Power grid load dynamic prediction and energy storage cooperative adjustment method based on double-layer model virtual rehearsal

Abstract

The invention discloses a power grid load dynamic prediction and energy storage coordinated regulation method based on double-layer model virtual rehearsal, and relates to the technical field of novel power system operation control. The method comprises the following steps: constructing a multi-scale online simulation model synchronous with a physical system; generating a probability prediction interval of load and output, and solving the model prediction control optimization model to obtain a candidate charging and discharging strategy; and performing virtual preview and safety verification correction on the candidate strategy in the multi-scale online simulation model, outputting a control instruction, and adjusting the energy storage system. The method is used for solving the problems that the internal damage risk of the battery is uncontrollable and the capability of coping with the random fluctuation of the source load is insufficient due to the lack of the micromechanism sensing capability in the existing scheme.

Description

Technical Field

[0001] This invention relates to the field of novel power system operation and control technology, and more specifically, to a method for dynamic prediction of power grid load and coordinated regulation of energy storage based on a two-layer model virtual simulation. Background Technology

[0002] Currently, advanced solutions for grid load forecasting and energy storage regulation in the industry mostly adopt a collaborative architecture of "prediction + optimization". Typically, variational mode decomposition (VMD) combined with deep learning models (such as LSTM and Transformer) is used to extract time-series features and make short-term predictions of load sequences. Based on this, model predictive control (MPC) algorithms are used for rolling optimization to formulate the scheduling strategy for the energy storage system. This approach alleviates, to some extent, the regulation deviation problem caused by response lag in traditional scheduling modes, thus improving the system's operating efficiency.

[0003] For example, the invention patent announcement CN118572757B discloses a method for regulating and maintaining an intelligent energy storage system based on digital twins. This method establishes a black-box digital twin model of the energy storage system, uses historical data to train the SOX intelligent algorithm to predict parameters, and adjusts the parameters of the black-box model by comparing simulation results with measured results. Based on this model, a charging and discharging regulation strategy for the energy storage system is formulated.

[0004] The above-disclosed technical solutions have at least the following technical problems: 1. The "black box" model upon which the solution relies is essentially a data-driven input-output fitting model, lacking a description of the internal electrochemical reaction mechanism of the energy storage battery. Although this model can estimate macroscopic external indicators such as state of charge (SOC) and terminal voltage, it cannot perceive the microscopic electrochemical states inside the battery, such as the solid-phase potential distribution, liquid-phase potential difference, and internal temperature field of the cell, in real time. This means that when the control system executes high-power charge and discharge commands, it is in a "blind spot" regarding potential microscopic risks such as lithium plating and side reactions inside the battery.

[0005] 2. Due to the lack of safety constraints at the microscopic mechanism level, existing solutions often only indirectly mitigate risks by setting large macroscopic safety margins (such as limiting the scope of SOC use) when formulating control strategies. This not only results in a waste of energy storage asset performance (unutilized available capacity), but also makes it difficult to fundamentally prevent irreversible battery damage caused by exceeding microscopic limits under extreme operating conditions such as severe grid load fluctuations. Consequently, it fails to meet the stringent requirements of new power systems for long lifespan and high safety of energy storage devices. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of existing technologies, embodiments of the present invention provide a method for dynamic prediction of grid load and coordinated regulation of energy storage based on virtual pre-simulation of a two-layer model. By constructing a multi-scale online simulation model integrating battery electrochemical and thermal coupling mechanism models, candidate charging and discharging strategies are generated using probability prediction intervals. The electrochemical reaction state is obtained through virtual pre-simulation in the multi-scale online simulation model, and the strategy is verified and corrected for safety. This addresses the problems of uncontrollable internal battery damage risk and insufficient ability to cope with random fluctuations in source and load caused by the lack of microscopic mechanism perception capabilities in existing solutions.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A method for dynamic prediction of grid load and coordinated regulation of energy storage based on virtual pre-simulation using a two-layer model includes the following steps: constructing a multi-scale online simulation model integrating battery electrochemical and thermal coupling mechanism models; generating probability prediction intervals for load and renewable energy output based on multi-source heterogeneous data; solving the prediction optimization model using the probability prediction intervals as constraints to obtain candidate charging and discharging strategies for the energy storage system; virtually pre-simulating the strategies in the multi-scale online simulation model to obtain the electrochemical reaction state; verifying the safety of the candidate charging and discharging strategies based on the electrochemical reaction state; when the verification fails, generating a penalty factor according to the degree of violation and feeding it back to the prediction optimization model, tightening the constraints, and resolving until the verification passes, generating control commands to coordinately regulate the energy storage system.

[0008] In a preferred embodiment, the multi-scale online simulation model includes a power grid topology model and an electrochemical-thermal coupling mechanism model of the energy storage battery.

[0009] In a preferred embodiment, the virtual simulation process includes using an online parameter identification algorithm to update the physical parameters of the multi-scale online simulation model in real time, and simulating the battery microstate under candidate charge-discharge strategies based on the updated parameters. If the microstate exceeds the limit, strategy correction is triggered.

[0010] In a preferred embodiment, the step of generating a probability prediction interval using a prediction model specifically includes: decomposing a historical load sequence into multiple intrinsic mode functions of different frequencies using variational mode decomposition technology; inputting the decomposed mode functions and meteorological characteristics into a time series prediction model based on deep learning, and training it using a quantile regression loss function; outputting the upper and lower bounds of power prediction at a preset confidence level at future times, and the probability prediction interval is constituted by the upper and lower bounds.

[0011] In a preferred embodiment, generating a penalty factor based on the degree of violation includes: when a lithium plating risk is triggered, calculating the difference between the minimum solid-liquid phase potential difference of the negative electrode and the lithium plating safety threshold as the degree of violation, and mapping the degree of violation to a power penalty factor using a pre-calibrated sensitivity coefficient of the negative electrode potential to the current; when a temperature over-limit risk is triggered, calculating the difference between the cell center temperature and the safety temperature threshold as the degree of violation, and calculating the power adjustment required to bring the temperature back to the safety temperature threshold as a power penalty factor using a pre-calibrated sensitivity coefficient of temperature to power.

[0012] In a preferred embodiment, the step of using an online parameter identification algorithm to update the physical parameters of the multi-scale online simulation model in real time specifically includes: acquiring real-time voltage, current, and temperature measurements of the physical power grid; constructing a state observer based on Kalman filtering; iteratively correcting the physical parameters in the multi-scale online simulation model based on the residuals between the measured values ​​and the output values ​​of the multi-scale online simulation model; and determining that the multi-scale online simulation model has completed parameter synchronization when the residuals converge to within a preset threshold.

[0013] In a preferred embodiment, the construction of the model predictive control optimization model includes establishing an optimization model with the charging and discharging power of the energy storage system at each time step as the decision variable, aiming to minimize the total system operating cost; the total system operating cost includes at least two parts: the grid transaction cost calculated based on the real-time electricity price and the power interaction with the grid, and the battery aging cost calculated based on the energy storage battery loss model; wherein, the battery aging cost is dynamically calculated by extracting charge and discharge cycle characteristics using the rainflow counting method and combining them with the current battery health status.

[0014] In a preferred embodiment, the virtual pre-simulation of the candidate charge-discharge strategy in a multi-scale online simulation model includes: inputting the candidate charge-discharge strategy into the electrochemical-thermal coupling mechanism model; calculating the electrochemical reaction state inside the battery, wherein the electrochemical reaction state includes at least the solid-phase potential, the liquid-phase potential, and the internal temperature field of the cell; determining whether the electrochemical reaction state satisfies microscopic safety constraints, wherein the microscopic safety constraints include at least a potential difference constraint for suppressing lithium plating reaction and a temperature constraint for preventing thermal runaway; wherein the potential difference constraint requires that the difference between the solid-phase potential and the liquid-phase potential be above a preset lithium plating safety threshold, and the temperature constraint requires that the internal temperature of the cell be below a preset safety temperature threshold.

[0015] In a preferred embodiment, the safety verification and correction of the candidate charging and discharging strategy based on the pre-simulation results includes: when the micro-safety constraints are met, marking the candidate charging and discharging strategy as a control command; when the micro-safety constraints are violated, calculating the penalty factor corresponding to the degree of violation, and feeding the penalty factor back to the constraints of the model predictive control framework; tightening the maximum allowable charging and discharging power boundary of the energy storage system, and resolving the optimization problem until the generated strategy passes the virtual pre-simulation.

[0016] In a preferred embodiment, the coordinated adjustment of the energy storage system based on the control command specifically involves: using a rolling time-domain control method, issuing the power setpoint of the first time step in the control command to the energy storage converter; and in the next control cycle, repeating the prediction, optimization, and virtual simulation steps based on the latest measured data and the updated multi-scale online simulation model state.

[0017] This invention provides a grid load dynamic prediction and energy storage coordinated regulation system based on a two-layer model virtual pre-simulation, comprising: an online simulation module for collecting multi-source heterogeneous data from the grid side, power source side, and energy storage side, and constructing a multi-scale online simulation model synchronized with the physical system; a probability prediction module for generating probability prediction intervals for load and renewable energy output within a preset time period based on the multi-source heterogeneous data and using the prediction model; a coordinated optimization module for constructing a model predictive control optimization model based on the probability prediction intervals and solving for candidate charging and discharging strategies for the energy storage system; a virtual-real pre-simulation module for performing virtual pre-simulation of the candidate charging and discharging strategies in the multi-scale online simulation model, verifying and correcting the safety of the candidate charging and discharging strategies based on the pre-simulation results, and obtaining control commands; and an execution control module for coordinating regulation of the energy storage system based on the control commands.

[0018] The technical effects and advantages of the present invention regarding the method for dynamic prediction of power grid load and coordinated regulation of energy storage based on two-layer model virtual pre-simulation are as follows: This invention effectively solves the technical challenge of existing "black box" models being unable to perceive the internal microscopic state of batteries by constructing a multi-scale online simulation model integrating battery electrochemical and thermal coupling mechanism models and combining it with optimization solutions based on probability prediction intervals. Its core advantage lies in the fact that by inputting the candidate charging and discharging strategies obtained from the solution into the multi-scale online simulation model for virtual pre-running, it can directly obtain the electrochemical reaction state reflecting the changes in the physical field inside the battery, and accordingly conduct forward-looking safety verification and correction of the strategy. This proactive closed-loop defense mechanism not only overcomes the risk of microscopic damage caused by traditional solutions that rely solely on macroscopic indicators for passive protection, but also quantifies the uncertainty of source-load fluctuations through probability prediction intervals. The resulting control commands can ensure the accuracy of grid coordinated regulation while guaranteeing the intrinsic safety and long-life operation of the energy storage system from a mechanistic perspective. Attached Figure Description

[0019] Figure 1 A schematic diagram of the method for dynamic prediction of power grid load and coordinated regulation of energy storage based on virtual pre-simulation of a two-layer model provided in an embodiment of the present invention; Figure 2 A schematic diagram of probability prediction results based on the VMD-TFT model provided in an embodiment of the present invention; Figure 3 The evolution curve of the health status penalty coefficient as a function of battery aging is provided in this embodiment of the invention. Figure 4 A dynamic simulation comparison diagram of the solid-liquid phase potential difference of the battery negative electrode before and after correction, provided in an embodiment of the present invention; Figure 5 This is a block diagram of a grid load dynamic prediction and energy storage coordinated regulation system based on a two-layer model virtual pre-simulation provided in an embodiment of the present invention. Detailed Implementation

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

[0021] Example 1, Figure 1 This invention presents a method for dynamic prediction of power grid load and coordinated regulation of energy storage based on a two-layer model virtual simulation, comprising the following steps: S1 collects multi-source heterogeneous data from the grid side, power source side, and energy storage side, and constructs a multi-scale online simulation model synchronized with the physical system.

[0022] It should be noted that, in order to comprehensively perceive the operating status of the new power system, this step involves data acquisition through sensor networks deployed in the physical system, including SCADA (Supervisory Control and Data Acquisition), PMU (Phasor Measurement Unit), and BMS (Battery Management System). The multi-source heterogeneous data specifically includes: 1. Grid-side data: mainly includes the voltage amplitude and phasor of each bus node, active / reactive power flow of line branches, frequency deviation, transformer tap position and circuit breaker switching status. 2. Power supply side data: For distributed photovoltaic or wind power, collect real-time data on solar irradiance, wind speed, wind direction, ambient temperature, backsheet temperature, and inverter output power curves; 3. Energy storage side data: Collect terminal voltage, total current, voltage difference of individual cells, temperature distribution of individual cells, coolant flow rate and ambient temperature of energy storage battery clusters.

[0023] To address the timestamp discrepancy issue caused by different data sources, the system sets a baseline simulation step size (e.g., 10ms) for the multi-scale online simulation model to establish a unified time reference axis. Specifically, for millisecond-level high-frequency PMU data, linear interpolation is used to fill in packet loss caused by network transmission; for minute-level SCADA or smart meter data, zero-order hold is used for time-domain filling, and a Kalman filter algorithm is employed to iteratively update the measurement noise covariance matrix to smooth out step interference in low-frequency data. This approach effectively suppresses artificial high-frequency noise generated during the mapping of low-frequency data to high-frequency data, ensuring the stability and accuracy of the input data.

[0024] The multi-scale online simulation model includes a power grid topology model and an electrochemical-thermal coupling mechanism model of energy storage batteries.

[0025] Specifically, the construction of multi-scale online simulation models aims to establish a high-fidelity mapping relationship between physical space and virtual space, and its core consists of the following two parts: 1. Power Grid Topology Model This model is based on the Common Information Model (CIM) file defined by the IEC 61970 / 61968 standard. The system reads the device connection relationships and attribute parameters from the CIM file and uses graph theory algorithms to construct the node admittance matrix of the physical power grid. This matrix not only reflects the static physical connections of the power grid, but also reconstructs the network topology in real time based on the collected circuit breaker switch states, thereby dynamically generating a mathematical model for power flow calculation.

[0026] 2. Electrochemical-thermal coupling mechanism model This is the core component describing the microscopic behavior of the energy storage unit in the multi-scale online simulation model. To balance computational accuracy and real-time operation, the multi-scale online simulation model in this embodiment adopts a multi-scale, two-layer modeling strategy, specifically including a control layer model and a verification layer model: (1) The first layer is the control layer model: This embodiment adopts a combination of a second-order RC equivalent circuit model and a lumped parameter thermal model. This model accurately describes the nonlinear characteristics of the battery internal resistance and open-circuit voltage as a function of temperature (T) and state of charge (SOC) through coupling relationships.

[0027] First, the electrical characteristic equation of the battery is established to describe the dynamic response of the battery's terminal voltage. Its mathematical expression is as follows: (1) (2) (3) In the formula, This indicates the battery's terminal voltage (unit: V, volt). Indicates the effect of SOC and temperature Changing open-circuit voltage; Indicates the charging and discharging current (unit: A, ampere, with discharge defined as positive). Indicates ohmic internal resistance (unit: ,ohm); and These represent the electrochemical polarization voltage and the concentration polarization voltage, respectively. , and , Polarization resistances corresponding to the first and second order RC networks (unit: ) and polarization capacitance (unit: F, farad).

[0028] Specifically, the parameters of the aforementioned binary function , The steps to obtain etc. are as follows: i. Perform hybrid pulse power characteristic (HPPC) testing on the energy storage cells; ii. Under preset temperature gradients (e.g., -10℃, 0℃, 25℃, 45℃) and SOC nodes (e.g., one point every 10%), collect the voltage response curves of the battery cells under charge and discharge pulse excitation; iii. The least squares method is used to identify the parameters of the above electrical differential equations offline, and a multi-dimensional lookup table is constructed and stored in the multi-scale online simulation model database for real-time simulation.

[0029] Secondly, the thermal balance equation for the battery is established. To more accurately describe the temperature change of the battery under high-rate charging and discharging, this embodiment introduces a reversible entropy-heat term based on Joule heating, and its modified mathematical expression is as follows: (4) In the formula, Indicates the mass of a single battery cell (unit: kg); This indicates the specific heat capacity of the battery (unit: J / (kg·K)). This indicates the average internal temperature of the battery cell (unit: Kelvin). Indicates time; The equivalent total internal resistance of the battery for heat generation is approximately equal to [value missing] in steady state. The sum; This represents the entropic heat produced by a reversible reaction, where... The entropy coefficient reflects the sensitivity of open-circuit voltage to temperature changes. This factor is either heat-absorbing or heat-releasing during battery charging and discharging, and is crucial for accurately predicting the internal temperature field of the battery. This represents the convective heat transfer coefficient between the battery surface and the environment (unit: W / (m²·K)). Indicates the effective heat dissipation area of ​​the battery (unit: m²). This represents the ambient temperature (unit: K). This layer of the model can calculate the terminal voltage fluctuation and average temperature rise in real time, providing basic battery state constraints for macroscopic power scheduling.

[0030] (2) Second layer: Validation layer model: This layer model has high computational complexity but can accurately reflect the microscopic mechanism and is used for subsequent virtual simulation of the microscopic safety of candidate strategies. This embodiment adopts a quasi-two-dimensional (P2D) electrochemical model.

[0031] This model, based on porous electrode theory, considers the battery interior as three regions: the positive electrode, the separator, and the negative electrode. It describes the solid-phase diffusion process of lithium ions within the active particles and their liquid-phase migration process in the electrolyte using a set of partial differential equations. Its core mechanism equations include: i. Solid-phase lithium-ion diffusion equation (Fick's law): describes the concentration distribution of lithium ions in the radial direction of spherical active particles; ii. Liquid phase charge conservation equation: describes the potential distribution and ion migration in the electrolyte.

[0032] By integrating this verification layer model into a multi-scale online simulation model, the system can output the spatiotemporal distribution of solid-phase potential, liquid-phase potential, and cell center temperature inside the battery, thereby achieving accurate capture of lithium plating risk and thermal runaway risk.

[0033] S2, Based on the multi-source heterogeneous data, a prediction model is used to generate a probability prediction range for load and new energy output within a preset time period in the future.

[0034] The steps for generating probability prediction intervals using a prediction model specifically include: The first step is to use variational mode decomposition (VMD) to decompose the historical load sequence into multiple eigenmode functions of different frequencies.

[0035] Specifically, considering the non-stationarity of power grid load data, this embodiment uses the variational mode decomposition (VMD) algorithm to transform the original load sequence. Decomposed into frequencies around the center Oscillating intrinsic mode function (IMF) components .

[0036] The VMD algorithm achieves signal decomposition by solving a constrained variational problem, which is expressed as follows: (5) (6) In the formula, and These represent the set of modal components and the corresponding set of center frequencies, respectively. The Dirac distribution function; This represents the convolution operation; express The square of the norm. This problem is solved by introducing a quadratic penalty factor. Using Lagrange multipliers and the alternating direction multiplier method (ADMM) for iterative solution, modal components of different frequencies can be separated.

[0037] To ensure effective decomposition and adapt to online real-time prediction, this embodiment adopts the following specific strategies: 1. Parameter determination strategy: Number of mode decompositions The value of directly affects the decomposition effect. This embodiment uses the center frequency observation method to determine . Value, i.e., observing different The distribution of the center frequencies of the main modes under the given value is considered. The critical value at which the center frequencies are similar (i.e., mode aliasing occurs) is used as the upper limit, or it can be set based on experience. Up to 8; simultaneously, the secondary penalty factor will be... Set to a moderate value (e.g.) This is to ensure the accuracy of the reconstruction.

[0038] 2. Sliding Window Decomposition Strategy: Considering the real-time requirements of online prediction and the potential endpoint effect of VMD, this embodiment adopts a sliding window strategy. A fixed-length time window is set (e.g., data from the past 24 hours). Whenever a new sampling point is generated, the time window is slid backward and VMD decomposition is re-executed. The values ​​of each IMF component obtained from the decomposition at the current time are extracted as feature inputs for subsequent models.

[0039] The second step involves inputting the decomposed modal functions and meteorological features into a deep learning-based time series prediction model, and training it using a quantile regression loss function.

[0040] Specifically, this step aims to construct a mapping from the feature space to the power probability distribution, including: 1. Data Preprocessing: First, the meteorological data (temperature, humidity, irradiance, etc.) and each decomposed IMF component are processed using Max-Min normalization and mapped to... The interval is used to eliminate dimensional differences and accelerate the convergence of the neural network. At the same time, the target load sequence is normalized in the same way, and inverse normalization is performed after the model output to recover the actual power value.

[0041] 2. Model Structure and Input / Output: This embodiment uses a temporal fusion Transformer (TFT) or LSTM-Attention model as the prediction backbone. The model's input layer receives a multi-dimensional feature matrix composed of normalized IMF components, meteorological feature vectors, and calendar features. The model's output target (Label) is set as the total raw load power or total renewable energy output at the next time step. That is, the model directly maps the probability distribution of the total power by learning the nonlinear combination relationship between each frequency component (IMF) and meteorological factors, rather than predicting each component separately.

[0042] 3. Loss Function: To quantify the uncertainty of the prediction, the model is trained using the quantile loss function. By minimizing this loss function, the model output approximates a specific quantile of the true distribution. The mathematical formula for the quantile loss function is as follows: (7) In the formula, Indicates quantiles The loss value below; This represents the number of samples (i.e., the training batch size, BatchSize). For the first The true total power value of each sample; The model predicts the first quantile values; For the preset quantiles (e.g.) ).

[0043] The third step is to output the upper and lower bounds of the power prediction at a preset confidence level at future times, and the probability prediction interval is formed by the upper and lower bounds.

[0044] Specifically, based on the trained model, for any future time... The model simultaneously outputs predicted values ​​corresponding to different quantiles. In this embodiment, the confidence level is set to [value missing]. (For example (i.e., 80% confidence level), select quantiles. (Right now The corresponding output value serves as the lower bound for prediction. Select quantiles (Right now The corresponding output value serves as the upper bound for prediction. .

[0045] The final probability prediction interval is represented as follows: (8) This range indicates that there is an 80% probability that the actual power value at future times will fall within this range, providing a reliable boundary constraint for robust optimization control in subsequent steps.

[0046] This embodiment uses measured load data from a typical day in an industrial park for verification. Figure 2 A schematic diagram showing the probability prediction results based on the VMD-TFT model is presented. Figure 2 The horizontal axis represents time (0:00-24:00), and the vertical axis represents load power (kW). The solid black line in the graph represents the actual load curve; the dashed blue line represents the deterministic point prediction (i.e., quantile) output by the model. The shaded area represents the area bounded by the upper bound. and the lower realm The 80% confidence interval is formed.

[0047] from Figure 1 As can be seen, during the period of severe photovoltaic fluctuations from 10:00 to 14:00, the shaded area (prediction interval) automatically widens, indicating that the model successfully captures the high uncertainty risk during this period; while during the stable load period at night, the prediction interval narrows. The actual load curve falls within this shaded area 96.5% of the time points, proving the effectiveness of the probabilistic prediction.

[0048] S3. Based on the probability prediction interval, construct a model prediction control optimization model and solve it to obtain the candidate charging and discharging strategies of the energy storage system.

[0049] The aforementioned model prediction control optimization model includes establishing an optimization model with the charging and discharging power of the energy storage system at each time step as the decision variable, and aiming to minimize the total operating cost of the system.

[0050] Specifically, this embodiment employs a retceding horizon control mechanism. The prediction horizon length is set to... The time interval for control decisions is Defining energy storage systems for the future Each time step ( The charging and discharging power is the decision variable. This includes: 1. Continuous variable: Charging power and discharge power (All are non-negative values); 2. Binary state variables: Charging status flags and discharge status flags (Value can be 0 or 1); 3. Auxiliary variables: used for power grid interaction and semi-cycle markers used for aging calculations .

[0051] The standard form of the optimization model is constructed as follows: (9) (10) Specific constraints include: 1. Charging and discharging mutual exclusion and physical constraints: To avoid the physically impossible phenomenon of "simultaneous charging and discharging," mutual exclusion constraints must be introduced: (11) (12) (13) In the formula, and These are the maximum allowable charging and discharging power of the energy storage converter, respectively.

[0052] 1. Real-time power balance constraint (i.e., equality constraint) ): Real-time power balance among power sources, grid, load, and energy storage must be achieved. (14) In the formula, and These are the load power and renewable energy output power predicted based on step S2, respectively. This refers to the power exchange between power grid interconnections.

[0053] 3. Robust constraints (i.e., inequality constraints) based on probability prediction intervals ): To address the uncertainty of forecasts and ensure that the power grid interaction does not exceed limits under the worst-case operating conditions, the following interval constraints are constructed: (15) (16) In the formula, and These correspond to the lower and upper bounds of the probability prediction output in S2, respectively. and This represents the permissible tie-line power exchange limit for the power grid. These constraints ensure that regardless of load and renewable energy availability within the forecast range... Regardless of internal fluctuations, the power exchange between the power grids remains within a safe range. Inside.

[0054] The total operating cost of the system includes at least two parts: the grid transaction cost calculated based on real-time electricity prices and grid interaction power, and the battery aging cost calculated based on the energy storage battery loss model.

[0055] Specifically, the objective function formula is as follows: (17) in: 1. Grid transaction costs The calculation formula is: (18) In the formula, This represents the time-of-use electricity price vector.

[0056] 2. Battery aging cost The aim is to monetize and quantify the physical losses of batteries, and its calculation relies on a detailed model of battery cycle life.

[0057] The battery aging cost is obtained by extracting charge-discharge cycle characteristics using the rainflow counting method and dynamically calculating them in conjunction with the current battery health status.

[0058] Specifically, considering that rainflow counting involves identifying extreme points, which is a nonlinear and non-differentiable computational process, it is difficult to directly embed it into conventional convex optimization solvers. Therefore, this embodiment uses mixed integer linear programming (MILP) to approximate the aging cost in a piecewise linear manner.

[0059] The first step is to introduce auxiliary variables to capture the "semi-loop" switching point.

[0060] Define auxiliary binary variables Used to identify at time Has a reversal of the charging / discharging direction occurred? Establish the following linear inequality constraint: (19) (20) This constraint ensures that when the energy storage system undergoes a state transition (such as from charging to discharging), If forced to be set to 1, count one half-cycle.

[0061] The second step is to construct a piecewise linearized aging cost function.

[0062] The aging cost is broken down into two parts: energy flow loss and mechanical stress loss. The calculation formula is as follows: (twenty one) In the formula: (Unit: Yuan / kWh): Characterizes the average energy throughput loss coefficient caused by the electrochemical reaction of the battery; (Unit: Yuan / cycle): This parameter represents the fixed depreciation factor corresponding to the change in mechanical stress caused by state switching. It is obtained through offline rainflow counting analysis. Specifically, using the known cycle life curve of the battery, typical operating conditions are discretized and simulated using the rainflow counting method. The total loss amount is then regressed to energy throughput and cycle number using the least squares method to calibrate the depreciation factor. and The specific value; Health status penalty factor, used to dynamically increase cost weight in the later stages of battery aging, forcing the control strategy to optimize and extend remaining lifespan; among which The current battery health status, This is a weighting factor. The role of this coefficient is to non-linearly increase the weight of aging costs as the battery aging degree deepens (SOH decreases), thereby forcing the control strategy to automatically tend towards a low-loss operation mode (such as shallow charging and discharging) at the end of the battery's life, achieving the optimal balance between economy and safety throughout the entire life cycle.

[0063] Figure 3 The graph shows the evolution curve of the health status penalty coefficient as a function of battery aging. The horizontal axis represents the current health status of the battery, ranging from 100% (new battery) to 70% (retired); the vertical axis represents the penalty factor. As shown, when the state of health (SOH) is in the "young and vigorous" stage (90%-100%), Maintaining a value around 1.0 allows the battery to actively participate in deep charge-discharge arbitrage; when the State of Harm (SOH) drops below 80% and enters its "old age," the curve rises exponentially. This will lead to aging costs in the optimization model. The surge in power forces the solver to mathematically favor shallow charging and discharging or reduce the frequency of operations, thus achieving the effect of "maintaining the health of old batteries".

[0064] It should be noted that, due to It captures the instant of each charging or discharging action, corresponding to a half-cycle in rainflow counting. Therefore, The value of should correspond to the depreciation value of the equivalent mechanical stress generated in a single half-cycle.

[0065] Through the above transformation, the complex rainflow counting logic is mathematically equivalently mapped to a weighted sum of energy flow terms and state transition terms. This allows the model to retain physical sensitivity to "cycle depth" and "cycle count" while fully conforming to the solution specifications of mixed integer linear programming, ensuring the computability and real-time performance of the control strategy at the millisecond time scale.

[0066] S4. The candidate charging and discharging strategy is virtually pre-simulated in the multi-scale online simulation model. Based on the pre-simulation results, the candidate charging and discharging strategy is verified and corrected for safety, and control commands are obtained.

[0067] Specifically, in order to balance the solution efficiency of model predictive control with the microscopic accuracy of battery internal state verification, the multi-scale online simulation model in this embodiment adopts a multi-scale two-layer modeling strategy: 1. Control layer model: The second-order RC equivalent circuit model described in S1 is adopted. This model has a fast calculation speed and is used for the rapid iterative solution of the MPC optimization problem in step S3. 2. Verification layer model: A quasi-two-dimensional (P2D) electrochemical-thermal coupling mechanism model is adopted. This model contains a set of partial differential equations describing the lithium ion concentration distribution and potential distribution. It has high computational complexity but can accurately reflect the microscopic mechanism. It is used to perform high-precision safety verification of the strategy in this step (S4).

[0068] The virtual pre-simulation process includes updating the physical parameters of the multi-scale online simulation model in real time using an online parameter identification algorithm. Specifically, it includes: Real-time voltage, current, and temperature measurements of the physical power grid are obtained; a state observer based on Kalman filtering is constructed, and the physical parameters in the multi-scale online simulation model are iteratively corrected by using the residual between the measured values ​​and the output values ​​of the multi-scale online simulation model as the driving force.

[0069] Specifically, this step mainly focuses on high-frequency correction of the parameters of the control layer model (RC model) to ensure the accuracy of the MPC optimization foundation. The Extended Kalman Filter (EKF) algorithm is employed, with the state vector set as... The state update equation is as follows: (twenty two) In the formula, For Kalman gain, This is the measured terminal voltage. for The system input at any given time is the measured current of the energy storage battery. This is a nonlinear observation function based on a second-order RC equivalent circuit, used to calculate the predicted terminal voltage of the model; This refers to the voltage residual. The algorithm is used to correct the internal resistance of the battery's equivalent circuit in real time.

[0070] It should be noted that for the microscopic parameters (such as the solid-phase diffusion coefficient and reaction rate constant) in the verification layer model (P2D model), since their changes on a slow time scale, this embodiment uses parameters based on the current temperature. and health status The multidimensional lookup table method is used for synchronous updates. Specifically, the microscopic parameters in the multidimensional lookup table are obtained through pre-calibration experiments of GITT (giant current intermittent titration) and EIS (electrochemical impedance spectroscopy) on the same type of battery cell. These parameters are updated synchronously using temperature. and Establish mapping relationships for indexes, for example in Below, the typical range of values ​​for the solid-phase diffusion coefficient is: .

[0071] During the simulation, the system uses real-time temperature feedback from the sensors and The estimated values ​​are used to calculate the model parameters for the current step size through interpolation. This ensures that the mechanistic model reflects the current electrochemical characteristics. When the absolute value of the voltage residual in the above EKF algorithm remains below the preset convergence threshold (e.g., 0.01V), the multi-scale online simulation model is considered to have completed parameter synchronization.

[0072] Subsequently, the battery microstate under candidate charge-discharge strategies was simulated based on the updated parameters. Specifically, this included: The candidate charge-discharge strategy is input into the electrochemical-thermal coupling mechanism model; the electrochemical reaction state inside the battery is calculated, and the electrochemical reaction state includes at least the solid phase potential, liquid phase potential and the internal temperature field of the cell.

[0073] Specifically, firstly, based on the candidate strategies output in step S3 ( ), calculate net input current The current is converted into current density and input into the P2D model for solution. This model is based on porous electrode theory, and its core equations include: 1. Solid-phase lithium-ion diffusion equation (Fick's law): (twenty three) 2. Liquid phase charge conservation equation: (twenty four) In the formula, This represents the concentration of lithium in the solid phase. These are the coordinates of the active particle radius. The solid-phase diffusion coefficient is... The liquid phase potential, The coordinates are along the thickness of the electrode plate. Effective conductivity; This refers to the lithium ion concentration in the electrolyte. The effective diffusion conductivity of the electrolyte reflects the effect of the concentration gradient on the potential. The total reaction current density per unit volume of electrode (i.e., volume current density) describes the electrochemical reaction rate at the solid-liquid interface.

[0074] By numerically solving the above equations, the global solid-state potential distribution is obtained. Liquid phase potential distribution and the cell center temperature calculated by the coupled thermal model .

[0075] Determine whether the electrochemical reaction state satisfies the microscopic safety constraints, which include at least a potential difference constraint for suppressing the lithium plating reaction and a temperature constraint for preventing thermal runaway.

[0076] The potential difference constraint requires that the difference between the solid phase potential and the liquid phase potential be above a preset lithium plating safety threshold, and the temperature constraint requires that the internal temperature of the cell be below a preset safety temperature threshold.

[0077] Specifically, the inequalities are as follows: 1. Lithium plating constraint: In the anode region, the local potential difference must always be greater than the safety threshold. (like ): (25) 2. Temperature constraint: The center temperature of the battery cell must be below the limit temperature. : (26) Finally, based on the simulation results, the candidate charge-discharge strategies are subjected to safety verification and correction, including: When the microscopic safety constraints are met, the candidate charging and discharging strategy is marked as a control command; When the micro-safety constraints are violated, the penalty factor corresponding to the degree of violation is calculated and fed back to the constraints of the model predictive control framework; the maximum allowable charge and discharge power boundary of the energy storage system is tightened, and the optimization problem is solved again until the generated strategy passes the virtual simulation.

[0078] Specifically, a sensitivity-based directional correction strategy is adopted to correct the power boundary in step S3 according to the type of constraint violation: Scenario 1: Triggering the risk of lithium plating If the virtual rehearsal discovers the first There is a constant risk of lithium plating (i.e., the local potential difference is less than the safety threshold). If so, calculate the degree of violation. Using a pre-calibrated sensitivity coefficient, the voltage violation is mapped to a power penalty factor. : (27) In the formula, For the safety margin factor (e.g., take 1.2), the denominator is... The sensitivity coefficient of the negative electrode potential to the current is determined by pre-calibration or online calculation; specifically, due to the strong nonlinearity of the P2D model, the sensitivity coefficient is obtained through any of the following methods: 1. Online Finite Difference Method: At the current simulation moment, for candidate currents... Apply small perturbation (For example, taking 1% of the rated current), recalculate the potential difference response under this small step size using the P2D model, and then use the formula... Obtain the local linear derivative; 2. Offline sensitivity lookup table method: The P2D model is used to simulate under different temperatures, SOC and rate conditions in advance, and the potential-current response slope under each condition is extracted to construct a multi-dimensional lookup table of sensitivity coefficients. The table is then retrieved in real time based on the current state during the simulation.

[0079] Using this penalty factor, tighten the first The upper bound constraint on power at time t is to update formula (11) as follows: (28) in, Candidate charging power values ​​that could lead to this risk.

[0080] Scenario 2: Risk of triggering temperature exceeding limits If the virtual rehearsal discovers the first When the cell temperature exceeds the limit, the temperature-power sensitivity coefficient, either pre-calibrated or calculated online, is used. (Represents the rate of temperature rise caused by a unit power change), calculate the temperature-corrected power penalty factor. : (29) If the current state is a discharge state, then tighten the first... The upper limit of discharge power at any given time is to update formula (12) as follows: (30) If the current state is charging, update formula (11) as follows: (31) In the formula, The candidate discharge power values ​​are then used. The updated dynamic constraints are substituted into the optimization model of step S3 and solved again until all micro-constraints are satisfied or the maximum number of iterations is reached, ultimately outputting a safe control command.

[0081] This embodiment uses a high-power fast charging scenario as an example to demonstrate the correction effect of the virtual simulation. Figure 4 The dynamic simulation comparison charts before and after the correction of the potential difference between the solid and liquid phases of the battery negative electrode are shown.

[0082] In the figure, the left Y-axis represents the charging current (A), and the right Y-axis represents the minimum solid-liquid phase potential difference at the negative electrode.

[0083] In addition, to prevent online computation from timeout, a maximum number of correction iterations is set. (For example, 3 times). If the strategy still fails the microscopic safety verification after 3 corrections and recalculations, a safety degradation protection mechanism is triggered, directly forcing the allowable charging and discharging power at that moment to 0 or switching to the minimum safe power mode to ensure the absolute safety of the physical system. The dashed safety threshold represents the lithium plating danger zone (potential difference < 0.05V). Dashed curve (before correction): corresponds to the aggressive charging strategy initially generated in S3. It can be seen that at the end of charging (SOC > 80%), due to the continuous injection of large current, the negative electrode potential difference drops rapidly and crosses the 0.05V red line, even approaching 0V at t=15min, posing an extremely high risk of lithium plating and fire. Solid curve (after correction): corresponds to the strategy after correction by virtual pre-simulation feedback in S4. It can be seen that after the multi-scale online simulation model predicts the potential exceeding the limit trend, it forcibly limits the peak value of the charging current. The corrected potential difference curve shows a clear "clamping" characteristic when approaching 0.05V, always remaining above the safety threshold. Figure 4 It is evident that this invention, through microscopic mechanism simulation, can avoid potential lithium plating accidents in batteries.

[0084] S5, based on the control commands, coordinately adjust the energy storage system. Specifically: The rolling time-domain control method is adopted, and the power setpoint of the first time step in the control command is sent to the energy storage converter only.

[0085] Specifically, the core of this embodiment lies in utilizing the retceding horizon characteristic of model predictive control (MPC) to introduce a real-time feedback mechanism into the regulation loop. In step S4, the optimal charge-discharge control sequence obtained after virtual pre-simulation correction can be expressed as: (32) In the formula, Indicates the current The optimal control vector obtained at each time step; Indicates in The first time predicted The optimal decision value of energy storage power at any given time (unit: kW).

[0086] According to the execution principle of the rolling time domain, the system does not execute the entire control sequence sequentially, but only extracts the first component of the sequence. As the current execution target, the remaining ones in this sequence Each instruction component will be treated as an intermediate redundant result and discarded by the system.

[0087] During the process of issuing commands to the power conversion system (PCS), in order to ensure the operational safety of physical equipment, the power setpoint issuance process also includes a power smoothing process. Specifically, the system pre-sets the rated power ramp-up limit (Rate-of-change limit) of the energy storage power station, denoted as... (Unit: kW / s). The PCS controller receives the optimal decision value. Then, it will be combined with the actual output power at the previous moment. A smooth transition calculation is performed to ensure that the power adjustment amplitude of the current step size does not exceed the ramp constraint. The command will smoothly transition to the target value with a preset slope, thereby effectively avoiding mechanical stress damage to the internal chemical structure of the battery caused by instantaneous high current surges and reducing the negative impact of high-power switching on grid voltage stability.

[0088] In the next control cycle, based on the latest measured data and the updated state of the multi-scale online simulation model, the prediction, optimization, and virtual simulation steps are repeated.

[0089] Specifically, after a preset control step size (For example (minutes) later, the system timeline changed from Advance to This initiates a new round of closed-loop regulation. The entire process, which is repeated repeatedly, includes: returning to step S1 to obtain the latest grid and battery measurement data, and correcting the physical parameters of the multi-scale online simulation model; executing step S2 to update the power prediction range; and executing steps S3 and S4 to resolve and verify the optimal control sequence. This iterative mechanism can correct random errors in load forecasting and model biases in the multi-scale online simulation model in real time.

[0090] It should be noted that, to ensure the real-time performance and reliability of the system in industrial settings, a computation time window constraint is set in this embodiment. Specifically, the system monitors the total computation time of steps S1 to S4 in real time. and require it to meet ,in A preset safety factor (e.g., 0.8) is used. If, within the current control cycle, network fluctuations or excessively complex optimization problems prevent the solver from providing the optimal solution within a specified time threshold, the system automatically executes a fault-tolerant degradation strategy. Under this strategy, the system will maintain the power command from the previous moment and operate smoothly, or switch to a preset constant power safety mode determined based on the current SOC state (e.g., if SOC > 80%, perform a discharge at 20% of the rated power; if SOC < 20%, perform a charge at 20% of the rated power; the remaining intervals remain in standby mode), and simultaneously record the system alarm log until the solver resumes normal output in the next cycle. Through the combination of the above real-time guarantee mechanism and rolling optimization logic, this invention achieves long-term and stable operation of the energy storage system in complex dynamic environments.

[0091] Example 2, Figure 2 A power grid load dynamic prediction and energy storage coordinated regulation system based on a two-layer model virtual simulation is presented, including: The online simulation module is used to collect multi-source heterogeneous data from the grid side, power source side, and energy storage side, and to build a multi-scale online simulation model that is synchronized with the physical system. The probability prediction module is used to generate a probability prediction range for load and new energy output within a preset time period based on the multi-source heterogeneous data and using a prediction model. The collaborative optimization module is used to construct a model prediction control optimization model based on the probability prediction interval, and solve for the candidate charging and discharging strategies of the energy storage system. The virtual-real pre-simulation module is used to perform virtual pre-simulation of the candidate charging and discharging strategies in the multi-scale online simulation model, and to perform safety verification and correction of the candidate charging and discharging strategies based on the pre-simulation results to obtain control commands; An execution control module is used to coordinate and regulate the energy storage system based on the control commands.

[0092] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0093] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0094] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0095] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0096] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0097] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. 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. A method for dynamic prediction of power grid load and coordinated regulation of energy storage based on a two-layer model virtual simulation, characterized in that, Includes the following steps: Construct a multi-scale online simulation model that integrates battery electrochemical and thermal coupling mechanism models; Based on multi-source heterogeneous data, a probability prediction interval for load and renewable energy output is generated. By using the probability prediction interval as a constraint, the prediction optimization model is solved to obtain candidate charging and discharging strategies for the energy storage system. The strategy is virtually pre-simulated in a multi-scale online simulation model to obtain the electrochemical reaction state; Safety verification of candidate charge-discharge strategies based on electrochemical reaction states; When the verification fails, a penalty factor is generated based on the degree of violation and fed back to the prediction optimization model. The constraints are tightened and the solution is recalculated until the verification passes. Then, control commands are generated to coordinate the adjustment of the energy storage system.

2. The method for dynamic prediction of power grid load and coordinated regulation of energy storage based on two-layer model virtual pre-simulation as described in claim 1, characterized in that, The battery electrochemical and thermal coupling mechanism model adopts a two-layer architecture, including a control layer model and a verification layer model; The control layer model is constructed based on a second-order RC equivalent circuit and is used to provide state constraints when solving the predictive optimization model. The verification layer model is constructed based on the quasi-two-dimensional P2D electrochemical mechanism and is used to perform virtual pre-simulation of the candidate charge-discharge strategies in a multi-scale online simulation model to obtain the solid-phase lithium-ion concentration distribution and liquid-phase potential distribution inside the battery.

3. The method for dynamic prediction of power grid load and coordinated regulation of energy storage based on two-layer model virtual pre-simulation as described in claim 2, characterized in that, The process of virtually rehearsing the strategy in a multi-scale online simulation model to obtain the electrochemical reaction state includes: In the multi-scale online simulation model, the electrochemical reaction state inside the battery is obtained based on the electrochemical and thermal coupling mechanism model. Determine whether the electrochemical reaction state meets the microscopic safety constraints, which include at least the potential difference constraint for suppressing the lithium plating reaction and the temperature constraint for preventing thermal runaway.

4. The method for dynamic prediction of power grid load and coordinated regulation of energy storage based on two-layer model virtual pre-simulation as described in claim 3, characterized in that, The safety verification of candidate charge-discharge strategies based on electrochemical reaction states includes: If the electrochemical reaction state satisfies the microscopic safety constraints, then the candidate charge-discharge strategy will be used as the final control command. If the micro-safety constraints are violated, a penalty factor corresponding to the degree of violation is calculated and fed back into the constraints of the model predictive control. Tighten the charging and discharging power constraints and resolve until a charging and discharging strategy that satisfies the microscopic safety constraints is obtained.

5. The method for dynamic prediction of power grid load and coordinated regulation of energy storage based on two-layer model virtual pre-simulation as described in claim 1, characterized in that, The generation of penalty factors based on the degree of violation includes: When the risk of lithium plating is triggered, the difference between the minimum solid-liquid phase potential difference of the negative electrode and the lithium plating safety threshold is calculated as the degree of violation. The degree of violation is mapped to a power penalty factor using a pre-calibrated sensitivity coefficient of the negative electrode potential to the current. When a temperature exceedance risk is triggered, the difference between the cell center temperature and the safe temperature threshold is calculated as the degree of violation. Using a pre-calibrated temperature-to-power sensitivity coefficient, the power adjustment required to bring the temperature back to the safe temperature threshold is calculated as a power penalty factor.

6. The method for dynamic prediction of power grid load and coordinated regulation of energy storage based on two-layer model virtual pre-simulation as described in claim 1, characterized in that, The virtual simulation process includes using an online parameter identification algorithm to update the physical parameters of the multi-scale online simulation model in real time, and calculating the electrochemical reaction state based on the updated parameters. If the electrochemical reaction state exceeds the limit, a strategy correction is triggered.

7. The method for dynamic prediction of power grid load and coordinated regulation of energy storage based on two-layer model virtual pre-simulation as described in claim 1, characterized in that, Based on multi-source heterogeneous data, probability prediction intervals for load and renewable energy output are generated, including: The historical load sequence is decomposed into multiple intrinsic mode functions using the variational mode decomposition method; A deep learning model trained with intrinsic mode functions and meteorological features using a quantile regression loss function is used to obtain the upper and lower bounds of power prediction at a preset time and a preset confidence level, thus forming a probability prediction interval.

8. The method for dynamic prediction of power grid load and coordinated regulation of energy storage based on two-layer model virtual pre-simulation as described in claim 6, characterized in that, The method of using an online parameter identification algorithm to update the physical parameters of a multi-scale online simulation model in real time specifically includes: Obtain real-time measurements of the physical power grid; A state observer is constructed, and the physical parameters in the multi-scale online simulation model are iteratively corrected based on the residual between the measured values ​​and the output values ​​of the multi-scale online simulation model. When the residual converges to the preset threshold, the parameter synchronization of the multi-scale online simulation model is completed.

9. The method for dynamic prediction of power grid load and coordinated regulation of energy storage based on two-layer model virtual pre-simulation as described in claim 1, characterized in that, The step of solving the prediction optimization model using the probability prediction interval as a constraint includes constructing a model prediction control optimization model, specifically: An optimization model is established with the charging and discharging power of the energy storage system at each time step as the decision variable, aiming to minimize the total operating cost of the system. The total operating cost of the system includes at least two parts: the grid transaction cost calculated based on real-time electricity price and grid interaction power, and the battery aging cost calculated based on the energy storage battery loss model. The battery aging cost is calculated by extracting charge-discharge cycle characteristics using the rainflow counting method and combining them with the current battery health status.

10. The method for dynamic prediction of power grid load and coordinated regulation of energy storage based on a two-layer model virtual simulation as described in claim 1, characterized in that, The aforementioned coordinated adjustment of the energy storage system via control commands specifically includes: The rolling time-domain control method is adopted, and the power setpoint of the first time step in the control command is only sent to the energy storage converter. In the next control cycle, based on the latest measured data and the updated state of the multi-scale online simulation model, the prediction, optimization, and virtual simulation steps are repeated.

Images