Method and system for establishing power supply and energy storage remote monitoring model

Abstract

The invention relates to the technical field of computers, discloses a power supply and energy storage remote monitoring model establishment method and system, and aims to solve the problems of slow response, low efficiency and safety risk caused by splitting of power grid dispatching, power station control, battery cluster equalization and battery cell management on multiple time scales in the prior art. According to the method, a hierarchical digital twin model system covering four levels of a power grid, a power station, a battery cluster and a battery cell is constructed, and closed-loop collaborative optimization is formed through top-down predictive control instruction propagation and bottom-up aggregative state feedback. The system comprises a multi-scale data acquisition module, a layered digital twin model construction module, a cross-scale state synchronous analysis module and a global cooperative control instruction generation module. According to the invention, the safety, the energy efficiency and the service life of the energy storage system are comprehensively improved.

Description

Technical Field

[0001] This invention belongs to the field of computer technology, specifically relating to a method and system for establishing a remote monitoring model for power supply and energy storage. Background Technology

[0002] As the demand for high-proportion renewable energy consumption and flexible regulation capabilities in new power systems continues to increase, electrochemical energy storage, as a key supporting technology, has been deployed on a large scale on the grid side, power source side, and user side.

[0003] The safe and efficient operation of energy storage systems depends heavily on accurate perception of their status throughout their entire life cycle and multi-level collaborative control.

[0004] Current mainstream monitoring systems generally adopt a centralized or hierarchical independent architecture. Although they can achieve basic data collection and local protection functions, they have limitations when facing complex and dynamic operating scenarios: there is a lack of a unified time scale coordination mechanism between control levels, which makes it difficult for upper-level scheduling instructions to be effectively decomposed and implemented in the lower-level execution units.

[0005] The operation of power supply and energy storage systems involves a coupled process across multiple time scales, from second-level cell response to minute-level cluster balancing, then to hour-level power plant power regulation and even day-ahead grid dispatch.

[0006] Existing technologies typically delegate grid dispatch commands directly to the power plant controller, while battery clusters and cells rely on local BMS for passive responses. The control logic of each layer is isolated from each other, making it impossible to dynamically adjust the upper layer allocation strategy based on the real-time health status of the lower layer, and also making it difficult to achieve rapid cross-layer coordination under sudden operating conditions, resulting in power distribution imbalance, local overcharging and over-discharging, and even thermal runaway risks.

[0007] In existing technologies, on the one hand, there is a lack of systematic decoupling and reconstruction of multi-timescale control processes, and on the other hand, no model-based state prediction and constraint optimization mechanism is introduced in cross-layer instruction transmission.

[0008] This results in delayed response, significant energy loss, and potential safety hazards when the system is dealing with complex scenarios such as frequent power grid frequency adjustments, extreme weather, or equipment aging.

[0009] Especially in large-scale energy storage power stations, the inconsistency between battery clusters exacerbates the above problems. There is an urgent need for a remote monitoring model establishment method that can connect the four-layer structure of "network-station-cluster-chip" and integrate model prediction and closed-loop feedback to achieve safe, economical and efficient full-link collaborative control. Summary of the Invention

[0010] The technical problem to be solved by the present invention is to provide a method and system for establishing a remote monitoring model for power supply and energy storage, which aims to overcome the technical defects in the prior art where grid dispatching, power plant control, battery cluster balancing and cell management are disconnected at different time scales, resulting in slow system response, low energy utilization efficiency and potential safety risks.

[0011] To address the aforementioned technical problems, this invention provides a method for establishing a remote monitoring model for power supply and energy storage. The core of this method lies in constructing a layered digital twin model system that covers four physical levels—grid, power station, battery cluster, and battery cell—and is coupled across time scales.

[0012] This system achieves global collaborative optimization and precise control of the entire energy storage power station through a closed-loop interaction between a top-down predictive control command propagation mechanism and a bottom-up aggregated state feedback mechanism.

[0013] This invention provides a method for establishing a remote monitoring model for power supply and energy storage, comprising: Real-time acquisition of operational status data covering all levels and multiple time scales of the energy storage power station. The operational status data includes: grid-level interactive data, which is obtained through the energy management system interface, including grid dispatch instructions, electricity price signals, and external weather forecast information; power station-level operational data, which is obtained through the power station monitoring and data acquisition system, including the main circuit current, voltage, power factor, and AC / DC side energy conversion efficiency of the power conversion system; battery cluster-level management data, which is obtained through the battery management system bus, including the total voltage, total current, insulation resistance, and charge / discharge status of each battery cluster; and cell-level status data, which is obtained through the cell monitoring unit, including the measured values ​​of the terminal voltage, surface temperature, and internal impedance of each cell.

[0014] Furthermore, the method also includes constructing a hierarchical digital twin model system, which consists of four mutually coupled sub-models. The specific construction process includes: A power grid interaction layer prediction model is established. This model adopts a sequence prediction algorithm based on long short-term memory network. It takes historical power grid load data, real-time electricity price signals and meteorological forecast information as inputs and outputs the power demand prediction curve of energy storage power stations and the frequency regulation demand baseline within a preset time period.

[0015] A power flow optimization model for a power plant is established. This model uses the state-space method to mathematically model the power conversion system, transformer, and auxiliary power system within the plant. A model predictive control algorithm is then used, with the power demand prediction curve as the optimization objective and the operating constraints of the power plant equipment as the boundary conditions, to solve for the optimal sequence of total active and reactive power output commands for the power plant.

[0016] An inter-cluster collaborative balancing model is established, which abstracts each battery cluster as an independent intelligent agent. A multi-agent deep reinforcement learning environment is constructed, and each agent takes maximizing the consistency of the overall battery cluster health status and minimizing the energy loss during the balancing process as the common goal. The cluster-level power allocation command after the decomposition of the total active power output command sequence of the power station is used as input. Through distributed policy learning, an active balancing circuit control strategy for each battery cluster is output.

[0017] An electrochemical aging model for battery cells was established. This model created a second-order Tevinan equivalent circuit model for each battery cell and employed a dual extended Kalman filter algorithm. One filter was used to estimate the state of charge of the battery cell in real time, while the other filter was used to identify the ohmic internal resistance and polarization internal resistance parameters in the equivalent circuit model online, thereby quantitatively assessing the health status and remaining service life of the battery cell.

[0018] Furthermore, the method also includes performing a cross-scale state synchronization resolution process, which comprises two parallel data flow paths: The execution involves a top-down, predictive control command propagation process, specifically: The power demand prediction curve generated by the power grid interaction layer prediction model is passed to the power plant power flow optimization model as its rolling optimization objective function. The total power command of the power station calculated by the power flow optimization model is weighted and decomposed according to the real-time health status of each battery cluster to form the power allocation command of each battery cluster, and then sent to the inter-cluster collaborative equilibrium model. The equalization control strategy generated by the inter-cluster collaborative equalization model is converted into a target charge and discharge current limit for each cell and applied as a constraint to the cell electrochemical aging model.

[0019] A bottom-up, aggregated state information feedback process is executed, specifically as follows: The state of charge and health of each cell estimated by the cell electrochemical aging model are aggregated into a cluster-level state vector, including the average state of charge, state of charge inconsistency, minimum health, and maximum temperature of the battery cluster. The cluster-level state vectors of all battery clusters are uploaded to the power flow optimization model of the power station as dynamic system state constraints when performing model predictive control, in order to correct the power output command. The aggregated equivalent health status and available capacity information of the entire station are fed back to the power grid interaction layer prediction model to adjust its assessment of response capability to power grid dispatch instructions.

[0020] Finally, the method also includes generating and issuing physical equipment control commands based on the globally optimal state formed by the cross-scale state synchronization analysis process, specifically including: The optimal power command output by the power flow optimization model of the power plant is converted into a pulse width modulation signal for the insulated gate bipolar transistor drive module of the power conversion system. The active balancing circuit control strategy output by the inter-cluster collaborative balancing model is converted into on / off control commands for the relays or power switching devices in the balancing module of the battery management system. The cell safety operation boundaries, such as the lower limit voltage and upper limit temperature, updated by the cell electrochemical aging model are then updated in real time to the alarm threshold register of the cell monitoring unit.

[0021] Another aspect of the present invention provides a power supply and energy storage remote monitoring model establishment system, which includes: The multi-scale real-time data acquisition module is used to acquire and synchronize timestamps in real time through communication interfaces with the energy management system, power plant monitoring and data acquisition system, battery management system and cell monitoring unit, and to collect the grid-level interactive data, power plant-level operation data, battery cluster-level management data and cell-level status data.

[0022] The hierarchical digital twin model construction module integrates four sub-modules: the power grid interaction layer prediction model construction unit, which is used to run the sequence prediction algorithm based on the long short-term memory network; The power plant power flow optimization model construction unit is used to perform the power flow optimization calculation based on the model predictive control algorithm; The inter-cluster collaborative equilibrium model construction unit is used to deploy the multi-agent deep reinforcement learning environment and perform policy training and inference. And a cell electrochemical aging model building unit, used to run the dual extended Kalman filter algorithm in parallel for each cell.

[0023] The cross-scale state synchronization analysis module is configured with two parallel processing logics: a predictive control command propagation logic, which is used to implement the step-by-step distribution of control objectives and constraints from the grid layer model to the cell layer model; and an aggregated state feedback logic, which is used to implement the step-by-step aggregation and uploading of real-time state parameters from the cell layer model to the grid layer model, thereby forming a closed loop of control and feedback.

[0024] The global collaborative control command generation module is used to receive the global optimal solution output by the cross-scale state synchronization analysis module, parse and convert it into specific command messages that conform to the underlying hardware communication protocol, and send them to the power conversion system, battery management system and related actuators in the energy storage power station through industrial Ethernet or controller local area network bus.

[0025] Compared with the prior art, the beneficial effects of the present invention are as follows: By constructing a unified digital twin model spanning four levels—grid, power plant, battery cluster, and battery cell—this invention integrates multiple control links that were originally fragmented and had vastly different response time scales into a framework of global collaborative optimization.

[0026] Top-down predictive command propagation enables underlying devices to pre-adjust their operating status to cope with upcoming load changes, greatly improving the system's response speed and its ability to support power grid dispatch.

[0027] Bottom-up aggregated state feedback ensures that high-level decisions are always based on the actual physical state of the underlying cells, and directly incorporates cell-level safety boundaries and health state constraints into the power station-level power scheduling strategy. This avoids safety risks such as battery overcharging, over-discharging, and over-temperature caused by information lag or model mismatch, thereby improving the inherent safety level of the system.

[0028] The application of multi-agent reinforcement learning in inter-cluster balancing has enabled a shift from passive balancing to active predictive balancing, suppressing the accumulation of inconsistencies between battery clusters and extending the cycle life of the entire energy storage system.

[0029] This invention achieves a comprehensive improvement in the operational safety, energy conversion efficiency, and asset economic benefits of energy storage power stations through closed-loop collaborative control across multiple time scales and physical levels. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of the overall technical architecture of a method and system for establishing a remote monitoring model for power supply and energy storage proposed in this invention. Figure 2 This is a schematic diagram of the core principle framework of the hierarchical digital twin model system in this invention; Figure 3 This is a logical flowchart of the real-time acquisition and time synchronization of multi-scale data in this invention. Figure 4 This is a logical flow diagram of the top-down predictive control command propagation process in this invention. Figure 5 This is a logical flowchart of the bottom-up aggregated state information feedback process in this invention. Figure 6 This is a schematic diagram of the multi-level interaction relationship and data flow between the four levels of power grid, power station, battery cluster and battery cell in this invention. Detailed Implementation

[0031] Please refer to Figures 1 to 6This invention provides a method and system for establishing a remote monitoring model for power supply and energy storage, aiming to solve the problems of slow system response, low energy utilization efficiency, and potential safety risks caused by the disconnect between process control at different time scales, such as power grid dispatching instructions, power plant power control, battery cluster balancing, and cell internal state management.

[0032] This method constructs a hierarchical digital twin model system covering four physical levels: power grid, power station, battery cluster, and battery cell. It also introduces a top-down predictive control command propagation mechanism and a bottom-up aggregated state feedback mechanism to form a closed-loop collaborative optimization architecture, thereby achieving global precise control and dynamic health management of the entire energy storage power station.

[0033] The method for establishing the remote monitoring model for power supply and energy storage includes the following steps: S1 collects real-time operational status data covering all levels and multiple time scales of the energy storage power station; S2, constructing a hierarchical digital twin model system consisting of four mutually coupled sub-models; S3 executes a cross-scale state synchronization analysis process, which includes top-down predictive control command propagation and bottom-up aggregated state information feedback. S4. Based on the globally optimal state formed by the cross-scale state synchronization analysis process, generate and issue control commands for the physical equipment.

[0034] In step S1, real-time operational status data covering all levels and multiple time scales of the energy storage power station is collected. This data acquisition process synchronously acquires four types of core data through multi-source heterogeneous interfaces: grid-level interaction data, power station-level operational data, battery cluster-level management data, and cell-level status data.

[0035] Grid-level interactive data is obtained through the energy management system interface, including grid dispatch instructions, real-time electricity price signals, and external weather forecast information.

[0036] The power grid dispatch instructions are issued every 15 minutes, including the active power target value of dispatching one point every 5 minutes in the next 2 hours; the electricity price signal is updated every hour, reflecting the peak-valley-flat electricity price classification of the current period; the meteorological forecast information is provided by a third-party meteorological service platform, including the hourly temperature, humidity, wind speed and solar irradiance forecast values ​​for the next 24 hours.

[0037] Power plant-level operating data is acquired through the power plant monitoring and data acquisition system, including the main circuit current, voltage, power factor, and AC / DC side energy conversion efficiency of the power conversion system.

[0038] The main circuit current and voltage are sampled at a frequency of 1000 Hz and then converted from analog to digital after anti-aliasing filtering. The power factor is calculated by synchronously measuring the phase difference between voltage and current. The AC / DC side energy conversion efficiency is calculated in real time by the ratio of DC side output power to AC side input power. Battery cluster-level management data is acquired through the battery management system bus, including the total voltage, total current, insulation resistance, and charge / discharge status of each battery cluster.

[0039] The total voltage and total current are sampled at a frequency of 100 Hz. The insulation resistance is periodically tested by injecting a small test current and measuring the voltage drop, with a testing cycle of ten minutes. The charging and discharging status is determined by the current integration method combined with open-circuit voltage correction.

[0040] Cell-level status data is acquired through the cell monitoring unit, including measurements of the terminal voltage, surface temperature, and internal impedance of each cell.

[0041] The terminal voltage sampling accuracy is 1 millivolt, and the sampling frequency is 10 Hz; the surface temperature is measured by a thermistor attached to the cell casing with an accuracy of 0.5 degrees Celsius; the internal impedance is identified online by applying a small-amplitude AC excitation signal and measuring the response amplitude and phase, with an excitation frequency of 1 Hz and an identification period of 30 minutes.

[0042] All collected data are appended with a unified timestamp, and time synchronization adopts a precise time protocol, with synchronization error controlled within 100 microseconds, ensuring strict alignment of multi-source data in the time dimension.

[0043] In step S2, a hierarchical digital twin model system is constructed, which consists of four mutually coupled sub-models: the power grid interaction layer prediction model, the power plant power flow optimization model, the inter-cluster collaborative equilibrium model, and the cell electrochemical aging model.

[0044] The power grid interaction layer prediction model adopts a sequence prediction algorithm based on long short-term memory networks.

[0045] The model's input layer receives historical power grid load data, real-time electricity price signals, and weather forecast information.

[0046] Historical grid load data is a sequence of actual charging and discharging power of energy storage power stations recorded every 5 minutes over the past seven days; real-time electricity price signal is the electricity price category identifier for the current period; meteorological forecast information is the hourly temperature, humidity, wind speed and solar irradiance for the next 24 hours.

[0047] The Long Short-Term Memory Network (LSTM) consists of three hidden layers, each with 256 neurons, and uses the hyperbolic tangent function as the activation function.

[0048] The model outputs power demand forecast curves for energy storage power stations at 5-minute intervals over the next 2 hours, as well as a frequency regulation demand baseline. The frequency regulation demand baseline is derived by modeling the correlation between historical grid frequency deviation data and load fluctuations, and is used to characterize the capability boundary of the energy storage system to participate in the primary frequency regulation of the grid.

[0049] The model is retrained before each power grid dispatch command is issued, and the training data window is updated on a rolling basis to ensure the model's adaptability to the latest operating conditions.

[0050] The power flow optimization model of the power plant is based on the state-space method to mathematically model the power conversion system, transformer and auxiliary power system in the plant.

[0051] The state variables include DC bus voltage, AC side current, transformer winding temperature, and auxiliary load power; the control input is the active and reactive power commands of the power conversion system; and the output is the actual active and reactive power injected into the power grid.

[0052] The model is described using discrete-time state-space equations: ; ; For the first The state vector at time t, To control the input vector, For the output vector, , , , The system matrix is ​​calibrated through system identification experiments.

[0053] Based on this model, a model predictive control algorithm is used for rolling optimization.

[0054] The objective function is to minimize the weighted sum of tracking error and control energy consumption: ; The power demand forecast curve provided for the power grid interaction layer prediction model For the prediction time domain, we set it to 24 steps, corresponding to 2 hours. To control the time domain, it is set to 6 steps. and This is the weight matrix.

[0055] The constraints include the maximum current limit of the equipment, the upper limit of the transformer temperature rise, the range of DC bus voltage fluctuation, and the minimum guaranteed power of the auxiliary power system.

[0056] The solver uses a quadratic programming algorithm, performs an optimization calculation every 5 minutes, and outputs the optimal total active power and reactive power output command sequence of the power plant at a point every 5 minutes for the next 2 hours.

[0057] The inter-cluster collaborative equilibrium model abstracts each battery cluster as an independent intelligent agent, constructing a multi-agent deep reinforcement learning environment.

[0058] The environmental status includes the state of charge, health status, temperature, and current power allocation instructions of all battery clusters; the action space of each agent is the combination of the switching states of its active balancing circuit, including bypass switches, bidirectional converter duty cycles, etc.; the reward function is defined as the weighted difference between the improvement in the consistency of the health status of the entire battery cluster and the energy loss during the balancing process.

[0059] Health status consistency is measured by the standard deviation of the health status of all battery clusters; the smaller the standard deviation, the higher the consistency.

[0060] The multi-agent policy network adopts a centralized training and distributed execution architecture. The policy network of each agent is a three-layer fully connected neural network. The input layer dimension is the cluster-level state vector dimension, the number of hidden layer neurons is 128, and the output layer is the action probability distribution.

[0061] During training, the experience replay buffer stores the joint experience tuples of all agents, and the parameters are updated using a deep deterministic policy gradient algorithm.

[0062] During the inference phase, each agent makes independent decisions based solely on its own state and the received cluster-level power allocation instructions, outputting an active balancing circuit control strategy for each battery cluster, including the balancing start timing, balancing current magnitude, and duration.

[0063] The cell electrochemical aging model establishes a second-order Tvinan equivalent circuit model for each cell. This model includes a voltage source representing the open-circuit voltage, an ohmic resistor representing the internal resistance, and two parallel RC branches representing the fast and slow polarization processes, respectively.

[0064] The open-circuit voltage is a nonlinear function of the charged state and is determined by a lookup table. The model's state equations are: ; and For the voltage of the two polarization capacitors, In a charged state, For current, For Coulomb efficiency, This is the rated capacity.

[0065] A dual extended Kalman filter algorithm is used for state estimation and parameter identification.

[0066] The first filter uses the terminal voltage and current as observations to estimate the state of charge and polarization voltage; the second filter uses the residual sequence of the first filter as input to identify the ohmic internal resistance. Polarization resistance , and capacitors , The health status is quantified by the growth rate of the ohmic internal resistance, and the remaining lifetime is predicted by the time required for the health status to decay to the 80% threshold.

[0067] The model operates independently for each battery cell, with a calculation cycle of 1 second.

[0068] In step S3, a cross-scale state synchronization resolution process is performed, which includes two parallel data flow paths.

[0069] The top-down propagation process of predictive control commands is as follows: The power demand forecast curve generated by the power grid interaction layer prediction model is passed to the power plant power flow optimization model, serving as the reference trajectory in its rolling optimization objective function. .

[0070] The total power command sequence of the power plant calculated by the power flow optimization model is weighted and decomposed according to the real-time health status of each battery cluster.

[0071] Weighting coefficients For the A battery cluster is defined as: ; For the first The health status of individual battery clusters This represents the total number of battery clusters.

[0072] This weighting method ensures that battery clusters in poorer health conditions bear less power load, thereby slowing down their aging process.

[0073] The decomposed cluster-level power allocation instructions are then sent to the inter-cluster collaborative balancing model.

[0074] The equalization control strategy generated by the inter-cluster collaborative equalization model is converted into target charge and discharge current limits for each cell.

[0075] This limit is dynamically adjusted based on the current state of charge, temperature, and health status of the battery cell. For example, when the battery cell temperature is above 45 degrees Celsius, the charging current limit is reduced by 20%; when the state of charge is above 90%, the charging current limit decreases exponentially.

[0076] These limits are applied as hard constraints to the current input of the cell electrochemical aging model to ensure that the simulation results are always within the safe operating boundaries.

[0077] The bottom-up aggregated state information feedback process is as follows: The state of charge and health of each cell estimated by the cell electrochemical aging model are aggregated into the cluster-level state vector of the battery cluster.

[0078] This vector includes the average state of charge, the inconsistency of the state of charge (defined as the standard deviation), the lowest health state, the highest temperature, and the average internal resistance growth rate.

[0079] The cluster-level state vectors of all battery clusters are uploaded to the power flow optimization model of the power plant.

[0080] In the constraints of model predictive control, a new dynamic system state constraint is added: the maximum allowable power of the DC bus is limited by both the minimum healthy state and the maximum temperature, specifically as follows: ; in and This is an empirical coefficient. The temperature safety threshold is used. The aggregated equivalent health status of the entire station (defined as the weighted average of the health status of all cells) and available capacity (defined as the sum of the available capacity of all cells) are fed back to the grid interaction layer prediction model.

[0081] By adding these two aggregated metrics as additional channels to the input features of the Long Short-Term Memory network, the model can assess the actual responsiveness of current energy storage assets and avoid outputting high power demand forecasts when the health status is severely deteriorated.

[0082] In step S4, based on the globally optimal state formed by the cross-scale state synchronization analysis process, control commands for the physical equipment are generated and issued.

[0083] The optimal power command output by the power flow optimization model of the power plant is converted into a pulse width modulation signal for the insulated gate bipolar transistor drive module of the power conversion system.

[0084] The conversion process uses a space vector modulation algorithm to calculate the three-phase modulation wave based on the active and reactive power commands, and then generates six complementary pulse width modulation signals through triangular carrier comparison. The dead time is set to 3 microseconds.

[0085] The active balancing circuit control strategy output by the inter-cluster collaborative balancing model is converted into on / off control commands for the relays or power switching devices in the balancing module of the battery management system.

[0086] For passive equalization, the instruction is the relay closing time; for active equalization, the instruction is the duty cycle and operating frequency of the bidirectional converter.

[0087] The cell safety operating boundaries updated by the cell electrochemical aging model, including the lower limit voltage (typically 2.5 volts), upper limit voltage (typically 3.65 volts), and upper limit temperature (typically 60 degrees Celsius), are written in real time into the alarm threshold register of the cell monitoring unit.

[0088] When the monitored value exceeds the threshold, the cell monitoring unit immediately triggers local protection and reports a fault code to the battery management system.

[0089] The power supply and energy storage remote monitoring model establishment system includes a multi-scale data real-time acquisition module, a hierarchical digital twin model construction module, a cross-scale state synchronization analysis module, and a global collaborative control command generation module.

[0090] The multi-scale data real-time acquisition module is equipped with four communication interface units: the first interface unit is connected to the energy management system, adopts the IEC61850 protocol, and periodically subscribes to power grid dispatch instructions, electricity price signals and meteorological forecast information; The second interface unit connects to the power plant monitoring and data acquisition system, and uses the Modbus TCP protocol to poll the electrical parameters of the main circuit of the power conversion system at a 10-millisecond cycle. The third interface unit connects to the battery management system, uses the CAN bus protocol, and sets the baud rate to 500 kilobits per second to receive real-time data on the total voltage, total current, and insulation resistance of each battery cluster. The fourth interface unit connects to the cell monitoring unit and uses daisy-chain serial communication to collect the terminal voltage and temperature of all cells at 100-millisecond intervals. This module has a built-in high-precision real-time clock chip that supports the IEEE 1588 precision time protocol, ensuring timestamp synchronization of all interface data.

[0091] The hierarchical digital twin model building module integrates four functional sub-units.

[0092] The predictive model building unit of the power grid interaction layer is deployed on a high-performance computing server and equipped with a graphics processor to accelerate the inference and training of the long short-term memory network. The power flow optimization model building unit runs on a real-time operating system and uses C language to implement the state-space model and quadratic programming solver to ensure the determinism of the control cycle. The inter-cluster collaborative equilibrium model building unit adopts a distributed architecture, with each battery cluster corresponding to an edge computing node. The nodes are interconnected through Ethernet switches to perform multi-agent policy inference. The cell electrochemical aging model building unit is an embedded software module deployed on the main control chip of the battery management system. It opens up an independent task thread for each cell and periodically executes the dual extended Kalman filter algorithm.

[0093] The cross-scale state synchronization parsing module is configured with two parallel processing logics.

[0094] The predictive control command propagation logic is executed by the central coordinating controller, which periodically reads the output of each sub-model, performs operations such as weighted decomposition and limit transformation, and sends the commands to the next layer model through a message queue; the aggregated state feedback logic is implemented by the state aggregator, which listens for the state update events of the underlying model, performs statistical calculations and constraint mapping, and pushes the results to the upper layer model.

[0095] The two logics share the same in-memory database to ensure data consistency.

[0096] The global collaborative control command generation module receives the globally optimal solution output by the cross-scale state synchronization analysis module and parses the power commands, equalization strategies, and safety boundary parameters within it. This module has a built-in protocol conversion engine that converts higher-level commands into specific command messages conforming to the underlying hardware communication protocol.

[0097] For the power conversion system, control commands conforming to the IEC61850 GOOSE message format are generated; for the battery management system, object dictionary write requests conforming to the CANopen protocol are generated; for the cell monitoring unit, register write instructions conforming to the dedicated serial protocol are generated.

[0098] All command messages are sent to the corresponding actuators via industrial Ethernet or controller area network bus, with the sending cycle strictly matching the control cycle of each device.

[0099] Through the collaborative work of the above methods and systems, this invention achieves unified modeling and closed-loop control of power supply and energy storage systems at multiple levels and time scales, solving the technical defects of fragmented control links in traditional monitoring systems and improving the system's response speed, operating efficiency, and intrinsic safety level.

[0100] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish an entity or operation from another entity or operation, 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.

[0101] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for establishing a remote monitoring model for power supply and energy storage, characterized in that, include: Real-time acquisition of operational status data covering all levels and multiple time scales of energy storage power stations, including grid-level interaction data, power station-level operational data, battery cluster-level management data, and cell-level status data; A hierarchical digital twin model system is constructed, which includes a power grid interaction layer prediction model, a power plant power flow optimization model, an inter-cluster collaborative equilibrium model, and a cell electrochemical aging model. The process of performing cross-scale state synchronization analysis includes a top-down predictive control command propagation process and a bottom-up aggregated state information feedback process. Based on the globally optimal state formed by the cross-scale state synchronization analysis process, control commands for physical equipment are generated and issued.

2. The method for establishing a remote monitoring model for power supply and energy storage according to claim 1, characterized in that, The real-time acquisition of operational status data covering all levels and multiple time scales of the energy storage power station includes: The power grid-level interactive data is obtained through the energy management system interface. The power grid-level interactive data includes power grid dispatch instructions, electricity price signals, and external meteorological forecast information. The power plant monitoring and data acquisition system acquires power plant-level operating data, which includes the main circuit current, voltage, power factor, and AC / DC side energy conversion efficiency of the power conversion system. Battery cluster-level management data is obtained through the battery management system bus. The battery cluster-level management data includes the total voltage, total current, insulation resistance, and charge / discharge status of each battery cluster. Cell-level status data is acquired through a cell monitoring unit. The cell-level status data includes the measured values ​​of the terminal voltage, surface temperature, and internal impedance of each cell.

3. The method for establishing a remote monitoring model for power supply and energy storage according to claim 1, characterized in that, The construction of the hierarchical digital twin model system includes: A power grid interaction layer prediction model is established, and a sequence prediction algorithm based on long short-term memory network is adopted. The model takes historical power grid load data, real-time electricity price signals and meteorological forecast information as inputs, and outputs the power demand prediction curve of energy storage power station and the frequency regulation demand baseline within the future preset time period. A power flow optimization model for a power plant is established. Based on the state-space method, mathematical models are performed on the power conversion system, transformer, and auxiliary power system within the plant. A model predictive control algorithm is then used, with the power demand prediction curve as the optimization objective and the operating constraints of the power plant equipment as the boundary conditions, to solve for the optimal sequence of total active and reactive power output commands for the power plant. A cluster collaborative balancing model is established, in which each battery cluster is abstracted as an independent intelligent agent. A multi-agent deep reinforcement learning environment is constructed. Each intelligent agent takes maximizing the consistency of the health status of the entire battery cluster and minimizing the energy loss in the balancing process as the common goal. The cluster-level power allocation command after the decomposition of the total active power output command sequence of the power station is used as input. Through distributed strategy learning, an active balancing circuit control strategy for each battery cluster is output. An electrochemical aging model for battery cells was established, and a second-order Tevinan equivalent circuit model was created for each cell. A dual extended Kalman filter algorithm was adopted, with one filter used to estimate the state of charge of the cell in real time, and another filter used to identify the ohmic internal resistance and polarization internal resistance parameters in the equivalent circuit model online, thereby quantitatively assessing the health status and remaining service life of the cell.

4. The method for establishing a remote monitoring model for power supply and energy storage according to claim 1, characterized in that, The top-down predictive control command propagation process includes: The power demand prediction curve generated by the power grid interaction layer prediction model is passed to the power plant power flow optimization model as its rolling optimization objective function. The total power command of the power station calculated by the power flow optimization model is weighted and decomposed according to the real-time health status of each battery cluster to form the power allocation command of each battery cluster, and then sent to the inter-cluster collaborative equilibrium model. The equalization control strategy generated by the inter-cluster collaborative equalization model is converted into a target charge and discharge current limit for each cell and applied as a constraint to the cell electrochemical aging model.

5. The method for establishing a remote monitoring model for power supply and energy storage according to claim 1, characterized in that, The bottom-up, aggregated state information feedback process includes: The state of charge and health of each cell estimated by the cell electrochemical aging model are aggregated into a cluster-level state vector, including the average state of charge, state of charge inconsistency, minimum health, and maximum temperature of the battery cluster. The cluster-level state vectors of all battery clusters are uploaded to the power flow optimization model of the power station as dynamic system state constraints when performing model predictive control, in order to correct the power output command. The aggregated equivalent health status and available capacity information of the entire station are fed back to the power grid interaction layer prediction model to adjust its assessment of response capability to power grid dispatch instructions.

6. The method for establishing a remote monitoring model for power supply and energy storage according to claim 4, characterized in that, The total power command of the power station calculated by the power flow optimization model is weighted and decomposed according to the real-time health status of each battery cluster, including: For the Each battery cluster defines a weighting coefficient: ； in For the first The health status of individual battery clusters This represents the total number of battery clusters. The total power command of the power station is decomposed based on the weighting coefficients to obtain the power allocation command for each battery cluster.

7. The method for establishing a remote monitoring model for power supply and energy storage according to claim 5, characterized in that, The state of charge and health status of each cell estimated by the cell electrochemical aging model are aggregated into a cluster-level state vector, including the average state of charge, state of charge inconsistency, lowest health status, and highest temperature of the battery cluster. The average state of charge, state of charge inconsistency, minimum health state, maximum temperature, and average internal resistance growth rate are calculated as components of the cluster-level state vector.

8. The method for establishing a remote monitoring model for power supply and energy storage according to claim 1, characterized in that, The global optimal state formed based on the cross-scale state synchronization analysis process generates and issues control commands for physical equipment, including: The optimal power command output by the power flow optimization model of the power plant is converted into a pulse width modulation signal for the insulated gate bipolar transistor drive module of the power conversion system. The active balancing circuit control strategy output by the inter-cluster collaborative balancing model is converted into on / off control commands for the relays or power switching devices in the balancing module of the battery management system. The cell safety operation boundary updated by the cell electrochemical aging model is updated in real time to the alarm threshold register of the cell monitoring unit.

9. A system for establishing a remote monitoring model for power supply and energy storage, characterized in that, The power supply and energy storage remote monitoring model establishment method according to any one of claims 1 to 8 is used to establish the power supply and energy storage remote monitoring model, wherein the power supply and energy storage remote monitoring model establishment system includes: The multi-scale real-time data acquisition module is used to acquire and synchronize timestamps in real time through communication interfaces with the energy management system, power plant monitoring and data acquisition system, battery management system and cell monitoring unit, and to collect the grid-level interactive data, power plant-level operation data, battery cluster-level management data and cell-level status data. The hierarchical digital twin model building module integrates four sub-modules: The power grid interaction layer prediction model construction unit is used to run the sequence prediction algorithm based on the long short-term memory network; The power plant power flow optimization model construction unit is used to perform the power flow optimization calculation based on the model predictive control algorithm; The inter-cluster collaborative equilibrium model construction unit is used to deploy the multi-agent deep reinforcement learning environment and perform policy training and inference. And a cell electrochemical aging model building unit, used to run the dual extended Kalman filter algorithm in parallel for each cell; The cross-scale state synchronization parsing module is configured with two parallel processing logics: Predictive control command propagation logic is used to implement the hierarchical distribution of control objectives and constraints from the power grid layer model to the cell layer model; Aggregated state feedback logic is used to realize the hierarchical aggregation and uploading of real-time state parameters from the cell layer model to the grid layer model, thereby forming a closed loop of control and feedback. The global collaborative control command generation module is used to receive the global optimal solution output by the cross-scale state synchronization analysis module, parse and convert it into specific command messages that conform to the underlying hardware communication protocol, and send them to the power conversion system, battery management system and related actuators in the energy storage power station through industrial Ethernet or controller local area network bus.

10. The power supply and energy storage remote monitoring model establishment system according to claim 9, characterized in that, The multi-scale real-time data acquisition module is equipped with four communication interface units: The first interface unit connects to the energy management system and adopts the IEC61850 protocol to periodically subscribe to grid dispatch instructions, electricity price signals and weather forecast information. The second interface unit connects to the power plant monitoring and data acquisition system, and uses the Modbus TCP protocol to poll the electrical parameters of the main circuit of the power conversion system at a fixed period. The third interface unit connects to the battery management system and uses the CAN bus protocol to receive real-time data on the total voltage, total current and insulation resistance of each battery cluster. The fourth interface unit connects to the cell monitoring unit and uses daisy-chain serial communication to collect the terminal voltage and temperature of all cells.

Images