second intelligent agent with a second preset time scale as the control period to determine a power correction coefficient based on the current wind and solar power output data, current SOC, target SOC, bus voltage deviation, and short-term fluctuation components; and generating a target energy storage charging / discharging power sequence based on the power correction coefficient and the charging / discharging reference power sequence. By performing time series decomposition on the wind and solar power output data, extracting the long-term trend component reflecting the macro trend, and having the first intelligent agent perform energy scheduling, efficient energy transfer is achieved. Short-period fluctuation components are extracted, and a second agent provides a real-time response output power correction coefficient to compensate for the baseline command, resulting in a more stable output power. Through dual-timescale task decoupling, the first agent is responsible for outputting stable charge and discharge power, while the second agent makes minor corrections based on this, thereby significantly reducing the equivalent charge and discharge cycles of the battery.

[0018] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0019] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a flowchart of the steps of an energy storage control method for an integrated energy system provided by the present invention; Figure 2 This is a flowchart of the steps of another integrated energy system energy storage control method provided by the present invention; Figure 3 This is a logic block diagram of an energy storage control method for an integrated energy system provided by the present invention; Figure 4 This is a structural block diagram of an energy storage control device for an integrated energy system provided by the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Figure 1 This is a flowchart illustrating the steps of an energy storage control method for an integrated energy system according to an embodiment of the present invention. See also... Figure 1 The method specifically includes the following steps: Step 101: Obtain time-of-use electricity price information, bus voltage deviation, and current wind and solar power output data and current state of charge of the wind-solar-storage combined power generation system; the current wind and solar power output data is a data sequence of the active power output of the wind turbine generator and photovoltaic array of the wind-solar-storage combined power generation system changing over time. To address the core shortcomings of energy storage scheduling conflicts, low efficiency, and accelerated lifespan degradation caused by the mismatch between wind and solar power output time periods, this invention proposes a solution for a combined power generation system with significant wind and solar power output time period mismatch. This solution aims to effectively decouple the rigid coupling relationship between the energy storage system and the short-cycle power smoothing task, thereby maximizing the combined utilization rate of wind and solar energy and extending the effective lifespan of the energy storage system while ensuring grid security and load power supply reliability.

[0022] Historical and real-time data from the wind-solar-storage combined power generation system are collected, including: minute-level output data of the photovoltaic array and wind turbine, load power data, state of charge and health status of the energy storage system, and time-of-use electricity price information. In the main controller of the wind-solar-storage combined power generation system, the following data are collected in real time: active power sequence output by the photovoltaic inverter (sampling period of 1 minute); active power and wind speed sequence of the wind turbine (sampling period of 1 minute); load power and frequency / voltage data at the grid connection point (sampling period of 1 second); SOC and SOH (health status) data reported by the energy storage battery management system (sampling period of 1 second); and time-of-use electricity price information from the external grid (updated hourly).

[0023] Time-of-use (TOU) pricing refers to the differentiated electricity purchase and sales prices set by the electricity market or grid company based on the electricity supply and demand relationship at different times of the day. A day is typically divided into peak, normal, and off-peak periods, with different prices for each period. For example, electricity prices are higher during peak daytime hours and lower during off-peak nighttime hours. Bus voltage deviation refers to the difference between the actual measured bus voltage at the grid connection point of a wind-solar-storage combined generation system and the rated voltage, usually expressed as a per-unit value or percentage. For example, if the rated voltage is 10kV and the actual measured value is 10.2kV, the voltage deviation is +0.2kV or +2%. Current wind and solar power output data refers to the data sequence of active power output from wind turbine generators and photovoltaic arrays in a wind-solar-storage combined generation system over time. Wind power output is affected by wind speed and direction, while photovoltaic output is affected by sunlight intensity and temperature; both exhibit intermittent and fluctuating characteristics. Current state of charge (SBC) refers to the percentage of the current remaining capacity of the energy storage battery relative to its rated capacity, typically ranging from 0% to 100%. For example, SOC=80% means that the battery has 80% of its rated capacity remaining and can continue to discharge; SOC=20% means that the battery capacity is low and deep discharge should be avoided to protect the battery.

[0024] In this embodiment of the invention, the data acquisition and initialization process is first initiated to comprehensively acquire the multi-source heterogeneous data required for the operation of the wind-solar-storage combined power generation system. Specifically, the following key information is read in real time through the data acquisition interface: time-of-use electricity price information released by the external power grid, which is usually updated on an hourly basis, reflecting the price difference between purchasing and selling electricity at different times, and providing an economic optimization basis for long-term energy management; the current state of charge reported by the energy storage converter or battery management system, which represents the current remaining capacity of the energy storage battery in percentage form, and is the core constraint variable that determines the charging and discharging capacity of the energy storage and prevents overcharging and over-discharging; the current wind and solar power output data output by the wind turbine generator and photovoltaic array in the wind-solar-storage combined power generation system, which is a time series of active power changes over time, with a sampling period of usually 1 minute, reflecting the real-time power generation capacity of wind and solar resources; and the bus voltage deviation parameter, which is the difference between the actual measured bus voltage and the rated voltage, used to characterize the power quality status on the grid side, and providing a basis for judging the voltage support requirements for short-term power response.

[0025] After data cleaning and outlier removal, the aforementioned data provides the foundation for subsequent time-series decomposition and agent decision-making. This step is the data entry point for the entire energy storage control method. The completeness and accuracy of the collected data directly determine whether long-cycle and short-cycle agents can make reasonable decisions, thereby affecting global performance indicators such as wind and solar energy transfer efficiency, power fluctuation mitigation effect, and energy storage battery life degradation.

[0026] Step 102: Perform time series decomposition on the current wind and solar power output data to obtain the long-cycle trend component and the short-cycle fluctuation component of the current wind and solar power output data; the long-cycle trend component is the power component in the current wind and solar power output data whose rate of change is lower than a first rate threshold and whose duration is longer than a first duration; the short-cycle fluctuation component is the power component in the current wind and solar power output data whose rate of change is higher than a second threshold and whose duration is shorter than a second duration. Time series decomposition is performed on the wind and solar power output data to extract long-term trend components (hourly smooth curves) and short-term fluctuation components (second / minute high-frequency noise). Wavelet decomposition or moving average filtering is then applied to the collected wind and solar power output data to obtain long-term trend components (cutoff frequency 1 / 60Hz, i.e., retaining hourly variations) and short-term fluctuation components, which are used as feature inputs for long-term and short-term agents, respectively.

[0027] In this embodiment of the invention, the collected current wind and solar power output data (i.e., the time series of active power output from wind turbine generators and photovoltaic arrays) is decomposed into a time series, decoupling it into a long-period trend component and a short-period fluctuation component. The wind and solar power output data is essentially a non-stationary time series, containing two different time-scale components: a slowly changing long-period trend component dominated by macroscopic factors such as sunrise and sunset, and weather processes, for example, the daily cycle of photovoltaic power output peaking at noon and approaching zero at night, and the continuous hourly changes in wind power output influenced by large-scale weather systems; and a drastically changing high-frequency fluctuation component caused by microscopic factors such as rapid cloud movement, atmospheric turbulence, and wake effects, for example, a cloud blocking a photovoltaic array can reduce output by more than 30% within seconds, or second-level pulsations in wind speed can cause instantaneous jumps in wind power.

[0028] The long-cycle trend component is the power component in the current wind and solar power output data whose rate of change is lower than the first rate threshold and whose duration is longer than the first duration, typically characterized by smooth changes on an hourly scale. The short-cycle fluctuation component is the power component whose rate of change is higher than the second rate threshold and whose duration is shorter than the second duration, typically characterized by high-frequency noise on the second / minute scale. The decomposition can be implemented using wavelet decomposition (such as Daubechies wavelet or Symlets wavelet, which decomposes the signal into different frequency bands through multi-resolution analysis) or moving average filtering (calculating a moving average with an hourly window width as the trend component, and the difference between the original signal and the trend component as the fluctuation component). Wavelet decomposition better preserves the local features of the signal, while moving average filtering has the advantages of simple calculation and suitability for real-time processing. After decomposition, the long-cycle trend component is sent to the first intelligent agent (long-cycle energy management module) as the core basis for planning the 24-hour energy storage charging and discharging "outline plan"; the short-cycle fluctuation component is sent to the second intelligent agent (short-cycle power response module) as a key input for its real-time monitoring and smoothing of instantaneous fluctuations in wind and solar power output. This decomposition operation effectively separates the "slow-changing trends" and "fast-changing fluctuations" in the wind and solar power output data, laying a data foundation for the subsequent decoupling of dual time-scale tasks. This allows long-period agents to focus on the transfer of energy between time periods without being disturbed by instantaneous noise, and short-period agents to focus on the real-time smoothing of high-frequency fluctuations without being constrained by macro trends.

[0029] Step 103: Invoke the first intelligent agent to determine the target state of charge and charging / discharging reference power sequence based on the time-of-use electricity price information, the current state of charge, and the long-cycle trend component, using a first preset time scale as the control period; the first preset time scale is matched with the period of the long-cycle trend component. The first intelligent agent is the long-cycle energy management module in the dual-time-scale collaborative control architecture of this invention. It is responsible for macroscopically planning the energy storage charging and discharging behavior of the wind-solar-storage combined power generation system on an hourly timescale. Its core task is to formulate the energy storage charging and discharging "outline plan" (i.e., charging and discharging reference power sequence) and the target trajectory of the state of charge for the next 24 hours, realize the transfer of energy between wind and solar surplus periods and wind and solar shortage periods, and optimize peak-valley arbitrage profits by combining time-of-use electricity price information.

[0030] In this embodiment of the invention, a first intelligent agent (long-cycle energy management module) that has been trained is invoked. Using a first preset time scale as the control period, and based on time-of-use electricity price information, the current state of charge, and the decomposed long-cycle trend components, the target state of charge trajectory and the charging / discharging reference power sequence are determined for a future period. The first preset time scale matches the hourly variation period of the long-cycle trend components, ensuring that the first intelligent agent can respond to macroscopic trend changes in wind and solar power output while ignoring instantaneous noise at the second / minute level.

[0031] The first intelligent agent first acquires predicted wind and solar power output data for a predetermined future time period (e.g., the next 24 hours). This predicted data, together with long-term trend components, reflects the macroscopic changes in wind and solar resources within the future period. Subsequently, the first intelligent agent uses time-of-use electricity price information (reflecting the differences in electricity purchase and sale prices at different times), the current state of charge (representing the current remaining capacity of energy storage), and the long-term trend components as input states, and performs a forward inference calculation through its internal deep neural network. The first intelligent agent adopts a deep deterministic policy gradient algorithm, whose Actor network directly outputs continuous action values, namely, the energy storage charging and discharging reference power sequence for each hour (24 time points) within the next 24 hours, while simultaneously planning the corresponding energy storage state of charge target trajectory. In the output charging and discharging reference power sequence, positive values ​​represent discharging (power transmission to the grid), and negative values ​​represent charging (power absorption from the grid or wind and solar systems), with amplitudes between -0.5C and 0.5C, satisfying the power limit constraints of the energy storage system. This benchmark power sequence reflects the scheduling of energy storage charging during the midday period when photovoltaic power is abundant, and the scheduling of energy storage discharging during the nighttime period when wind power is insufficient. It also incorporates time-of-use pricing information, prioritizing charging during off-peak hours and discharging during peak hours, thereby maximizing energy movement over time and peak-valley arbitrage profits. After determining the target state of charge and the benchmark charging / discharging power sequence, the first intelligent agent transmits these decision results to the second intelligent agent (short-cycle power response module) in the form of commands, serving as the benchmark for subsequent real-time adjustments. Through this step, the energy management task of the wind-solar-storage combined power generation system is decoupled from the task of suppressing instantaneous fluctuations. The first intelligent agent is then dedicated to hourly macro-level scheduling and planning, providing top-level decision support for the economical operation and efficient energy utilization of the entire system.

[0032] Step 104: Invoke the second intelligent agent to determine the power correction coefficient based on the current wind and solar power output data, the current state of charge, the target state of charge, the bus voltage deviation, and the short-period fluctuation component, using the second preset time scale as the control period; the second preset time scale is matched with the period of the short-period fluctuation component. This invention establishes a decoupled energy storage scheduling and control architecture, comprising a long-cycle energy management module and a short-cycle power response module. The long-cycle energy management module, with a 1-hour time resolution, plans the target trajectory of energy storage SOC and the charging / discharging reference power sequence for the next 24 hours. The goal of this module is to maximize the time-shifting efficiency of wind and solar energy, i.e., "storing electricity when solar power is abundant and discharging when wind power is insufficient." The short-cycle power response module, with a 1-second to 1-minute time resolution, receives the reference power command issued by the long-cycle module. This module monitors in real time the deviation between the actual wind and solar power output and the predicted values, as well as the instantaneous fluctuations in grid-side frequency / voltage, and outputs a corrected power value superimposed on the reference command to achieve immediate smoothing of fluctuations.

[0033] The two modules are modeled as a collaborative reinforcement learning agent. The long-term agent's state includes the 24-hour wind and solar power forecast sequence, the time-of-use electricity price sequence, and the current state of charge (SOC). The short-term agent's state includes the real-time wind and solar power output deviation, SOC deviation, bus voltage deviation, and energy storage power limitation. The long-term agent's action is the energy storage reference charge and discharge power for the next hour (a continuous value, between -0.5C and 0.5C); the short-term agent's action is the power correction coefficient (ranging from -0.2 to 0.2, multiplicatively added to the reference power).

[0034] The second intelligent agent is the short-cycle power response module in the dual-timescale collaborative control architecture. It is responsible for real-time correction of the energy storage charging and discharging power of the wind-solar-storage combined generation system, using a second-level timescale. Its core task is to dynamically output a power correction coefficient based on the charging and discharging reference power issued by the first intelligent agent (long-cycle energy management module), taking into account the instantaneous fluctuations in wind and solar power output, the tracking error of the energy storage state of charge, and the bus voltage deviation. This achieves immediate smoothing of power fluctuations at the second / minute level. The first intelligent agent outputs macroscopic reference commands, and the second intelligent agent performs real-time microscopic corrections based on these commands. Together, they complete the full-timescale control of the energy storage system. The second intelligent agent is deployed on the local energy storage controller (energy storage converter side), possessing independent operating capabilities and not relying on real-time cloud communication, ensuring the real-time performance and reliability of the control.

[0035] In this embodiment of the invention, a trained second agent (short-cycle power response module) is invoked. Using a second preset time scale as the control period, and based on the current wind and solar power output data, the current state of charge, the target state of charge issued by the first agent, the bus voltage deviation, and the decomposed short-cycle fluctuation components, the power correction coefficient is determined in real time. The second preset time scale matches the second / minute-level variation period of the short-cycle fluctuation components, ensuring that the second agent can capture and respond to the instantaneous high-frequency fluctuations in wind and solar power output.

[0036] The second agent first processes the input data in real time, calculating the real-time deviation between the current wind and solar power output data and the predicted wind and solar power output data (reflecting real-time fluctuation amplitude), calculating the real-time deviation between the current state of charge (SOC) and the target SOC trajectory issued by the first agent (reflecting SOC tracking error), and simultaneously acquiring the real-time bus voltage deviation (reflecting the power quality of the grid) and short-period fluctuation components (reflecting the high-frequency noise characteristics of wind and solar power output). Subsequently, the second agent inputs these state variables into its internal deep neural network for a forward inference calculation. The second agent employs a dual-delay deep deterministic strategy gradient algorithm, whose Actor network directly outputs continuous power correction coefficients. These coefficients are multiplicative factors, typically limited to the range [-0.2, 0.2]. The power correction coefficient represents a relative proportional adjustment based on the charging / discharging reference power issued by the long-period agent; a positive correction coefficient indicates increased discharging or decreased charging, while a negative value indicates decreased discharging or increased charging. For example, when rapid cloud movement causes a sudden drop in photovoltaic output within seconds, the real-time deviation between the current wind and solar output data and the predicted value is a large negative value. The second intelligent agent then outputs a positive correction coefficient, increasing the actual discharge power of the energy storage (or decreasing the charging power) to compensate for the power drop. Conversely, when wind and solar output surges, the second intelligent agent outputs a negative correction coefficient, causing the energy storage to increase charging to absorb excess power. Through this mechanism, the second intelligent agent can achieve real-time smoothing of second / minute-level fluctuations in wind and solar output. Even if cloud communication is briefly interrupted, the second intelligent agent in the local controller can continue to operate based on the most recently received reference power curve, ensuring the system's control reliability.

[0037] Step 105: Generate the target energy storage charge and discharge power sequence based on the power correction coefficient and the charge and discharge reference power sequence.

[0038] In this embodiment of the invention, the charging and discharging reference power sequence determined by the first intelligent agent is fused with the power correction coefficient output in real time by the second intelligent agent to generate the final target energy storage charging and discharging power sequence, which serves as the actual control command for the energy storage converter. The power correction coefficient is applied to the reference power using a multiplicative superposition method, and the specific calculation formula is: Target energy storage charging and discharging power = Charging and discharging reference power × (1 + Power correction coefficient). The charging and discharging reference power is updated by the first intelligent agent on an hourly basis, reflecting the system's macro-planning for energy transfer within the next 24 hours; the power correction coefficient is calculated in real time by the second intelligent agent on a second-by-second basis, reflecting the system's immediate response to current wind and solar power output fluctuations, state of charge tracking errors, and bus voltage deviations. Since the range of the power correction coefficient is limited to [-0.2, 0.2], the correction effect of the second intelligent agent is controlled within ±20% of the reference power, ensuring that short-cycle responses only make limited adjustments to the reference commands without fundamentally altering the long-cycle energy transfer intent.

[0039] This superposition method has a clear physical meaning: when wind and solar power output basically matches the prediction and the state of charge tracking is good, the correction coefficient approaches 0, and the actual power strictly follows the benchmark power sequence, realizing the energy shifting planned by the first intelligent agent; when wind and solar power output drops sharply by seconds (such as due to cloud cover), the second intelligent agent outputs a positive correction coefficient, increasing the actual discharge power or decreasing the charging power to compensate for the power gap in real time; when wind and solar power output rises sharply, the second intelligent agent outputs a negative correction coefficient, increasing the actual charging power to absorb excess power. The target energy storage charge and discharge power sequence generated after fusion is sent to the power execution unit of the energy storage converter, and finally converted into the actual charge and discharge current of the energy storage battery.

[0040] Figure 2 This is a flowchart illustrating the steps of another integrated energy system energy storage control method provided in this embodiment of the invention. See also... Figure 2 The method specifically includes the following steps: Step 201: Obtain time-of-use electricity price information, bus voltage deviation, and current wind and solar power output data and current state of charge of the wind-solar-storage combined power generation system; the current wind and solar power output data is a data sequence of the active power output of the wind turbine generator and photovoltaic array of the wind-solar-storage combined power generation system changing over time. In this embodiment of the invention, the data acquisition and initialization process is first initiated to comprehensively acquire the multi-source heterogeneous data required for the operation of the wind-solar-storage combined power generation system. Specifically, the following key information is read in real time through the data acquisition interface: time-of-use electricity price information released by the external power grid, which is usually updated on an hourly basis, reflecting the price difference between purchasing and selling electricity at different times, and providing an economic optimization basis for long-term energy management; the current state of charge reported by the energy storage converter or battery management system, which represents the current remaining capacity of the energy storage battery in percentage form, and is the core constraint variable that determines the charging and discharging capacity of the energy storage and prevents overcharging and over-discharging; the current wind and solar power output data output by the wind turbine generator and photovoltaic array in the wind-solar-storage combined power generation system, which is a time series of active power changes over time, with a sampling period of usually 1 minute, reflecting the real-time power generation capacity of wind and solar resources; and the bus voltage deviation parameter, which is the difference between the actual measured bus voltage and the rated voltage, used to characterize the power quality status on the grid side, and providing a basis for judging the voltage support requirements for short-term power response.

[0041] Step 202: Perform time series decomposition on the current wind and solar power output data to obtain the long-cycle trend component and the short-cycle fluctuation component of the current wind and solar power output data; the long-cycle trend component is the power component in the current wind and solar power output data whose rate of change is lower than a first rate threshold and whose duration is longer than a first duration; the short-cycle fluctuation component is the power component in the current wind and solar power output data whose rate of change is higher than a second threshold and whose duration is shorter than a second duration. In this embodiment of the invention, the collected current wind and solar power output data (i.e., the time series of active power output from wind turbine generators and photovoltaic arrays) is decomposed into a time series, decoupling it into a long-period trend component and a short-period fluctuation component. The long-period trend component is the power component in the current wind and solar power output data whose rate of change is lower than a first rate threshold and whose duration is longer than a first duration, typically characterized by hourly smooth changes; the short-period fluctuation component is the power component whose rate of change is higher than a second rate threshold and whose duration is shorter than a second duration, typically characterized by high-frequency noise at the second / minute level.

[0042] In some embodiments, step 202 may include the following sub-steps: Sub-step S11: Perform wavelet decomposition or moving average filtering on the current wind and solar power output data to extract the long-period trend component. Sub-step S12: Based on the long-cycle trend component and the current wind and solar power output data, determine the short-cycle fluctuation component.

[0043] In this embodiment of the invention, the collected current wind and solar power output data (time series of active power output from wind turbine generators and photovoltaic arrays) is processed by wavelet decomposition or moving average filtering to extract long-period trend components. Wavelet decomposition employs multi-resolution analysis technology, selecting appropriate wavelet basis functions and decomposition levels to progressively decompose the original signal into high-frequency detail components and low-frequency approximate components. The low-frequency approximate components represent the long-period trend components characterizing macroscopic changes. Moving average filtering uses a fixed window width to calculate a moving average, which is then used as the long-period trend component. Regardless of whether wavelet decomposition or moving average filtering is used, the extracted long-period trend component corresponds to the power component in the wind and solar power output data whose rate of change is lower than a first rate threshold and whose duration is longer than a first duration. Typical characteristics include hourly smooth changes, reflecting macroscopic patterns such as sunrise and sunset, and weather evolution. This component is input into the first intelligent agent as the core basis for formulating its charging and discharging "outline plan."

[0044] Based on the long-cycle trend component, the short-cycle fluctuation component is determined using this long-cycle trend component and the original wind and solar power output data. Specifically, the short-cycle fluctuation component is obtained by subtracting the long-cycle trend component from the original wind and solar power output data, i.e.: Short-cycle fluctuation component = Current wind and solar power output data - Long-cycle trend component. The physical meaning of this difference is the high-frequency residual part remaining in the original signal after removing the macro trend. It corresponds to the power component in the wind and solar power output data whose rate of change is higher than the second rate threshold and whose duration is shorter than the second duration. Its typical characteristic is high-frequency noise at the second / minute level, reflecting the instantaneous power fluctuation caused by micro-disturbance factors such as rapid cloud movement, atmospheric turbulence, and wake effects. To ensure the physical rationality of the decomposition results, the system can further perform zero-mean processing on the short-cycle fluctuation component (i.e., subtract its mean), causing it to fluctuate around zero, representing the instantaneous deviation of power relative to the trend component. At the same time, to prevent misjudgment caused by noise amplification, an energy threshold can be set, retaining only the fluctuation component with an amplitude exceeding the threshold, and ignoring the small fluctuations below the threshold as measurement noise.

[0045] Step 203: Invoke the first intelligent agent to determine the target state of charge and charging / discharging reference power sequence based on the time-of-use electricity price information, the current state of charge, and the long-cycle trend component, using a first preset time scale as the control period; the first preset time scale and the long-cycle trend component are periodically matched. In this embodiment of the invention, a first intelligent agent (long-cycle energy management module) that has been trained is invoked. Using a first preset time scale as the control period, and based on time-of-use electricity price information, the current state of charge (SOC), and the decomposed long-cycle trend components, the target SOC trajectory and the charging / discharging reference power sequence are determined for a future period. Predicted wind and solar power output data for the next first preset time period (e.g., the next 24 hours) are obtained. This predicted data, along with the long-cycle trend components, reflects the macroscopic changes in wind and solar resources over the future period. Subsequently, the first intelligent agent uses time-of-use electricity price information (reflecting the differences in electricity purchase and sale prices at different times), the current SOC (characterizing the current remaining capacity of energy storage), and the long-cycle trend components as input states, and performs a forward inference calculation through its internal deep neural network. The first intelligent agent employs a deep deterministic policy gradient algorithm, whose Actor network directly outputs continuous action values, i.e., the energy storage charging / discharging reference power sequence for each hour (24 time points) within the next 24 hours, while simultaneously planning the corresponding energy storage SOC target trajectory.

[0046] In some embodiments, step 203 may include the following sub-steps: Sub-step S21: Obtain the predicted wind and solar power output data within the first preset time period in the future; Sub-step S22: The first intelligent agent is invoked to perform inference based on the predicted wind and solar power output data within the first preset time period, the time-of-use electricity price information, the current state of charge and the long-term trend component, with the first preset time scale as the control period, and outputs the target state of charge and the charging and discharging reference power sequence.

[0047] The first intelligent agent operates on a cloud-based or site-level energy management server, employing a 1-hour control cycle. Its core is a pre-trained deep deterministic policy gradient network. Inputs include: a 24-hour wind and solar power forecast trend sequence, a time-of-use electricity price sequence (length 24), and the current energy storage state of charge (scalar). Outputs include: an hourly reference charge / discharge power sequence for energy storage (length 24, positive values ​​for discharging, negative values ​​for charging). This sequence, after smoothing and filtering, is converted into a continuous reference power curve and sent to the local controller.

[0048] In this embodiment of the invention, predicted wind and solar power output data for a first preset time period is obtained from the prediction model. The first preset time period is typically set to 24 hours, matching the daily periodicity of wind and solar power output and the daily fluctuation period of time-of-use electricity prices. The predicted wind and solar power output data includes the predicted output of photovoltaic arrays and wind turbine generators at each time point within the next 24 hours, and the sum of the two yields the total predicted wind and solar power output sequence. This prediction data can be obtained based on physical methods of meteorological satellite cloud images and numerical weather prediction, time series prediction based on historical data and machine learning (such as LSTM, Transformer, etc.), or a combination of the above methods. The long-term trend component is a macroscopic regularity feature extracted from historical data, reflecting the typical daily variation pattern of wind and solar power output; the predicted wind and solar power output data is an estimate of the actual output in the future, reflecting the specific weather conditions and resource status on a specific future date.

[0049] After acquiring the predicted wind and solar power output data for the next 24 hours, the trained first agent is invoked to perform forward inference with a first preset time scale (typically 1 hour) as the control period. The input state space of the first agent includes: the predicted wind and solar power output data for the next 24 hours, time-of-use electricity price information, the current state of charge, and the long-term trend components obtained from the decomposition. The first agent internally employs a deep deterministic policy gradient algorithm, and its Actor network maps these input states into continuous action outputs: the hourly charging and discharging baseline power sequence for each hour in the next 24 hours (positive values ​​represent discharging, negative values ​​represent charging, and the value range is between -0.5C and 0.5C), and the corresponding energy storage state of charge target trajectory (reflecting the desired SOC change path, usually between 0.1 and 0.9). During the inference process, the first agent simultaneously considers maximizing energy transfer efficiency (charging during periods of surplus photovoltaic power and discharging during periods of insufficient wind power) and maximizing economic benefits (charging during periods of low electricity prices and discharging during periods of high electricity prices), and outputs a Pareto optimal charging and discharging scheduling scheme. After inference is completed, the first agent transmits the output target state of charge trajectory and charge / discharge reference power sequence to the second agent in the form of instructions, as the reference for its subsequent real-time correction.

[0050] Step 204: Invoke the second intelligent agent to determine the power correction coefficient based on the current wind and solar power output data, the current state of charge, the target state of charge, the bus voltage deviation, and the short-period fluctuation component, using the second preset time scale as the control period; the second preset time scale is matched with the period of the short-period fluctuation component. In this embodiment of the invention, a trained second agent is invoked, with a second preset time scale as the control period. Based on the current wind and solar power output data, the current state of charge (SOC), the target SOC trajectory issued by the first agent, the bus voltage deviation, and the decomposed short-period fluctuation components, the power correction coefficient is determined in real time. The input data is processed in real time to calculate the real-time deviation between the current and predicted wind and solar power output data (reflecting the real-time fluctuation amplitude), the real-time deviation between the current SOC and the target SOC trajectory issued by the first agent (reflecting the SOC tracking error), and simultaneously acquire the real-time bus voltage deviation (reflecting the power quality of the power grid) and the short-period fluctuation components (reflecting the high-frequency noise characteristics of wind and solar power output). Subsequently, the second agent inputs these state variables into its internal deep neural network for a forward inference calculation, directly outputting continuous power correction coefficients.

[0051] In some embodiments, step 204 may include the following sub-steps: Sub-step S31: Determine the deviation between the current wind and solar power output data and the predicted wind and solar power output data; Sub-step S32: Determine the deviation between the current state of charge and the target state of charge; Sub-step S33: Invoke the second intelligent agent with the second preset time scale as the control period, and infer based on the deviation between the current wind and solar power output data and the predicted wind and solar power output data, the deviation between the current state of charge and the target state of charge, the bus voltage deviation and the short-period fluctuation component, and output the power correction coefficient.

[0052] The second intelligent agent operates within the local controller of the energy storage converter, employing a 1-second control cycle. Its core is a pre-trained dual-delay deep deterministic policy gradient network. Inputs include: the current reference power issued over a long period, the deviation between the measured total wind and solar power output and the predicted value, the deviation between the energy storage SOC and the target SOC trajectory, and the grid connection point voltage deviation. Outputs include: a power correction coefficient (range [-0.2, 0.2]).

[0053] In this embodiment of the invention, the real-time deviation between the current wind and solar power output data and the corresponding predicted wind and solar power output data is calculated. This deviation reflects the instantaneous degree of deviation between the actual wind and solar power output and the predicted value, and is a key indicator characterizing whether the system is affected by sudden disturbances such as cloud cover, wind speed turbulence, etc. When the deviation is a large negative value (actual output is much lower than the predicted value), it indicates that a power drop event has occurred, and the second intelligent agent should output a positive correction coefficient to increase discharge or reduce charging to compensate for the power gap; when the deviation is a large positive value (actual output is much higher than the predicted value), it indicates that a power surge event has occurred, and the second intelligent agent should output a negative correction coefficient to increase charging or reduce discharge to absorb the excess power. This deviation term is the core input for the second intelligent agent to achieve the fluctuation smoothing function.

[0054] Simultaneously, the real-time deviation between the current state of charge (SOC) and the target SOC trajectory issued by the first agent is calculated. This deviation reflects the degree of deviation of the actual remaining energy storage capacity from the expected planned path and is a key indicator of whether the long-term energy management plan is being executed accurately. When the SOC deviation is positive (actual SOC is higher than the target value), it indicates that the stored energy capacity is too high, and the second agent can appropriately bias towards discharging to return to the target. When the SOC deviation is negative (actual SOC is lower than the target value), it indicates that the stored energy capacity is too low, and the second agent should appropriately bias towards charging. If the absolute value of the SOC deviation exceeds a preset deviation threshold (e.g., ±5%), the system will trigger the SOC correction mode. The second agent will automatically adjust its decision weights and prioritize outputting correction instructions to bring the SOC back to the target trajectory until the deviation is eliminated.

[0055] After calculating the two deviation terms mentioned above, the wind and solar power output deviation, charge deviation, real-time bus voltage deviation, and the decomposed short-cycle fluctuation components are used as the state input. The trained second agent is then invoked to perform forward inference with a second preset time scale (typically 1 second) as the control period. The second agent employs a dual-delay deep deterministic policy gradient algorithm, and its Actor network maps these state variables into continuous power correction coefficient outputs. This correction coefficient is a multiplicative factor, with a value range limited to [-0.2, 0.2], and acts on the charging and discharging reference power issued by the first agent. The second agent's inference process is executed once per second, ensuring that the system can respond to instantaneous fluctuations in wind and solar power output on a second-level time scale. During inference, the second agent simultaneously considers multiple optimization objectives: minimizing grid-connected power fluctuations (guided by Δ wind and solar power output deviation and short-cycle fluctuation components), maintaining SOC tracking error within a reasonable range (guided by charge deviation), and supporting bus voltage stability (guided by bus voltage deviation).

[0056] Step 205: If the deviation between the current state of charge and the target state of charge exceeds a preset deviation threshold, the second agent is invoked to output a state of charge correction instruction in order to make the deviation between the current state of charge and the target state of charge less than or equal to the preset deviation threshold.

[0057] In this embodiment of the invention, the deviation between the current state of charge (SOC) and the target SOC is monitored in real time, and its absolute value is calculated. When the absolute value exceeds a preset deviation threshold, it is determined that the energy storage charge state has significantly deviated from the planned path, and the SOC correction mode is automatically triggered. This deviation may be caused by the following reasons: the cumulative effect of wind and solar ultra-short-term forecast errors (such as actual sunshine being consistently stronger than predicted, resulting in insufficient charging), unplanned charging and discharging caused by load fluctuations, or passive power regulation caused by grid commands. Once the correction mode is triggered, the decision logic of the second intelligent agent switches from "mainly focusing on fluctuation mitigation" to "mainly focusing on SOC regression," prioritizing the output of correction commands to bring the SOC back to the target trajectory.

[0058] In correction mode, the second agent automatically adjusts its internal decision weight allocation. Specifically, the coupled reward function or state-action mapping of the second agent is dynamically corrected: the weight coefficient of the state of charge tracking deviation term is significantly increased, while the weights of the wind and solar power output deviation term and the bus voltage deviation term are correspondingly reduced. This means that when outputting power correction coefficients, the second agent will prioritize "how to bring the state of charge back to the target trajectory" rather than "how to smooth out the current power fluctuations." For example, when the state of charge is low (actual value is more than 5% lower than the target value), even if the current wind and solar power output is normal or even slightly excessive, the second agent will prioritize outputting a negative correction coefficient (biased towards charging), instructing the energy storage to increase charging or reduce discharging to make up for the power gap; when the state of charge is high (actual value is more than 5% higher than the target value), even if the current wind and solar power output is normal or slightly insufficient, the second agent will prioritize outputting a positive correction coefficient (biased towards discharging), instructing the energy storage to increase discharging or reduce charging to release excess power. During this process, the second agent will still try its best to balance fluctuation smoothing and voltage support while prioritizing the correction of the state of charge, but the correction objective will take the lead.

[0059] When the deviation gradually decreases and falls back to within the preset deviation threshold, the correction target is determined to have been achieved, and the second agent automatically exits the correction mode and returns to normal operation mode. After returning to normal mode, the decision weights of the second agent are restored to their original configuration, and fluctuation smoothing becomes the primary optimization objective again. To prevent control jitter caused by frequent mode switching, the system can set a hysteresis interval (e.g., trigger threshold ±5%, exit threshold ±3%), meaning that the correction mode is only entered when the deviation exceeds 5% and exited when it falls below 3%, avoiding repeated switching at the threshold boundaries. In addition, if the correction mode continues for a long time (e.g., more than 2 hours) and still cannot pull the state of charge back to the target trajectory, the system can issue a warning to the upper layer, indicating that there may be a prediction model deviation or equipment failure, and trigger the rolling optimization mechanism of the long-cycle agent to replan the subsequent target trajectory of the state of charge.

[0060] Step 206: Generate the target energy storage charge and discharge power sequence based on the power correction coefficient and the charge and discharge reference power sequence.

[0061] In this embodiment of the invention, the charging and discharging reference power sequence determined by the first intelligent agent is fused with the power correction coefficient output in real time by the second intelligent agent to generate the final target energy storage charging and discharging power sequence, which serves as the actual control command for the energy storage converter. The power correction coefficient is applied to the reference power in a multiplicative superposition manner.

[0062] In some embodiments, step 206 may include the following sub-steps: Sub-step S41: When the deviation between the current wind and solar power output data and the predicted wind and solar power output data is less than or equal to a preset deviation threshold, the charging and discharging reference power sequence is used as the target energy storage charging and discharging power sequence. Sub-step S42: When the deviation between the current wind and solar power output data and the predicted wind and solar power output data is greater than the preset deviation threshold, the power correction coefficient is superimposed on the charging and discharging reference power sequence to generate the target energy storage charging and discharging power sequence.

[0063] In this embodiment of the invention, when the absolute value of the deviation between the current wind and solar power output data and the predicted wind and solar power output data is less than or equal to a preset deviation threshold, it is determined that the current wind and solar power output is basically consistent with the prediction, and there is no significant power fluctuation that needs to be compensated. In this case, the power correction coefficient output by the second intelligent agent approaches 0 (or although it has a small value, it is not enough to have a substantial impact), and the charging and discharging reference power sequence issued by the first intelligent agent is directly used as the target energy storage charging and discharging power sequence. When the wind and solar power output is stable and no correction is needed, the superposition calculation of the correction coefficient is skipped (or although it has been calculated, it is not actually superimposed), so that the energy storage strictly follows the energy transfer plan planned by the long-term intelligent agent, which saves computing resources and avoids command jitter caused by small noise.

[0064] When the absolute value of the deviation between the current wind and solar power output data and the predicted wind and solar power output data exceeds a preset deviation threshold, a significant power fluctuation event is determined to have occurred (such as a power drop caused by rapid cloud cover or a power surge caused by a sudden increase in wind speed). In this case, corrections from a second intelligent agent are needed to smooth out the fluctuations. The power correction coefficients output in real-time by the second intelligent agent are multiplicatively added to the charging and discharging reference power sequence to generate the target energy storage charging and discharging power sequence. When the actual output is lower than the predicted value (negative deviation, power drop), the correction coefficient is usually positive, increasing the actual discharge power or decreasing the charging power to compensate for the power shortfall. When the actual output is higher than the predicted value (positive deviation, power surge), the correction coefficient is usually negative, increasing the actual charging power or decreasing the discharge power to absorb excess power. Real-time power correction smooths out power fluctuations at the grid connection point, ensuring that the power delivered to the grid is stable and controllable.

[0065] Step 207: Obtain the actual operating results of executing the target energy storage charging and discharging power sequence to update the historical wind and solar power output dataset; Step 208: Based on the updated wind and solar power output dataset, optimize the training of the first and second agents, and evaluate the performance of the optimized first and second agents with the current first and second agents. Step 209: If the performance of the first and second agents after optimization training is better than the performance of the currently deployed first and second agents, then update the current first and second agents.

[0066] In this embodiment of the invention, after executing the target energy storage charging and discharging power sequence, actual operating result data is acquired, and the historical wind and solar power output dataset is updated accordingly. The actual operating results include: actual wind and solar power output data (the actual active power output of wind turbine generators and photovoltaic arrays), actual state of charge (SOC) change trajectory (the SOC change curve of the energy storage battery during charging and discharging), actual grid connection point power fluctuation rate (reflecting the stability of the power transmitted from the system to the grid), and actual peak-valley arbitrage revenue and wind and solar curtailment (reflecting the energy transfer effect and economic efficiency). These actual operating data, after data cleaning (removing outliers and supplementing missing values) and normalization, are added to the historical wind and solar power output dataset, continuously enriching and expanding the dataset over time.

[0067] Based on the updated wind and solar power output dataset, the first and second agents are optimized and trained offline or semi-offline. The optimized models are then compared with the currently deployed models for comprehensive performance evaluation. Specific optimization training methods include: periodically (e.g., weekly or monthly) incrementally training or fully retraining the original models using accumulated new data; using the same coupling reward function and constraints as the initial training to ensure consistency in the optimization objectives of the new and old models. Constraints include upper and lower limits for energy storage state of charge (typically 0.1~0.9); energy storage charging and discharging power limits (not exceeding 0.5C); a baseline power change rate constraint for long-cycle agents (not exceeding 0.2C per hour); and a correction coefficient limit for short-cycle agents (±0.2). After training, the optimized models are compared with the currently deployed models on a validation set. Evaluation metrics include, but are not limited to: comprehensive reward value (total score of the coupled reward function), energy transfer efficiency (ratio of actual wind and solar power consumption to theoretically generated total power), power smoothing effect (root mean square value of the power change rate at the grid connection point), state of charge tracking accuracy (mean absolute error between the actual state of charge and the target state of charge trajectory), and equivalent charge-discharge cycles (reflecting the degree of battery life degradation).

[0068] If the combined performance of the first and second agents after optimization training is better than the current deployment version, the system performs a model update operation: loading the optimized model parameters into the cloud scheduling platform (replacing the first agent) and the local energy storage controller (replacing the second agent), enabling the new model to run online. This achieves continuous learning and adaptive evolution of the energy storage control strategy, maintaining optimal control performance over a long period and adapting to complex dynamic environments such as seasonal changes in wind and solar resources, equipment aging and degradation, and adjustments in electricity pricing policies.

[0069] In some embodiments, the first agent and the second agent are trained in the following manner: Obtain the coupled reward function and historical landscape output data; The historical power output data is preprocessed, and a dataset is constructed based on the preprocessed historical power output data. The initial first agent and the initial second agent are jointly trained based on the dataset and the coupled reward function. The joint training takes the coupled reward function as the optimization objective, the target state of charge and the charge-discharge reference power sequence output by the first agent as the input constraints of the second agent, and the power correction coefficient output by the second agent adjusts the charge-discharge reference power sequence. After training is completed, the first agent and the second agent are obtained.

[0070] During the offline phase, long-cycle and short-cycle agents were jointly trained in a simulation environment using one year's worth of historical wind and solar load data. A multi-agent deep deterministic policy gradient algorithm was employed to update the network parameters. After training, the long-cycle agent was deployed on a cloud-based scheduling platform (executing once per hour), while the short-cycle agent was deployed on the local energy storage controller (with millisecond-level response). During online operation, the long-cycle agent sends a baseline power curve to the short-cycle agent every hour. The short-cycle agent operates independently without real-time communication with the cloud, ensuring control reliability.

[0071] In a simulation environment, using one year's worth of historical wind load data, long-cycle agents (Actor networks), short-cycle agents (Actor networks), and their respective Critic networks were jointly trained. The training algorithm employed a multi-agent deep deterministic policy gradient. Specifically, each training epoch simulated 24 hours of operation, with time steps of 1 hour (long-cycle step) and 1 second (short-cycle step nesting). During training, the Critic network acquired the global state and the actions of all agents, while the Actor networks could only acquire local observations during execution. After approximately 100,000 training epochs, the Critic network's Q-value converged, and the model parameters of both Actor networks were saved.

[0072] In this embodiment of the invention, a pre-designed coupled reward function and a historical wind and solar power output dataset for training are obtained. The coupled reward function is the core optimization objective of the joint training, and its expression is R = R1 + R2 + R3, where: R1 is the time-shifting efficiency reward, used to encourage the first agent to move energy at different time periods, specifically including a wind and solar curtailment penalty and a peak-valley arbitrage benefit; R2 is the power smoothing reward, used to encourage the second agent to smooth power fluctuations at the second / minute level, with the reciprocal of the root mean square value of the power change rate at the grid connection point as the indicator; R3 is the lifetime decay penalty, used to suppress excessive charging and discharging switching of the energy storage system, applying a negative reward based on the equivalent charging and discharging number model. The historical wind and solar power output data includes the wind turbine output sequence, photovoltaic array output sequence, load power sequence, energy storage state of charge sequence, and corresponding time-of-use electricity price information of the wind-solar-storage combined power generation system over a period of time (e.g., one year).

[0073] Comprehensive preprocessing was performed on the collected historical wind and solar power output data to improve data quality and training efficiency. The preprocessing steps included: data cleaning, removing outliers caused by sensor malfunctions, communication interruptions, etc., and filling them with interpolation; normalization, scaling all types of data to the [0,1] or [-1,1] interval to eliminate the influence of different units on neural network training; time series decomposition, performing wavelet decomposition or moving average filtering on the wind and solar power output data to extract long-cycle trend components and short-cycle fluctuation components, which were used as feature inputs for the first and second agents, respectively; and sample construction, using a sliding window method to divide the continuous time series into multiple training samples. Each sample contained the predicted wind and solar power output, time-of-use electricity price, state of charge, and corresponding long-cycle / short-cycle components for the next 24 hours. The sample label was the corresponding optimal control action (obtainable through interaction with the simulation environment and reward function). After preprocessing, the dataset was divided into training, validation, and test sets for model training, hyperparameter tuning, and final performance evaluation, respectively.

[0074] In a simulation environment, the first and second agents are jointly trained using a multi-agent deep deterministic policy gradient framework. The core feature of the joint training lies in the interactive coupling between the two agents: in each training round (simulating 24 hours of operation), the first agent outputs a target state of charge and a charging / discharging reference power sequence at intervals of 1 hour, based on the current state (predicted wind and solar power output, time-of-use electricity price, current state of charge, and long-cycle trend components); this reference power sequence is passed to the second agent as an input constraint; the second agent, at intervals of 1 second, outputs a power correction coefficient based on wind and solar power output deviation, state of charge tracking deviation, bus voltage deviation, and short-cycle fluctuation components, multiplicatively adjusting the reference power to generate the actual charging / discharging power, which is then applied to the simulation environment. After the environment executes an action, an immediate reward value is calculated according to the coupled reward function, and the reward is fed back to both agents to update their network parameters. The Critic network can acquire the global state and the actions of all agents during training, while the Actor network can only acquire local observations during execution. Through iterative training, the first intelligent agent learns to plan a macro-schedule strategy for charging during off-peak hours and discharging during peak hours, while the second intelligent agent learns to make minor adjustments based on the baseline power to smooth out instantaneous fluctuations.

[0075] After joint training is completed, the system saves the network parameters of the first and second agents (including the weights and biases of the Actor network) and generates a model that can be deployed online.

[0076] In some embodiments, the coupled reward function is obtained by weighted summation of time-shifting efficiency reward, power smoothing reward, and lifetime decay penalty; the time-shifting efficiency reward is determined based on the energy transfer efficiency and peak-valley arbitrage revenue within each scheduling cycle; the power smoothing reward is determined based on the root mean square value of the rate of change of grid-connected power within the evaluation time interval of the power smoothing effect; and the lifetime decay penalty is determined based on the equivalent number of charge-discharge cycles.

[0077] The coupled reward function is: Total Reward = R1 + R2 + R3. R1 rewards the successful shifting of long-cycle periods (penalties for wind and solar curtailment, and peak-valley arbitrage gains); R2 rewards the suppression of power fluctuations in short-cycle periods (indicated by the power change rate at the grid connection point); R3 penalizes frequent charging and discharging based on an equivalent charge-discharge cycle model to extend battery life. Energy transfer efficiency reward R1: After each scheduling cycle (24 hours), the ratio of the actual absorbed total wind and solar power to the theoretically generated total power is calculated and multiplied by a coefficient of 0.5; simultaneously, peak-valley arbitrage gains (discharge revenue - charging cost) are calculated and multiplied by a coefficient of 0.3. The sum of these two amounts constitutes R1. Power smoothing reward R2: The root mean square value of the power change rate at the grid connection point is calculated every 15 minutes, and its reciprocal is multiplied by a coefficient of 0.2 as a reward to encourage the system to deliver power to the grid at a stable power level. Lifetime decay penalty R3: Based on the real-time monitored equivalent charge-discharge cycles (considering depth of discharge and temperature correction), a negative reward value (e.g., -0.05) is given for each equivalent full charge-discharge cycle to suppress unnecessary frequent switching.

[0078] In this embodiment of the invention, a coupled reward function is used to train the core optimization objectives of the first agent (long-cycle energy management module) and the second agent (short-cycle power response module). This function achieves a unified quantitative evaluation of three interrelated and even conflicting objectives: "energy transfer effect," "fluctuation smoothing effect," and "battery lifespan protection," by weighted summing of time-shifting efficiency reward, power smoothing reward, and lifetime degradation penalty. The time-shifting efficiency reward R1 evaluates the effect of the first agent in achieving energy transfer over a long-cycle timescale, serving as the core incentive signal guiding the energy storage system to charge during periods of photovoltaic surplus and discharge during periods of insufficient wind power. The power smoothing reward R2 evaluates the effect of the second agent in smoothing power fluctuations at the grid connection point over a short-cycle timescale, serving as the core incentive signal guiding the system to deliver stable power to the grid. The lifetime degradation penalty R3 suppresses frequent charge-discharge switching of the energy storage system, serving as the core penalty signal guiding the system to extend battery lifespan.

[0079] Reference Figure 3 The diagram shows a logic block diagram of an energy storage control method for an integrated energy system provided by an embodiment of the present invention. Figure 3The core control flow of this invention is described as follows: First, time-of-use electricity price information, bus voltage deviation, current wind and solar power output data, and current state of charge are acquired. The current wind and solar power output data is then decomposed into a time series to obtain long-term trend components and short-term fluctuation components. Subsequently, a first and second intelligent agent, which have been trained in parallel, are invoked. The first intelligent agent takes the predicted wind and solar power output data, time-of-use electricity price information, current state of charge, and long-term trend components within a first preset time period as input and outputs the target state of charge and the charging and discharging reference power sequence. The second intelligent agent takes the deviation between the current wind and solar power output data and the predicted wind and solar power output data, the deviation between the current state of charge and the target state of charge, the bus voltage deviation, and the short-term fluctuation components as input and outputs a power correction coefficient. Finally, the power correction coefficient is fused with the charging and discharging reference power sequence to generate the target energy storage charging and discharging power sequence, thereby realizing the dual-time-scale collaborative control of the energy storage system.

[0080] It should be noted that, for the sake of simplicity, the method embodiments are described as a series of actions. However, those skilled in the art should understand that the embodiments of the present invention are not limited to the described order of actions, because according to the embodiments of the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions involved are not necessarily essential to the embodiments of the present invention.

[0081] Figure 4 This is a structural block diagram of an energy storage control device for an integrated energy system provided in an embodiment of the present invention. (Refer to...) Figure 4 The device specifically includes the following modules: The parameter acquisition module 301 is used to acquire time-of-use electricity price information, bus voltage deviation, and the current wind and solar power output data and current state of charge of the wind-solar-storage combined power generation system; the current wind and solar power output data is a data sequence of the active power output of the wind turbine generator and photovoltaic array of the wind-solar-storage combined power generation system changing over time. The time-series decomposition module 302 is used to perform time-series decomposition on the current wind and solar power output data to obtain the long-cycle trend component and the short-cycle fluctuation component of the current wind and solar power output data; the long-cycle trend component is the power component in the current wind and solar power output data whose rate of change is lower than a first rate threshold and whose duration is longer than a first duration; the short-cycle fluctuation component is the power component in the current wind and solar power output data whose rate of change is higher than a second threshold and whose duration is shorter than a second duration. The power sequence determination module 303 is used to call a first intelligent agent to determine the target state of charge and the charging and discharging reference power sequence based on the time-of-charge price information, the current state of charge and the long-period trend component, with a first preset time scale as the control period; the first preset time scale and the long-period trend component are periodically matched. The correction coefficient determination module 304 is used to call the second intelligent agent to determine the power correction coefficient based on the current wind and solar power output data, the current state of charge, the target state of charge, the bus voltage deviation, and the short-period fluctuation component, with the second preset time scale as the control period; the second preset time scale is matched with the period of the short-period fluctuation component. The target power sequence generation module 305 is used to generate a target energy storage charge and discharge power sequence based on the power correction coefficient and the charge and discharge reference power sequence.

[0082] In some embodiments, the time-series decomposition module 302 includes: The long-period component extraction submodule is used to perform wavelet decomposition or moving average filtering on the current wind and solar power output data to extract the long-period trend component. The short-cycle component determination submodule is used to determine the short-cycle fluctuation component based on the long-cycle trend component and the current wind and solar power output data.

[0083] In some embodiments, the first agent and the second agent are trained using the following modules: The historical parameter acquisition module is used to acquire the coupled reward function and historical landscape output data; The dataset construction module is used to preprocess the historical landscape power output data and construct a dataset based on the preprocessed historical landscape power output data. The agent training module is used to jointly train the constructed initial first agent and initial second agent based on the dataset and the coupled reward function; the joint training takes the coupled reward function as the optimization objective, the target state of charge and the charge-discharge reference power sequence output by the first agent are used as the input constraints of the second agent, and the power correction coefficient output by the second agent is used to adjust the charge-discharge reference power sequence; The agent determination module is used to obtain the first agent and the second agent after training is completed.

[0084] In some embodiments, the coupled reward function is obtained by weighted summation of time-shifting efficiency reward, power smoothing reward, and lifetime decay penalty; the time-shifting efficiency reward is determined based on the energy transfer efficiency and peak-valley arbitrage revenue within each scheduling cycle; the power smoothing reward is determined based on the root mean square value of the rate of change of grid-connected power within the evaluation time interval of the power smoothing effect; and the lifetime decay penalty is determined based on the equivalent number of charge-discharge cycles.

[0085] In some embodiments, the power sequence determination module 303 includes: The predictive data acquisition submodule is used to acquire the predicted wind and solar power output data within the first preset time period in the future; The power sequence output submodule is used to call the first intelligent agent to perform inference based on the predicted wind and solar power output data within the first preset time period, the time-of-use electricity price information, the current state of charge and the long-term trend component, and output the target state of charge and the charging and discharging reference power sequence, with the first preset time scale as the control period.

[0086] In some embodiments, the correction coefficient determination module 304 includes: The wind and solar power output data deviation determination submodule is used to determine the deviation between the current wind and solar power output data and the predicted wind and solar power output data; The state of charge deviation determination submodule is used to determine the deviation between the current state of charge and the target state of charge; The correction coefficient output submodule is used to call the second intelligent agent to perform inference based on the deviation between the current wind and solar power output data and the predicted wind and solar power output data, the deviation between the current state of charge and the target state of charge, the bus voltage deviation, and the short-period fluctuation component, and output the power correction coefficient.

[0087] In some embodiments, the target power sequence generation module 305 includes: The target power sequence determination submodule is used to take the charging and discharging reference power sequence as the target energy storage charging and discharging power sequence when the deviation between the current wind and solar power output data and the predicted wind and solar power output data is less than or equal to a preset deviation threshold; and to add the power correction coefficient to the charging and discharging reference power sequence to generate the target energy storage charging and discharging power sequence when the deviation between the current wind and solar power output data and the predicted wind and solar power output data is greater than the preset deviation threshold.

[0088] In some embodiments, before generating the target energy storage charge / discharge power sequence, the method further includes: The state of charge correction module is used to call the second intelligent agent to output a state of charge correction instruction first if the deviation between the current state of charge and the target state of charge exceeds a preset deviation threshold, so that the deviation between the current state of charge and the target state of charge is less than or equal to the preset deviation threshold.

[0089] In some embodiments, the apparatus further includes: The result acquisition module is used to acquire the actual operation results of executing the target energy storage charging and discharging power sequence in order to update the historical wind and solar power output dataset; The agent evaluation module is used to optimize and train the first agent and the second agent based on the updated wind and solar power output dataset, and to evaluate the performance of the optimized first agent and the second agent with the current first agent and the second agent. The agent update module is used to update the current first and second agents if the performance of the optimized and trained first and second agents is better than the performance of the currently deployed first and second agents.

[0090] As the apparatus embodiment is basically similar to the method embodiment, it is described in a relatively simple manner. For relevant details, please refer to the description of the method embodiment.

[0091] This invention also provides an electronic device, including: a processor, a memory, and a computer program stored in the memory and capable of running on the processor. When the computer program is executed by the processor, it implements the various processes of the above-described integrated energy system energy storage control method embodiments and achieves the same technical effects. To avoid repetition, it will not be described again here.

[0092] This invention also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the various processes of the above-described embodiments of the energy storage control method for the integrated energy system and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. An energy storage control method for an integrated energy system, characterized in that, The method includes: The system acquires time-of-use electricity price information, bus voltage deviation, and current wind and solar power output data and current state of charge of the wind-solar-storage combined power generation system; the current wind and solar power output data is a data sequence of the active power output of the wind turbine generator and photovoltaic array of the wind-solar-storage combined power generation system over time. The current wind and solar power output data is decomposed into a time series to obtain the long-term trend component and the short-term fluctuation component of the current wind and solar power output data; the long-term trend component is the power component in the current wind and solar power output data whose rate of change is lower than a first rate threshold and whose duration is longer than a first duration; the short-term fluctuation component is the power component in the current wind and solar power output data whose rate of change is higher than a second threshold and whose duration is shorter than a second duration. The first intelligent agent is invoked with a first preset time scale as the control period, and based on the time-of-use electricity price information, the current state of charge, and the long-period trend component, the target state of charge and the charging / discharging reference power sequence are determined; the first preset time scale and the long-period trend component are periodically matched. The second intelligent agent is invoked with a second preset time scale as the control period, and a power correction coefficient is determined based on the current wind and solar power output data, the current state of charge, the target state of charge, the bus voltage deviation, and the short-period fluctuation component; the second preset time scale is matched with the period of the short-period fluctuation component. Based on the power correction coefficient and the charge / discharge reference power sequence, a target energy storage charge / discharge power sequence is generated.

2. The energy storage control method for an integrated energy system according to claim 1, characterized in that, The step of performing time series decomposition on the current wind and solar power output data to obtain the long-term trend component and short-term fluctuation component of the current wind and solar power output data includes: The long-period trend component is extracted by performing wavelet decomposition or moving average filtering on the current wind and solar power output data. Based on the long-cycle trend component and the current wind and solar power output data, the short-cycle fluctuation component is determined.

3. The energy storage control method for an integrated energy system according to claim 1, characterized in that, The first and second agents were trained in the following manner: Obtain the coupled reward function and historical landscape output data; The historical power output data is preprocessed, and a dataset is constructed based on the preprocessed historical power output data. The initial first agent and the initial second agent are jointly trained based on the dataset and the coupled reward function. The joint training takes the coupled reward function as the optimization objective, the target state of charge and the charge-discharge reference power sequence output by the first agent as the input constraints of the second agent, and the power correction coefficient output by the second agent adjusts the charge-discharge reference power sequence. After training is completed, the first agent and the second agent are obtained.

4. The energy storage control method for an integrated energy system according to claim 3, characterized in that, The coupled reward function is obtained by weighted summation of time-shifting efficiency reward, power smoothing reward, and lifetime decay penalty; the time-shifting efficiency reward is determined based on the energy transfer efficiency and peak-valley arbitrage revenue within each scheduling cycle; the power smoothing reward is determined based on the root mean square value of the rate of change of grid-connected power within the evaluation time interval of the power smoothing effect; and the lifetime decay penalty is determined based on the equivalent number of charge-discharge cycles.

5. The energy storage control method for an integrated energy system according to claim 1, characterized in that, The invocation of the first intelligent agent, using a first preset time scale as the control period, determines the target state of charge and charging / discharging reference power sequence based on the time-of-use electricity price information, the current state of charge, and the long-term trend component, including: Obtain the predicted wind and solar power output data within the first preset time period in the future; The first intelligent agent is invoked to perform inference based on the predicted wind and solar power output data within the first preset time period, the time-of-use electricity price information, the current state of charge and the long-term trend component, with the first preset time scale as the control period, and outputs the target state of charge and the charging and discharging reference power sequence.

6. The energy storage control method for an integrated energy system according to claim 5, characterized in that, The invocation of the second intelligent agent, using a second preset time scale as the control period, determines the power correction coefficient based on the current wind and solar power output data, the current state of charge, the target state of charge, the bus voltage deviation, and the short-period fluctuation component, including: Determine the deviation between the current wind and solar power output data and the predicted wind and solar power output data; Determine the deviation between the current state of charge and the target state of charge; The second intelligent agent is invoked with a second preset time scale as the control period. Based on the deviation between the current wind and solar power output data and the predicted wind and solar power output data, the deviation between the current state of charge and the target state of charge, the bus voltage deviation, and the short-period fluctuation component, inference is performed, and a power correction coefficient is output.

7. The energy storage control method for an integrated energy system according to claim 6, characterized in that, The step of generating the target energy storage charge-discharge power sequence based on the power correction coefficient and the charge-discharge reference power sequence includes: When the deviation between the current wind and solar power output data and the predicted wind and solar power output data is less than or equal to a preset deviation threshold, the charging and discharging reference power sequence is used as the target energy storage charging and discharging power sequence. When the deviation between the current wind and solar power output data and the predicted wind and solar power output data is greater than the preset deviation threshold, the power correction coefficient is superimposed on the charging and discharging reference power sequence to generate the target energy storage charging and discharging power sequence.

8. The energy storage control method for an integrated energy system according to claim 6, characterized in that, Before generating the target energy storage charge / discharge power sequence, the following steps are also included: If the deviation between the current state of charge and the target state of charge exceeds a preset deviation threshold, the second agent is invoked to output a state of charge correction instruction in order to make the deviation between the current state of charge and the target state of charge less than or equal to the preset deviation threshold.

9. The energy storage control method for an integrated energy system according to claim 1, characterized in that, The method further includes: Obtain the actual operating results of executing the target energy storage charging and discharging power sequence to update the historical wind and solar power output dataset; Based on the updated wind and solar power output dataset, the first and second agents are optimized and trained, and the performance of the optimized first and second agents is evaluated with that of the current first and second agents. If the performance of the first and second agents after optimization training is better than that of the currently deployed first and second agents, then the current first and second agents should be updated.

10. An energy storage control device for an integrated energy system, characterized in that, The device includes: The parameter acquisition module is used to acquire time-of-use electricity price information, bus voltage deviation, and the current wind and solar power output data and current state of charge of the wind-solar-storage combined power generation system; the current wind and solar power output data is a data sequence of the active power output of the wind turbine generator and photovoltaic array of the wind-solar-storage combined power generation system changing over time. The time-series decomposition module is used to perform time-series decomposition on the current wind and solar power output data to obtain the long-cycle trend component and the short-cycle fluctuation component of the current wind and solar power output data; the long-cycle trend component is the power component in the current wind and solar power output data whose rate of change is lower than a first rate threshold and whose duration is longer than a first duration; the short-cycle fluctuation component is the power component in the current wind and solar power output data whose rate of change is higher than a second threshold and whose duration is shorter than a second duration. A power sequence determination module is used to call a first intelligent agent to determine a target state of charge and a charging / discharging reference power sequence based on the time-of-charge price information, the current state of charge, and the long-period trend component, using a first preset time scale as the control period; the first preset time scale is matched with the period of the long-period trend component; The correction coefficient determination module is used to call the second intelligent agent to determine the power correction coefficient based on the current wind and solar power output data, the current state of charge, the target state of charge, the bus voltage deviation, and the short-period fluctuation component, with the second preset time scale as the control period; the second preset time scale is matched with the period of the short-period fluctuation component. The target power sequence generation module is used to generate a target energy storage charge and discharge power sequence based on the power correction coefficient and the charge and discharge reference power sequence.

11. An electronic device, characterized in that, include: A processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the energy storage control method for an integrated energy system as described in any one of claims 1-9.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the steps of the energy storage control method for the integrated energy system as described in any one of claims 1-9.

Images