Integrated energy management control system and method based on photovoltaic energy storage system

Abstract

The invention discloses an integrated energy management control system and method based on a photovoltaic energy storage system, and the system employs a three-layer four-ring management control architecture, and comprises the construction of a physical hardware layer, an intelligent edge control layer, a cloud collaborative management optimization layer, and the fusion of four closed loops of an energy flow, an information flow, a value flow and a safety flow which penetrate through the three layers. The physical hardware layer is used for hardware decoupling, fault isolation and state depth perception; the intelligent edge control layer is used for converting an optimization target of a cloud end into collaborative management and safety control on a hardware module on a set time scale; the intelligent edge control layer adopts a frame of digital twinning partner training enhanced learning, and the frame comprises an adaptive networking control module, a digital twinning simulation module and a model prediction control module; and the cloud collaborative management optimization layer is used for constructing a full life cycle optimizer and carrying out global value optimization management. According to the invention, stable management of photovoltaic energy storage energy and closed-loop collaborative control of multi-objective optimization are realized.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic energy storage management and control technology, and specifically to an integrated energy management and control system and method based on a photovoltaic energy storage system. Background Technology

[0002] In the process of managing the energy of photovoltaic energy storage, the current mainstream systems usually adopt a hierarchical control architecture. The upper layer uses model predictive control or artificial intelligence algorithms to carry out multi-timescale energy dispatch with the goals of economic optimization and grid demand response. The lower layer achieves real-time power balance and grid support through grid-connected or grid-linked converter control. In terms of hardware architecture, topologies such as AC coupling, DC coupling, and AC / DC hybrid bus coexist. At the same time, a battery management system is integrated to perform fine-grained monitoring and protection of energy storage units. With the development of aggregation forms such as virtual power plants, the system also needs to have the ability to interact with the upper-level grid dispatch center and coordinate operations.

[0003] However, at the control level, the deep integration of photovoltaics and energy storage introduces highly nonlinear and strongly coupled characteristics. This is particularly challenging when the system operates under weak grid conditions or requires grid support, as traditional control strategies face difficulties in ensuring wideband system stability, suppressing power oscillations, and achieving seamless switching between multiple operating modes. Regarding the management of key components, the lifespan degradation of battery storage units directly conflicts with system economics. Existing energy management strategies lack the ability to collaboratively optimize battery aging costs, thermal safety risks, and uneven operation among heterogeneous battery clusters. Furthermore, the system architecture itself presents an inherent contradiction between efficiency and reliability; multi-stage power electronic conversion increases losses, while highly integrated designs may expand the scope of failures. Although artificial intelligence algorithms offer new optimization approaches, they face challenges in implementing them in real-time power electronic control where safety is critical, including model reliability verification, computational resource constraints, and hardware deployment. Simultaneously, power quality issues such as harmonic pollution caused by high-proportion converter grid connection urgently require more effective suppression methods. Summary of the Invention

[0004] The purpose of this invention is to provide an integrated energy management control system and method based on a photovoltaic energy storage system to solve the problems mentioned in the background art.

[0005] The specific technical solution provided by this invention is as follows: an integrated energy management and control system and method based on photovoltaic energy storage system. The system adopts a "three-layer four-ring" management and control architecture, including the construction of a physical hardware layer, an intelligent edge control layer, a cloud-based collaborative management and optimization layer, and four closed loops that integrate and run through the three layers: energy flow, information flow, value flow, and security flow.

[0006] Preferably, the physical hardware layer is used for hardware decoupling, fault isolation, and deep state awareness. The physical hardware layer also includes: Standardized power modules: Each module has a built-in unified "modular controller", equipped with a multi-core processor and running a lightweight real-time operating system; It features a programmable power topology: the main power section uses general-purpose power units; Built-in miniature synchronous phasor measurement unit: Each module integrates an ADC chip to synchronously sample the input / output of this module at a set rate and timestamp it to provide measurement data to the upper layer; Distributed intelligent sensor network: Employing cell-level fusion sensing and magnetic-acoustic joint monitoring; Deterministic communication network: All sensor data is aggregated through a time-sensitive network switch; Hardware discovery and identification: When a standardized module is inserted, the "presence detection" pin of its signal contact is turned on first. The hardware manager reads the module's unique electronic ID, which contains the module type, rated parameters, serial number, and software version, through the I2C bus. The manager then registers the new module with the "digital twin emulator" in the edge control layer via Ethernet. Automatic topology reconstruction and parameter tuning: After receiving information about a new module, the digital twin simulator automatically adds the corresponding virtual module to its virtual model and calls the pre-stored detailed parameter model of that model. At the same time, the digital twin simulator recalculates the control parameters of the entire system online according to the new system topology and compiles and generates new control code. After passing the security verification, it is dynamically loaded into the corresponding module controller and the upper-level adaptive network controller.

[0007] Preferably, the intelligent edge control layer is used to transform the optimization goals in the cloud into collaborative management and security control of hardware modules at a set time scale; the intelligent edge control layer adopts a framework of digital twin training reinforcement learning, which includes: an adaptive network construction control module, a digital twin simulation module, and a model prediction control module; Preferably, the adaptive network control module enables the controller to autonomously adapt to complex and ever-changing power grid conditions; the digital twin simulation module is used to construct a multiphysics model and perform deduction and evolution; and the model predictive control module is used to embed battery aging costs and network reconfiguration capabilities into rolling optimization for joint optimization.

[0008] Preferably, the cloud-based collaborative management and optimization layer is used to build a full lifecycle optimizer for global value optimization management. The operation method of the cloud-based collaborative management and optimization layer includes: Step S1: Receive instructions through the model prediction controller; Step S2: The model predictive controller sends the preliminary power adjustment scheme to the twin in the digital twin model for pre-control simulation, and the twin in the digital twin model performs the simulation execution; Step S3: The twin in the digital twin model will feed back the warning to the model predictive controller. The model predictive controller will readjust the optimization problem, add corresponding constraints, and solve the output again. Step S4: Issue commands after optimization; Step S5: Record the process and results of this collaborative management decision-making, and use them as training data for the reinforcement learning agent and federated learning model to train for better future decisions.

[0009] Compared with the prior art, the beneficial effects achieved by the present invention are: (1) By deploying “software-defined, modular plug-and-play” power units and deep sensing networks at the physical hardware layer, the present invention realizes dynamic reconfiguration of hardware and accurate mirroring of state, solving the inherent problems of poor system scalability, easy fault propagation and insufficient state perception.

[0010] (2) This invention constructs a real-time autonomous loop of "perception-simulation-decision-verification" with "digital twin simulator" as the core in the intelligent edge control layer, and integrates an adaptive network controller and a model prediction controller with embedded aging cost, thereby realizing closed-loop collaboration of millisecond-level stable control and minute-level multi-objective optimization. This solves the problems of stability challenges under weak power grids, conflicts between battery life and short-term economic efficiency, and the difficulty of implementing advanced algorithms in safety-critical scenarios.

[0011] (3) By running a full life cycle optimizer and a blockchain-based virtual power plant coordinator in the cloud collaborative management optimization layer, this invention realizes the accurate transmission of long-term economic goals to real-time control parameters and the interaction of trusted markets, solving the problems of disconnect between global optimization goals and local control goals, as well as the trust and efficiency issues of distributed resource aggregation participating in market transactions.

[0012] (4) This invention achieves integrated linkage of physical operation, data-driven, economic decision-making and active protection by deeply coupling the physical hardware layer, intelligent edge control layer and cloud collaborative management optimization layer with four closed loops of "energy flow, information flow, value flow and security flow". It solves the complex technical challenges faced by traditional photovoltaic energy storage integrated systems in control, management, architecture and algorithm levels, such as "difficulty in highly nonlinear control", "conflict between lifespan and economy", "contradiction between efficiency and reliability" and "difficulty in implementing intelligent algorithms". It provides a complete technical architecture for building a new generation of new power system nodes with high resilience, high efficiency and high returns. Attached Figure Description

[0013] Figure 1 This is a diagram of the "three-layer, four-ring" management and control architecture adopted by the system provided in this embodiment of the invention. Detailed Implementation

[0014] 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 embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention are within the scope of protection of the present invention.

[0015] Example 1: Combination Figure 1 As shown in the figure, the integrated energy management control system and method based on photovoltaic energy storage system described in this embodiment adopts a "three-layer four-ring" management control architecture: unlike the existing technical route with layered fragmentation, tight software and hardware coupling and single optimization goal, it constructs a physical hardware layer, an intelligent edge control layer, a cloud collaborative management and optimization layer, and four closed loops of energy flow, information flow, value flow and security flow that run through the three layers, so as to realize the paradigm shift of photovoltaic energy storage system energy management control from local optimization to global optimization, and from passive response to active prevention.

[0016] In this embodiment, the physical hardware layer serves as the physical carrier of the system, used for hardware decoupling, fault isolation, and deep state awareness, fundamentally alleviating the problems of "efficiency versus reliability contradiction" and "insufficient state awareness." All power modules involved in the system (photovoltaic optimizer, battery DC / DC converter, grid-connected inverter, load converter) adopt a unified standardized process, using two-sided forward / reverse blind-plug liquid-cooled plate interfaces to automatically complete the coolant circuit connection upon module insertion into the rack, ensuring heat dissipation efficiency. Two high-current contacts, "Power+" and "Power-" (supporting up to 1500Vdc, 400A), and one multi-pin signal contact are defined. The signal contact integrates Gigabit Ethernet, CAN FD, and low-voltage power supply lines, achieving a three-in-one blind-plug connection for power, communication, and power supply.

[0017] For example, each module in the physical hardware layer has a built-in unified "Modular Controller (MC)," equipped with a high-performance multi-core processor (such as an ARM Cortex-A+R5 dual-core), running a lightweight real-time operating system, and the main power section uses a general-purpose power unit (GPU). Taking the battery DC / DC module as an example, its core is an H-bridge or three-level topology composed of SiC MOSFETs, but its control algorithm and logic can be defined by software. The same hardware module can be configured in Buck, Boost, or Buck-Boost modes by loading different software to adapt to different battery voltage ranges, achieving hardware reuse. At the same time, each module integrates a high-precision ADC chip to synchronously sample the input / output voltage and current of the module at a rate of ≥100kHz, and stamp it with a nanosecond-level timestamp based on the IEEE 1588 precision time protocol, providing the upper layer with network-wide synchronized broadband measurement data.

[0018] In this embodiment, in addition to traditional voltage and temperature sampling lines, a distributed fiber optic temperature sensing (DTS) cable and a flexible thin-film pressure sensor are pre-embedded in each battery module. The DTS provides a continuous, linear temperature distribution map (accuracy...). Pressure sensors monitor cell expansion forces, providing multi-physical fusion criteria for early thermal runaway and lithium plating warnings. High-frequency Rogowski coils and ultrasonic microphone arrays are installed at critical power buses and transformers. The Rogowski coils monitor rapid transient currents >1MHz, and the ultrasonic microphones capture specific frequency sound waves emitted by power devices or magnetic components due to localized overheating or mechanical loosening, enabling predictive maintenance. All sensor data is aggregated through a Time-Sensitive Network (TSN) switch. The TSN protocol ensures that the transmission delay of critical status information (such as overcurrent and high temperature) is strictly limited to within 100 microseconds and has the highest priority, providing a channel for extremely rapid and safe response.

[0019] For example, when a standardized module is inserted into the rack, its "presence detection" pin is activated first. The "hardware manager" inside the rack then reads the unique electronic ID pre-programmed into the module's EEPROM via the I2C bus, containing: module type (e.g., "bidirectional DC / DC, battery-powered"), rated parameters (voltage, current, power), serial number, software version, etc. The manager registers the new module with the "Digital Twin Emulator (DTS)" at the edge control layer via Ethernet. After receiving the new module information, the DTS automatically adds the corresponding virtual module to its virtual model and calls the pre-stored detailed parameter model for that model. Simultaneously, based on the new system topology (e.g., adding a new battery cluster), the DTS recalculates the entire system control parameters (e.g., loop gain, virtual impedance) online and compiles new control code. After passing security verification, this code is dynamically loaded into the corresponding module controller and the upper-layer adaptive network controller.

[0020] In this embodiment, the present invention provides a process example of physical layer collaborative operation. For example, when a slight short circuit occurs inside battery module B2, causing an abnormal rise in the current of its secondary bus, the following operations are performed: Microsecond-level sensing and isolation: The miniature synchronization phasor unit built into module B2 detects a sudden current surge exceeding the local protection threshold within 50 microseconds. B2's controller immediately blocks its own drive pulses and sends a high-priority fault event message via the TSN network to the "Solid State Power Router" and the edge layer's Digital Twin Emulator (DTS). Within 100 microseconds of receiving the message, the power router drives the corresponding SSCB's MOSFET to turn off, physically isolating the faulty secondary bus. At this time, the main system power supply remains unaffected.

[0021] Millisecond-level state assessment and reconfiguration: Upon receiving an alarm, the DTS immediately initiates a "post-fault power flow reconfiguration simulation." Based on the latest system state, it simulates the feasibility of transferring the load from the faulty cluster to adjacent healthy battery clusters C2 and C3 in digital space, verifying that this operation will not cause overload or stability issues. After the simulation passes, the DTS sends a command to the hardware manager to close the interconnection switch between clusters C2 and C3 and the main bus, and adjusts the output power of the corresponding DC / DC modules. The entire power supply reconfiguration is completed within 200 milliseconds, with the load perception consisting of only a brief voltage dip.

[0022] Subsequent analysis and early warning: Pre-failure data recorded by the pressure sensor and DTS was uploaded to the cloud analysis platform. Cloud-based large-scale model analysis revealed that a slight increase in pressure and uneven temperature distribution had already occurred inside the module before the failure. These characteristics were marked as a new "early failure fingerprint." This fingerprint model was distributed to all edge nodes through federated learning to enhance the early warning capability for similar failures.

[0023] In this embodiment, the intelligent edge control layer is the core connecting "programmable physical hardware" and "cloud intelligence." It translates cloud-based optimization goals into precise, coordinated, and secure control of hundreds of hardware modules over a set time scale. The intelligent edge control layer in this invention also includes: an adaptive network control module that uses a digital twin-based training and reinforcement learning framework to enable the controller to autonomously adapt to complex and changing power grid conditions; a digital twin simulation module that constructs and evolves a multi-physics model; and a model predictive control module that embeds battery aging costs and network reconfiguration capabilities into rolling optimization to achieve online joint optimization of economy and reliability.

[0024] For example, the operation method of the adaptive network control module includes: Step A1: The controller receives network-wide synchronization phasor data from the physical layer TSN network at a rate of 10kHz, including the voltage of each node. Current (with timestamps) Calculates the active power at the grid connection point of the system in real time using instantaneous power theory. reactive power and grid voltage amplitude and phase ; Step A2: The controller continuously injects a series of non-invasive small-signal harmonic disturbances (such as 2Hz, 5Hz, 15Hz, etc.) into the power grid and observes the system's frequency response. The Thevenin equivalent impedance of the power grid is then fitted online using the recursive least squares (RLS) method. , in, Equivalent resistance For equivalent inductance, The angular frequency is used to update the data once per second, thus forming a dynamic understanding of the power grid strength. Step A3: Construct a virtual controller replica identical to the real controller and connect it to a real-time digital twin power grid model. Place it next to the real controller and run it. The virtual controller will adjust its operation based on the current state. Output a set of tentative control parameter adjustment actions For example, virtual inertia Sag coefficient Voltage loop bandwidth Performing actions in a digital twin model This allows us to deduce the system dynamics within a future timescale.

[0025] Step A4: Construct a reward function for a multi-objective trade-off : ,in, The weights of each item are dynamically adjusted by the cloud optimizer. For frequency and voltage deviation, The large deviation is penalized for the square term. Active power provided to the power grid (such as frequency regulation) is awarded bonus points if it is active power. The total harmonic distortion to be suppressed, This provides an estimate of converter efficiency to encourage efficient operation. The virtual controller maximizes the cumulative reward. In digital twins, thousands of "trial and error" training sessions are conducted per second, learning to perform different tasks. The system automatically selects the optimal control parameters for both strong and weak power grids and different operating points.

[0026] Step A5: The trained strategy parameters are synchronized to the real main controller every few seconds. The main controller applies the parameters to the classic dual-loop vector control or virtual synchronous machine control algorithm to generate the final PWM drive signal, which is then sent to the power module of the physical layer.

[0027] For example, the operation of a digital twin simulation module includes: Step B1: Using a detailed switching model based on the nodal admittance method, each power module is modeled as a combination of a controlled current source and an RLC network. A second-order RC equivalent circuit model is coupled with a three-dimensional thermal model, including an electrical model and a thermal model. The electrical model uses a voltage source. A resistor in series and two RC parallel networks ( and To simulate a battery equivalently, and This demonstrates that after a sudden change in current, the voltage does not stabilize immediately but relaxes exponentially with time; the state equations of the electrical and thermal models are expressed as:

[0028]

[0029] in This is the battery's terminal voltage. The operating current flowing through the battery, Open circuit voltage, In a charged state, , For polarization internal resistance, , This is expressed as the polarization capacitor voltage. For heat generation rate, This refers to the temperature of the core region inside the battery cell. The surface temperature of the battery cell. For thermal resistance, For heat capacity. Input current to the electrical model. Output voltage And calculate the key heat source—ohmic loss. The power is transmitted to the thermal model, which then receives the generated heat. and surface temperature Predicting internal temperature Temperature, in turn, affects the parameters in the electrical model (such as...). This forms a closed loop.

[0030] Step B2: The twin receives real data from the physical layer sensors at a rate of 1kHz and uses unscented Kalman filtering to continuously correct key state variables in the model (such as bus voltage, battery SOC, and cell temperature) to ensure the accuracy of the virtual world and the physical world. Step B3: DTS maintains one real shadow model and multiple hypothetical extrapolation models, performing hypothetical extrapolation in parallel; for example, when a pressure sensor detects the expansion force of a certain battery cell... When a slow upward trend is observed, the simulation model is immediately activated. Starting from the current state, it assumes the system will operate as planned over the next 24 hours and accelerates the simulation. If the simulation results show the hot spot temperature of the battery cell... It will exceed the safety threshold, and the expansion force growth rate To accelerate the process, the system will issue a Level 1 maintenance warning several hours in advance and provide a suggested strategy to "reduce the charging current of this cluster".

[0031] Step B4: DTS periodically uploads model prediction errors (such as SOC estimation errors and temperature prediction errors) to the cloud. The cloud aggregates error data from multiple systems and trains an error correction supernetwork. This network is then distributed to each edge node to dynamically correct the parameters of the local twin model.

[0032] For example, the operation method of the model prediction control module includes: Step C1: Construct a multi-timescale rolling optimization framework including long-term and short-term time-scale layers. In the long-term layer, receive the 24-hour power plan and economic weights from the cloud. The short-period layer receives power commands from the long-period layer. This allows for the rapid smoothing of second-level fluctuations in photovoltaic and load conditions, as well as the handling of hardware module switching. The optimization objective in the long-cycle layer is: ,in, To predict the time domain, This is the grid interaction cost coefficient for the k-th time period, typically represented by the time-of-use electricity price. Let be the power exchanged between the system and the main grid during the k-th time period. The duration of a single time step. This is a weighting factor for battery aging costs. The cost of capacity degradation caused by battery throughput in time period k. The penalty weighting factor for power plan tracking deviation. This represents the actual power reference command value of the system grid connection point during the k-th time period. This represents the long-term power plan value issued by the cloud or dispatch center for the k-th time period.

[0033] Step C2: Utilize the Model Predictive Controller (MPC) with its built-in online cycle counter and stress analyzer to analyze the battery current in real time. and voltage The depth of charge-discharge cycles was identified using an improved rainflow counting method. and average ; Step C3: Construct optimization constraints including power balancing and real-time network topology provided by digital twins; for example, when a section of secondary bus is isolated due to a fault, MPC will receive new topology information and automatically add real-time network topology constraints in the optimization to ensure that power allocation does not lead to overload.

[0034] Step C4: The optimization problem is transformed into a quadratic programming problem. Using the multi-core CPU of the edge computing box, the interior-point method is used to solve the problem within hundreds of milliseconds. The optimal solution is decomposed into power commands and voltage commands for each power module, which are then sent out after passing security verification.

[0035] In this embodiment, the cloud-based collaborative management and optimization layer is used to build a full lifecycle optimizer for global value optimization management. The operation method of the cloud-based collaborative management and optimization layer includes: Step S1: Receive instructions via the Model Prediction Controller (MPC) and set... A negative value (power supplied to the grid) and an increased economic weight. ; Step S2: The MPC sends the preliminary power adjustment plan to the digital twin for pre-control simulation, and the twin simulates the execution. For example, the plan is evaluated, and it is found that if this plan is followed, a severely aged battery cluster will bear most of the discharge task, and its internal temperature will rise rapidly. Step S3: The twin feeds back the temperature warning to the MPC. The MPC readjusts and optimizes the problem, adds the temperature constraint of the aging battery cluster, and solves it again. The new solution is to distribute the power more evenly among multiple battery clusters. Step S4: After optimization, the command is issued. At the same time, the adaptive grid controller senses the slight rise in local voltage caused by power reversal and automatically fine-tunes the reactive power output to absorb excess reactive power to stabilize the voltage and ensure the stability of the power grid during the execution of the command. Step S5: Record the process and results of this collaborative decision-making (such as the effect of the balancing strategy and the voltage adjustment amount) as training data for the reinforcement learning agent and the federated learning model, for better future decisions.

[0036] In this embodiment, the information flow is the foundation, providing data fuel for the precise control of the energy flow and the optimized calculation of the value flow; the value flow is the guide, setting economic boundaries for the scheduling of the energy flow and the protection level of the security flow; the security flow is the guarantee, providing a protective barrier for the stable and reliable operation of the energy flow and the information flow; and the energy flow is the final material carrier, through which the results of the other three flows ultimately realize their value.

[0037] It should be noted that, in this invention, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0038] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An integrated energy management and control system and method based on a photovoltaic energy storage system, characterized in that: The system adopts a "three-layer, four-ring" management and control architecture, which includes building a physical hardware layer, an intelligent edge control layer, a cloud-based collaborative management and optimization layer, and four closed loops that integrate energy flow, information flow, value flow, and security flow across the three layers. The physical hardware layer is used for hardware decoupling, fault isolation, and deep state awareness. The intelligent edge control layer is used to transform the optimization goals in the cloud into collaborative management and security control of hardware modules over a set time scale. The intelligent edge control layer adopts a framework of digital twin training reinforcement learning, which includes: an adaptive network construction control module, a digital twin simulation module, and a model prediction control module. The cloud-based collaborative management and optimization layer is used to build a full lifecycle optimizer for global value optimization management.

2. The integrated energy management control system and method based on photovoltaic energy storage system according to claim 1, characterized in that: The physical hardware layer also includes: Standardized power modules: Each module has a built-in unified "modular controller", equipped with a multi-core processor and running a lightweight real-time operating system; It features a programmable power topology: the main power section uses general-purpose power units; Built-in miniature synchronous phasor measurement unit: Each module integrates an ADC chip to synchronously sample the input / output of this module at a set rate and timestamp it to provide measurement data to the upper layer; Distributed intelligent sensor network: Employing cell-level fusion sensing and magnetic-acoustic joint monitoring; Deterministic communication network: All sensor data is aggregated through a time-sensitive network switch; Hardware discovery and identification: When a standardized module is inserted, the "presence detection" pin of its signal contact is turned on first. The hardware manager reads the module's unique electronic ID, which contains the module type, rated parameters, serial number, and software version, through the I2C bus. The manager then registers the new module with the "digital twin emulator" in the edge control layer via Ethernet. Automatic topology reconstruction and parameter tuning: After receiving information about a new module, the digital twin simulator automatically adds the corresponding virtual module to its virtual model and calls the pre-stored detailed parameter model of that model. At the same time, the digital twin simulator recalculates the control parameters of the entire system online according to the new system topology and compiles and generates new control code. After passing the security verification, it is dynamically loaded into the corresponding module controller and the upper-level adaptive network controller.

3. The integrated energy management control system and method based on photovoltaic energy storage system according to claim 2, characterized in that: The adaptive network control module is used to enable the controller to autonomously adapt to complex and ever-changing power grid conditions. The digital twin simulation module is used to construct a multiphysics model and perform deduction and evolution. The model prediction and control module is used to embed battery aging costs and network reconfiguration capabilities into rolling optimization for joint optimization.

4. The integrated energy management control system and method based on photovoltaic energy storage system according to claim 3, characterized in that: The operation method of the adaptive network construction control module includes: Step A1: The controller receives network-wide synchronous phasor data from the physical layer time-sensitive network at a set rate, including the voltage of each node, the current with timestamps, and calculates the active power, reactive power, and grid voltage amplitude and phase at the system grid connection point in real time. Step A2: The controller continuously injects non-invasive small-signal harmonic disturbances into the power grid, observes the frequency response of the system, and fits the Thevenin equivalent impedance of the power grid online using the recursive least squares method; Step A3: Construct a virtual controller copy that is exactly the same as the real controller and connect it to a real-time digital twin power grid model. Place it next to the real controller and run it. The virtual controller outputs a set of tentative control parameters based on the current state to adjust the action. The action is executed in the digital twin model to deduce the system dynamics within a future set time scale. Step A4: Construct a reward function that balances multiple objectives, maximizes the cumulative reward for the virtual controller, and performs thousands of "trial and error" training sessions per second in the digital twin to learn to automatically select the optimal control parameters; Step A5: The trained strategy parameters are synchronized to the real main controller every few seconds. The main controller applies the parameters to the classic dual-loop vector control or virtual synchronous machine control algorithm to generate the final PWM drive signal, which is then sent to the power module of the physical layer.

5. The integrated energy management control system and method based on photovoltaic energy storage system according to claim 4, characterized in that: Reward function in step A4 Represented as: in, Each item has its own weight. For frequency and voltage deviation, The large deviation is penalized for the square term. The active support power provided to the power grid The total harmonic distortion to be suppressed, The virtual controller maximizes the cumulative reward as the converter efficiency estimate. .

6. The integrated energy management control system and method based on photovoltaic energy storage system according to claim 5, characterized in that: The operation of the digital twin simulation module includes: Step B1: Using a switching model based on nodal admittance, each power module is modeled as a combination of a controlled current source and an RLC network. A second-order RC equivalent circuit model is coupled with a three-dimensional thermal model, including an electrical model and a thermal model. Step B2: The twin in the digital twin model receives real data from the physical layer sensors at a set rate, and uses unscented Kalman filtering to continuously correct the key state variables in the model; Step B3: The digital twin simulator maintains a real shadow model and multiple hypothetical extrapolation models, performing hypothetical extrapolation in parallel; Step B4: The digital twin simulator periodically uploads the model prediction error to the cloud. The cloud aggregates the error data from multiple systems and trains an error correction supernetwork. This network is then distributed to each edge node to dynamically correct the parameters of the local twin model.

7. The integrated energy management control system and method based on photovoltaic energy storage system according to claim 6, characterized in that: The state equations for the electrical and thermal models in step B1 are expressed as follows: in This is the battery's terminal voltage. The operating current flowing through the battery, Open circuit voltage, In a charged state, , For polarization internal resistance, , This is expressed as the polarization capacitor voltage. For heat generation rate, This refers to the temperature of the core region inside the battery cell. The surface temperature of the battery cell. For thermal resistance, Heat capacity; Input current of the electrical model Output voltage And the key heat source—ohmic loss—was calculated. The power is transmitted to the thermal model, which then receives the generated heat. and surface temperature Predicting internal temperature Temperature, in turn, affects the parameters in the electrical model, forming a closed loop.

8. The integrated energy management control system and method based on photovoltaic energy storage system according to claim 7, characterized in that: The operation method of the model prediction control module includes: Step C1: Construct a multi-timescale rolling optimization framework that includes long-time period layers and short-time period layers; Step C2: Utilize a model predictive controller with a built-in online cycle counter and stress analyzer to analyze the battery's current and voltage in real time, and use an improved rainflow counting method to identify the depth of charge-discharge cycles. and average ; Step C3: Construct optimization constraints including power balancing and real-time network topology provided by digital twins; Step C4: The optimization problem is transformed into a quadratic programming problem. Using the multi-core CPU of the edge computing box, the interior-point method is used to solve it. The optimal solution is decomposed into power commands and voltage commands for each power module, which are then sent out after passing security verification.

9. The integrated energy management control system and method based on photovoltaic energy storage system according to claim 8, characterized in that: The optimization objective in the long-period layer of step C1 is: in, To predict the time domain, Let be the power grid interaction cost coefficient for the k-th time period. Let be the power exchanged between the system and the main grid during the k-th time period. The duration of a single time step. This is a weighting factor for battery aging costs. The cost of capacity degradation caused by battery throughput in time period k. The penalty weighting factor for power plan tracking deviation. This represents the actual power reference command value of the system grid connection point during the k-th time period. This represents the long-term power plan value issued by the cloud or dispatch center for the k-th time period.

10. The integrated energy management control system and method based on a photovoltaic energy storage system according to claim 9, characterized in that: The operation methods of the cloud-based collaborative management optimization layer include: Step S1: Receive instructions through the model prediction controller; Step S2: The model predictive controller sends the preliminary power adjustment scheme to the twin in the digital twin model for pre-control simulation, and the twin in the digital twin model performs the simulation execution; Step S3: The twin in the digital twin model will feed back the warning to the model predictive controller. The model predictive controller will readjust the optimization problem, add corresponding constraints, and solve the output again. Step S4: Issue commands after optimization; Step S5: Record the process and results of this collaborative management decision-making, and use them as training data for the reinforcement learning agent and federated learning model to train for better future decisions.

Images